bumpity bump (#987)

* rebase

* nit

* fix eval in vmap

.pre-commit-config.yaml

@@ -1,11 +1,11 @@
 repos:
 -   repo: https://github.com/pre-commit/mirrors-clang-format
-    rev: v17.0.6
+    rev: v18.1.3
     hooks:
     -   id: clang-format
 # Using this mirror lets us use mypyc-compiled black, which is about 2x faster
 -   repo: https://github.com/psf/black-pre-commit-mirror
-    rev: 24.2.0
+    rev: 24.3.0
     hooks:
     -   id: black
 -   repo: https://github.com/pycqa/isort

ACKNOWLEDGMENTS.md

@@ -15,6 +15,8 @@ MLX was developed with contributions from the following individuals:
 - Hinrik Snær Guðmundsson: Added `atleast_1d`, `atleast_2d`, `atleast_3d` ops.
 - Luca Arnaboldi: Added `Ceil` and `Floor` ops; implemented pickling, copy and deepcopy for mlx arrays.
 - Brian Keene & Atila Orhon, with Argmax Inc.: Added `fast.scaled_dot_product_attention`
+- AmirHossein Razlighi: Added chaining support for some of the ops in `nn.Module`. Comparison works for non array objects in `mlx.core.array`. Exception handling for invalid operations in `mlx.core.array`.
+
 <a href="https://github.com/ml-explore/mlx/graphs/contributors">
   <img class="dark-light" src="https://contrib.rocks/image?repo=ml-explore/mlx&anon=0&columns=20&max=100&r=true" />
 </a>

CMakeLists.txt

@@ -15,31 +15,33 @@ option(MLX_BUILD_EXAMPLES "Build examples for mlx" ON)
 option(MLX_BUILD_BENCHMARKS "Build benchmarks for mlx" OFF)
 option(MLX_BUILD_PYTHON_BINDINGS "Build python bindings for mlx" OFF)
 option(MLX_BUILD_METAL "Build metal backend" ON)
+option(MLX_METAL_DEBUG "Enhance metal debug workflow" OFF)
+option(MLX_ENABLE_X64_MAC "Enable building for x64 macOS" OFF)
 option(BUILD_SHARED_LIBS "Build mlx as a shared library" OFF)
 
 if(NOT MLX_VERSION)
-  set(MLX_VERSION 0.8.1)
+  set(MLX_VERSION 0.10.0)
 endif()
 
 # --------------------- Processor tests -------------------------
 
-message(STATUS "Building MLX for ${CMAKE_HOST_SYSTEM_PROCESSOR} processor on ${CMAKE_SYSTEM_NAME}")
+message(STATUS "Building MLX for ${CMAKE_SYSTEM_PROCESSOR} processor on ${CMAKE_SYSTEM_NAME}")
 
 set(MLX_BUILD_ARM OFF)
 
 if (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
-  if (${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "x86_64" AND ${CMAKE_HOST_APPLE})
-    message(FATAL_ERROR
-      "Building for x86_64 on macOS is not supported."
-      " If you are on an Apple silicon system, check the build"
-      " documentation for possible fixes: "
-      "https://ml-explore.github.io/mlx/build/html/install.html#build-from-source")
-  elseif (${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "x86_64")
-    message(WARNING
-      "Building for x86_64 on macOS is not supported."
-      " If you are on an Apple silicon system, "
-      " make sure you are building for arm64.")
-  elseif(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "arm64")
+  if(${CMAKE_SYSTEM_PROCESSOR} MATCHES "x86_64")
+    if(NOT MLX_ENABLE_X64_MAC)
+      message(FATAL_ERROR
+        "Building for x86_64 on macOS is not supported."
+        " If you are on an Apple silicon system, check the build"
+        " documentation for possible fixes: "
+        "https://ml-explore.github.io/mlx/build/html/install.html#build-from-source")
+    else()
+      message(WARNING "Building for x86_64 arch is not officially supported.")
+    endif()
+    set(MLX_BUILD_METAL OFF)
+  elseif(${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64")
     set(MLX_BUILD_ARM ON)
   endif()
 
@@ -64,8 +66,14 @@ endif()
 if (MLX_BUILD_METAL AND NOT METAL_LIB)
   message(STATUS "Metal not found. Unable to build GPU")
   set(MLX_BUILD_METAL OFF)
+  set(MLX_METAL_DEBUG OFF)
 elseif (MLX_BUILD_METAL)
   message(STATUS "Building METAL sources")
+
+  if (MLX_METAL_DEBUG)
+    add_compile_definitions(MLX_METAL_DEBUG)
+  endif()
+
   # Throw an error if xcrun not found
   execute_process(COMMAND zsh "-c" "/usr/bin/xcrun -sdk macosx --show-sdk-version"
                   OUTPUT_VARIABLE MACOS_VERSION
@@ -108,7 +116,27 @@ if (MLX_BUILD_ARM AND ACCELERATE_LIBRARY)
 else()
   message(STATUS "Accelerate or arm neon not found, using default backend.")
   set(MLX_BUILD_ACCELERATE OFF)
-  #set(BLA_VENDOR Generic)
+  if(${CMAKE_HOST_APPLE})
+    # The blas shipped in macOS SDK is not supported, search homebrew for
+    # openblas instead.
+    set(BLA_VENDOR OpenBLAS)
+    set(LAPACK_ROOT "${LAPACK_ROOT};$ENV{LAPACK_ROOT};/usr/local/opt/openblas")
+  endif()
+  # Search and link with lapack.
+  find_package(LAPACK REQUIRED)
+  if (NOT LAPACK_FOUND)
+    message(FATAL_ERROR "Must have LAPACK installed")
+  endif()
+  find_path(LAPACK_INCLUDE_DIRS lapacke.h
+    /usr/include
+    /usr/local/include
+    /usr/local/opt/openblas/include)
+  message(STATUS "Lapack lib " ${LAPACK_LIBRARIES})
+  message(STATUS "Lapack include " ${LAPACK_INCLUDE_DIRS})
+  target_include_directories(mlx PRIVATE ${LAPACK_INCLUDE_DIRS})
+  target_link_libraries(mlx ${LAPACK_LIBRARIES})
+  # List blas after lapack otherwise we may accidentally incldue an old version
+  # of lapack.h from the include dirs of blas.
   find_package(BLAS REQUIRED)
   if (NOT BLAS_FOUND)
     message(FATAL_ERROR "Must have BLAS installed")
@@ -122,17 +150,6 @@ else()
   message(STATUS "Blas include " ${BLAS_INCLUDE_DIRS})
   target_include_directories(mlx PRIVATE ${BLAS_INCLUDE_DIRS})
   target_link_libraries(mlx ${BLAS_LIBRARIES})
-  find_package(LAPACK REQUIRED)
-  if (NOT LAPACK_FOUND)
-      message(FATAL_ERROR "Must have LAPACK installed")
-  endif()
-  find_path(LAPACK_INCLUDE_DIRS lapacke.h
-    /usr/include
-    /usr/local/include)
-  message(STATUS "Lapack lib " ${LAPACK_LIBRARIES})
-  message(STATUS "Lapack include " ${LAPACK_INCLUDE_DIRS})
-  target_include_directories(mlx PRIVATE ${LAPACK_INCLUDE_DIRS})
-  target_link_libraries(mlx ${LAPACK_LIBRARIES})
 endif()
 
 add_subdirectory(${CMAKE_CURRENT_LIST_DIR}/mlx)

benchmarks/cpp/time_utils.h

@@ -17,14 +17,13 @@
             << std::setprecision(5) << time_fn(FUNC, ##__VA_ARGS__) << " msec" \
             << std::endl;
 
-#define TIMEM(MSG, FUNC, ...)                                                  \
-  std::cout << "Timing "                                                       \
-            << "(" << MSG << ") " << #FUNC << " ... " << std::flush            \
-            << std::setprecision(5) << time_fn(FUNC, ##__VA_ARGS__) << " msec" \
-            << std::endl;
+#define TIMEM(MSG, FUNC, ...)                                      \
+  std::cout << "Timing " << "(" << MSG << ") " << #FUNC << " ... " \
+            << std::flush << std::setprecision(5)                  \
+            << time_fn(FUNC, ##__VA_ARGS__) << " msec" << std::endl;
 
 template <typename F, typename... Args>
-double time_fn(F fn, Args... args) {
+double time_fn(F fn, Args&&... args) {
   // warmup
   for (int i = 0; i < 5; ++i) {
     eval(fn(std::forward<Args>(args)...));

benchmarks/python/layer_norm_bench.py

@@ -0,0 +1,41 @@
+# Copyright © 2023-2024 Apple Inc.
+
+import mlx.core as mx
+import mlx.nn as nn
+from time_utils import time_fn
+
+
+def layer_norm(x, w, b, eps):
+    ot = x.dtype
+    x = x.astype(mx.float32)
+    mu = mx.mean(x, -1, keepdims=True)
+    v = mx.var(x, -1, keepdims=True)
+    return (x - mu) * mx.rsqrt(v + eps) * w + b
+
+
+def time_layer_norm():
+    f1 = lambda x, w, b, y: (layer_norm(x, w, b, 1e-5) * y).sum()
+    f2 = lambda x, w, b, y: (mx.fast.layer_norm(x, w, b, 1e-5) * y).sum()
+    g1 = mx.grad(f1, argnums=(0, 1, 2))
+    g2 = mx.grad(f2, argnums=(0, 1, 2))
+
+    x = mx.random.uniform(shape=(8, 1024, 4096)).astype(mx.float16)
+    w = mx.random.uniform(shape=(4096,)).astype(mx.float16)
+    b = mx.random.uniform(shape=(4096,)).astype(mx.float16)
+    y = mx.random.uniform(shape=(8, 1024, 4096)).astype(mx.float16)
+    mx.eval(x, w, b, y)
+
+    def layer_norm_loop(g, x, w, b):
+        gx, gw, gb = x, w, b
+        for _ in range(32):
+            gx, gw, gb = g(gx, gw, gb, y)
+        return gx, gw, gb
+
+    time_fn(layer_norm_loop, g1, x, w, b)
+    time_fn(layer_norm_loop, g2, x, w, b)
+    time_fn(layer_norm_loop, mx.compile(g1), x, w, b)
+    time_fn(layer_norm_loop, mx.compile(g2), x, w, b)
+
+
+if __name__ == "__main__":
+    time_layer_norm()

benchmarks/python/rms_norm_bench.py

@@ -0,0 +1,39 @@
+# Copyright © 2023-2024 Apple Inc.
+
+import mlx.core as mx
+import mlx.nn as nn
+from time_utils import time_fn
+
+
+def rms_norm(x, w, eps):
+    ot = x.dtype
+    x = x.astype(mx.float32)
+    n = mx.rsqrt(x.square().mean(-1, keepdims=True) + eps)
+    return (x * n).astype(ot) * w
+
+
+def time_rms_norm():
+    f1 = lambda x, w, y: (rms_norm(x, w, 1e-5) * y).sum()
+    f2 = lambda x, w, y: (mx.fast.rms_norm(x, w, 1e-5) * y).sum()
+    g1 = mx.grad(f1, argnums=(0, 1))
+    g2 = mx.grad(f2, argnums=(0, 1))
+
+    x = mx.random.uniform(shape=(8, 1024, 4096)).astype(mx.float16)
+    w = mx.random.uniform(shape=(4096,)).astype(mx.float16)
+    y = mx.random.uniform(shape=(8, 1024, 4096)).astype(mx.float16)
+    mx.eval(x, w, y)
+
+    def rms_norm_loop(g, x, w):
+        gx, gw = x, w
+        for _ in range(32):
+            gx, gw = g(gx, gw, y)
+        return gx, gw
+
+    time_fn(rms_norm_loop, g1, x, w)
+    time_fn(rms_norm_loop, g2, x, w)
+    time_fn(rms_norm_loop, mx.compile(g1), x, w)
+    time_fn(rms_norm_loop, mx.compile(g2), x, w)
+
+
+if __name__ == "__main__":
+    time_rms_norm()

docs/src/dev/extensions.rst

@@ -1,24 +1,16 @@
 Developer Documentation
 =======================
 
-MLX provides a open and flexible backend to which users may add operations 
-and specialized implementations without much hassle. While the library supplies
-efficient operations that can be used and composed for any number of 
-applications, there may arise cases where new functionalities or highly 
-optimized implementations are needed. For such cases, you may design and 
-implement your own operations that link to and build on top of :mod:`mlx.core`.
-We will introduce the inner-workings of MLX and go over a simple example to 
-learn the steps involved in adding new operations to MLX with your own CPU 
-and GPU implementations. 
+You can extend MLX with custom operations on the CPU or GPU. This guide
+explains how to do that with a simple example.
 
 Introducing the Example
 -----------------------
 
-Let's say that you would like an operation that takes in two arrays, 
-``x`` and ``y``, scales them both by some coefficients ``alpha`` and ``beta``
-respectively, and then adds them together to get the result 
-``z = alpha * x + beta * y``. Well, you can very easily do that by just 
-writing out a function as follows:
+Let's say you would like an operation that takes in two arrays, ``x`` and
+``y``, scales them both by coefficients ``alpha`` and ``beta`` respectively,
+and then adds them together to get the result ``z = alpha * x + beta * y``.
+You can do that in MLX directly:
 
 .. code-block:: python
 
@@ -27,44 +19,35 @@ writing out a function as follows:
     def simple_axpby(x: mx.array, y: mx.array, alpha: float, beta: float) -> mx.array:
         return alpha * x + beta * y
 
-This function performs that operation while leaving the implementations and 
-differentiation to MLX. 
+This function performs that operation while leaving the implementation and
+function transformations to MLX.
 
-However, you work with vector math libraries often and realize that the 
-``axpby`` routine defines the same operation ``Y = (alpha * X) + (beta * Y)``. 
-You would really like the part of your applications that does this operation 
-on the CPU to be very fast - so you decide that you want it to rely on the 
-``axpby`` routine provided by the Accelerate_ framework. Continuing to impose 
-our assumptions on to you, let's also assume that you want to learn how to add 
-your own implementation for the gradients of your new operation while going 
-over the ins-and-outs of the MLX framework. 
+However you may need to customize the underlying implementation, perhaps to
+make it faster or for custom differentiation. In this tutorial we will go
+through adding custom extensions. It will cover:
 
-Well, what a coincidence! You are in the right place. Over the course of this 
-example, we will learn:
-
-* The structure of the MLX library from the frontend API to the backend implementations.
-* How to implement your own CPU backend that redirects to Accelerate_ when appropriate (and a fallback if needed).
-* How to implement your own GPU implementation using metal.
-* How to add your own ``vjp`` and ``jvp``.
-* How to build your implementations, link them to MLX, and bind them to python.
+* The structure of the MLX library.
+* Implementing a CPU operation that redirects to Accelerate_ when appropriate.
+* Implementing a GPU operation using metal.
+* Adding the ``vjp`` and ``jvp`` function transformation.
+* Building a custom extension and binding it to python.
 
 Operations and Primitives
 -------------------------
 
-In one sentence, operations in MLX build the computation graph, and primitives 
-provide the rules for evaluation and transformations of said graph. Let's start 
-by discussing operations in more detail. 
+Operations in MLX build the computation graph. Primitives provide the rules for
+evaluating and transforming the graph. Let's start by discussing operations in
+more detail.
 
 Operations
 ^^^^^^^^^^^
 
-Operations are the frontend functions that operate on arrays. They are defined 
-in the C++ API (:ref:`cpp_ops`) and then we provide bindings to these 
-operations in the Python API (:ref:`ops`). 
+Operations are the front-end functions that operate on arrays. They are defined
+in the C++ API (:ref:`cpp_ops`), and the Python API (:ref:`ops`) binds them.
 
-We would like an operation, :meth:`axpby` that takes in two arrays ``x`` and ``y``,
-and two scalars, ``alpha`` and ``beta``. This is how we would define it in the 
-C++ API:
+We would like an operation, :meth:`axpby` that takes in two arrays ``x`` and
+``y``, and two scalars, ``alpha`` and ``beta``. This is how to define it in
+C++:
 
 .. code-block:: C++
 
@@ -83,10 +66,7 @@ C++ API:
         StreamOrDevice s = {} // Stream on which to schedule the operation
     );
 
-
-This operation itself can call other operations within it if needed. So, the 
-simplest way to go about implementing this operation would be do so in terms 
-of existing operations. 
+The simplest way to this operation is in terms of existing operations:
 
 .. code-block:: C++
 
@@ -100,25 +80,23 @@ of existing operations.
         // Scale x and y on the provided stream
         auto ax = multiply(array(alpha), x, s);
         auto by = multiply(array(beta), y, s);
-        
+
         // Add and return
         return add(ax, by, s);
     }
 
-However, as we discussed earlier, this is not our goal. The operations themselves 
-do not contain the implementations that act on the data, nor do they contain the
-rules of transformations. Rather, they are an easy to use interface that build 
-on top of the building blocks we call :class:`Primitive`. 
+The operations themselves do not contain the implementations that act on the
+data, nor do they contain the rules of transformations. Rather, they are an
+easy to use interface that use :class:`Primitive` building blocks.
 
 Primitives
 ^^^^^^^^^^^
 
-A :class:`Primitive` is part of the computation graph of an :class:`array`. It 
-defines how to create an output given a set of input :class:`array` . Further,
-a :class:`Primitive` is a class that contains rules on how it is evaluated 
-on the CPU or GPU, and how it acts under transformations such as ``vjp`` and 
-``jvp``. These words on their own can be a bit abstract, so lets take a step 
-back and go to our example to give ourselves a more concrete image. 
+A :class:`Primitive` is part of the computation graph of an :class:`array`. It
+defines how to create outputs arrays given a input arrays. Further, a
+:class:`Primitive` has methods to run on the CPU or GPU and for function
+transformations such as ``vjp`` and ``jvp``.  Lets go back to our example to be
+more concrete:
 
 .. code-block:: C++
 
@@ -134,11 +112,15 @@ back and go to our example to give ourselves a more concrete image.
         * To avoid unnecessary allocations, the evaluation function
         * is responsible for allocating space for the array.
         */
-        void eval_cpu(const std::vector<array>& inputs, array& out) override;
-        void eval_gpu(const std::vector<array>& inputs, array& out) override;
+        void eval_cpu(
+            const std::vector<array>& inputs,
+            std::vector<array>& outputs) override;
+        void eval_gpu(
+            const std::vector<array>& inputs,
+            std::vector<array>& outputs) override;
 
         /** The Jacobian-vector product. */
-        array jvp(
+        std::vector<array> jvp(
             const std::vector<array>& primals,
             const std::vector<array>& tangents,
             const std::vector<int>& argnums) override;
@@ -147,7 +129,8 @@ back and go to our example to give ourselves a more concrete image.
         std::vector<array> vjp(
             const std::vector<array>& primals,
             const array& cotan,
-            const std::vector<int>& argnums) override;
+            const std::vector<int>& argnums,
+            const std::vector<array>& outputs) override;
 
         /**
         * The primitive must know how to vectorize itself across
@@ -155,7 +138,7 @@ back and go to our example to give ourselves a more concrete image.
         * representing the vectorized computation and the axis which
         * corresponds to the output vectorized dimension.
         */
-        std::pair<array, int> vmap(
+        virtual std::pair<std::vector<array>, std::vector<int>> vmap(
             const std::vector<array>& inputs,
             const std::vector<int>& axes) override;
 
@@ -175,22 +158,22 @@ back and go to our example to give ourselves a more concrete image.
         void eval(const std::vector<array>& inputs, array& out);
     };
 
-The :class:`Axpby` class derives from the base :class:`Primitive` class and 
-follows the above demonstrated interface. :class:`Axpby` treats ``alpha`` and 
-``beta`` as parameters. It then provides implementations of how the array ``out`` 
-is produced given ``inputs`` through :meth:`Axpby::eval_cpu` and 
-:meth:`Axpby::eval_gpu`. Further, it provides rules of transformations in 
-:meth:`Axpby::jvp`, :meth:`Axpby::vjp`, and :meth:`Axpby::vmap`. 
+The :class:`Axpby` class derives from the base :class:`Primitive` class. The
+:class:`Axpby` treats ``alpha`` and ``beta`` as parameters. It then provides
+implementations of how the output array is produced given the inputs through
+:meth:`Axpby::eval_cpu` and :meth:`Axpby::eval_gpu`. It also provides rules
+of transformations in :meth:`Axpby::jvp`, :meth:`Axpby::vjp`, and
+:meth:`Axpby::vmap`.
 
-Using the Primitives
-^^^^^^^^^^^^^^^^^^^^^
+Using the Primitive
+^^^^^^^^^^^^^^^^^^^
 
-Operations can use this :class:`Primitive` to add a new :class:`array` to 
-the computation graph. An :class:`array` can be constructed by providing its 
-data type, shape, the :class:`Primitive` that computes it, and the 
-:class:`array` inputs that are passed to the primitive.
+Operations can use this :class:`Primitive` to add a new :class:`array` to the
+computation graph. An :class:`array` can be constructed by providing its data
+type, shape, the :class:`Primitive` that computes it, and the :class:`array`
+inputs that are passed to the primitive.
 
-Let's re-implement our operation now in terms of our :class:`Axpby` primitive.
+Let's reimplement our operation now in terms of our :class:`Axpby` primitive.
 
 .. code-block:: C++
 
@@ -223,7 +206,7 @@ Let's re-implement our operation now in terms of our :class:`Axpby` primitive.
             /* const std::vector<int>& shape = */ out_shape,
             /* Dtype dtype = */ out_dtype,
             /* std::unique_ptr<Primitive> primitive = */
-            std::make_unique<Axpby>(to_stream(s), alpha, beta),
+            std::make_shared<Axpby>(to_stream(s), alpha, beta),
             /* const std::vector<array>& inputs = */ broadcasted_inputs);
     }
 
@@ -238,27 +221,26 @@ This operation now handles the following:
 Implementing the Primitive
 --------------------------
 
-No computation happens when we call the operation alone. In effect, the 
-operation only builds the computation graph. When we evaluate the output 
-array, MLX schedules the execution of the computation graph, and calls
-:meth:`Axpby::eval_cpu` or :meth:`Axpby::eval_gpu` depending on the 
-stream/device specified by the user. 
+No computation happens when we call the operation alone. The operation only
+builds the computation graph. When we evaluate the output array, MLX schedules
+the execution of the computation graph, and calls :meth:`Axpby::eval_cpu` or
+:meth:`Axpby::eval_gpu` depending on the stream/device specified by the user.
 
 .. warning::
     When :meth:`Primitive::eval_cpu` or :meth:`Primitive::eval_gpu` are called,
     no memory has been allocated for the output array. It falls on the implementation
-    of these functions to allocate memory as needed
+    of these functions to allocate memory as needed.
 
-Implementing the CPU Backend
+Implementing the CPU Back-end
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-Let's start by trying to implement a naive and generic version of 
-:meth:`Axpby::eval_cpu`. We declared this as a private member function of 
-:class:`Axpby` earlier called :meth:`Axpby::eval`. 
+Let's start by implementing a naive and generic version of
+:meth:`Axpby::eval_cpu`. We declared this as a private member function of
+:class:`Axpby` earlier called :meth:`Axpby::eval`.
 
-Our naive method will go over each element of the output array, find the 
-corresponding input elements of ``x`` and ``y`` and perform the operation 
-pointwise. This is captured in the templated function :meth:`axpby_impl`. 
+Our naive method will go over each element of the output array, find the
+corresponding input elements of ``x`` and ``y`` and perform the operation
+point-wise. This is captured in the templated function :meth:`axpby_impl`.
 
 .. code-block:: C++
 
@@ -296,19 +278,19 @@ pointwise. This is captured in the templated function :meth:`axpby_impl`.
         }
     }
 
-Now, we would like our implementation to be able to do this pointwise operation 
-for all incoming floating point arrays. Accordingly, we add dispatches for 
-``float32``, ``float16``, ``bfloat16`` and ``complex64``. We throw an error 
-if we encounter an unexpected type.
+Our implementation should work for all incoming floating point arrays.
+Accordingly, we add dispatches for ``float32``, ``float16``, ``bfloat16`` and
+``complex64``. We throw an error if we encounter an unexpected type.
 
 .. code-block:: C++
 
     /** Fall back implementation for evaluation on CPU */
-    void Axpby::eval(const std::vector<array>& inputs, array& out) {
-        // Check the inputs (registered in the op while constructing the out array)
-        assert(inputs.size() == 2);
+    void Axpby::eval(
+      const std::vector<array>& inputs,
+      const std::vector<array>& outputs) {
         auto& x = inputs[0];
         auto& y = inputs[1];
+        auto& out = outputs[0];
 
         // Dispatch to the correct dtype
         if (out.dtype() == float32) {
@@ -321,28 +303,26 @@ if we encounter an unexpected type.
             return axpby_impl<complex64_t>(x, y, out, alpha_, beta_);
         } else {
             throw std::runtime_error(
-                "Axpby is only supported for floating point types.");
+                "[Axpby] Only supports floating point types.");
         }
     }
 
-We have a fallback implementation! Now, to do what we are really here to do. 
-Remember we wanted to use the ``axpby`` routine provided by the Accelerate_
-framework? Well, there are 3 complications to keep in mind:
+This is good as a fallback implementation. We can use the ``axpby`` routine
+provided by the Accelerate_ framework for a faster implementation in certain
+cases:
 
 #.  Accelerate does not provide implementations of ``axpby`` for half precision
-    floats. We can only direct to it for ``float32`` types 
-#.  Accelerate assumes the inputs ``x`` and ``y`` are contiguous and all elements
-    have fixed strides between them. Possibly due to broadcasts and transposes, 
-    we aren't guaranteed that the inputs fit this requirement. We can 
-    only direct to Accelerate if both ``x`` and ``y`` are row contiguous or 
-    column contiguous. 
-#.  Accelerate performs the routine ``Y = (alpha * X) + (beta * Y)`` inplace. 
-    MLX expects to write out the answer to a new array. We must copy the elements 
-    of ``y`` into the output array and use that as an input to ``axpby``
+    floats. We can only use it for ``float32`` types.
+#.  Accelerate assumes the inputs ``x`` and ``y`` are contiguous and all
+    elements have fixed strides between them. We only direct to Accelerate
+    if both ``x`` and ``y`` are row contiguous or column contiguous.
+#.  Accelerate performs the routine ``Y = (alpha * X) + (beta * Y)`` in-place.
+    MLX expects to write the output to a new array. We must copy the elements
+    of ``y`` into the output and use that as an input to ``axpby``.
 
-Let's write out an implementation that uses Accelerate in the right conditions. 
-It must simply allocate data for the output, copy elements of ``y`` into it, 
-and then call the :meth:`catlas_saxpby` from accelerate. 
+Let's write an implementation that uses Accelerate in the right conditions.
+It allocates data for the output, copies ``y`` into it, and then calls the
+:func:`catlas_saxpby` from accelerate.
 
 .. code-block:: C++
 
@@ -356,17 +336,7 @@ and then call the :meth:`catlas_saxpby` from accelerate.
         // Accelerate library provides catlas_saxpby which does
         // Y = (alpha * X) + (beta * Y) in place
         // To use it, we first copy the data in y over to the output array
-
-        // This specialization requires both x and y be contiguous in the same mode
-        // i.e: corresponding linear indices in both point to corresponding elements
-        // The data in the output array is allocated to match the strides in y
-        // such that x, y, and out are contiguous in the same mode and
-        // no transposition is needed
-        out.set_data(
-            allocator::malloc_or_wait(y.data_size() * out.itemsize()),
-            y.data_size(),
-            y.strides(),
-            y.flags());
+        out.set_data(allocator::malloc_or_wait(out.nbytes()));
 
         // We then copy over the elements using the contiguous vector specialization
         copy_inplace(y, out, CopyType::Vector);
@@ -389,18 +359,20 @@ and then call the :meth:`catlas_saxpby` from accelerate.
             /* INCY = */ 1);
     }
 
-Great! But what about the inputs that do not fit the criteria for accelerate?
-Luckily, we can always just direct back to :meth:`Axpby::eval`.
-
-With this in mind, lets finally implement our :meth:`Axpby::eval_cpu`.
+For inputs that do not fit the criteria for accelerate, we fall back to
+:meth:`Axpby::eval`. With this in mind, let's finish our
+:meth:`Axpby::eval_cpu`.
 
 .. code-block:: C++
 
     /** Evaluate primitive on CPU using accelerate specializations */
-    void Axpby::eval_cpu(const std::vector<array>& inputs, array& out) {
+    void Axpby::eval_cpu(
+      const std::vector<array>& inputs,
+      const std::vector<array>& outputs) {
         assert(inputs.size() == 2);
         auto& x = inputs[0];
         auto& y = inputs[1];
+        auto& out = outputs[0];
 
         // Accelerate specialization for contiguous single precision float arrays
         if (out.dtype() == float32 &&
@@ -410,35 +382,33 @@ With this in mind, lets finally implement our :meth:`Axpby::eval_cpu`.
             return;
         }
 
-        // Fall back to common backend if specializations are not available
-        eval(inputs, out);
+        // Fall back to common back-end if specializations are not available
+        eval(inputs, outputs);
     }
 
-We have now hit a milestone! Just this much is enough to run the operation 
-:meth:`axpby` on a CPU stream! 
+Just this much is enough to run the operation :meth:`axpby` on a CPU stream! If
+you do not plan on running the operation on the GPU or using transforms on
+computation graphs that contain :class:`Axpby`, you can stop implementing the
+primitive here and enjoy the speed-ups you get from the Accelerate library.
 
-If you do not plan on running the operation on the GPU or using transforms on 
-computation graphs that contain :class:`Axpby`, you can stop implementing the 
-primitive here and enjoy the speed-ups you get from the Accelerate library. 
-
-Implementing the GPU Backend
+Implementing the GPU Back-end
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-Apple silicon devices address their GPUs using the Metal_ shading language, and 
-all GPU kernels in MLX are written using metal. 
+Apple silicon devices address their GPUs using the Metal_ shading language, and
+GPU kernels in MLX are written using Metal.
 
 .. note::
 
-    Here are some helpful resources if you are new to metal!
+    Here are some helpful resources if you are new to Metal:
 
     * A walkthrough of the metal compute pipeline: `Metal Example`_
     * Documentation for metal shading language: `Metal Specification`_
     * Using metal from C++: `Metal-cpp`_
 
-Let's keep the GPU algorithm simple. We will launch exactly as many threads 
-as there are elements in the output. Each thread will pick the element it needs 
-from ``x`` and ``y``, do the pointwise operation, and then update its assigned 
-element in the output. 
+Let's keep the GPU kernel simple. We will launch exactly as many threads as
+there are elements in the output. Each thread will pick the element it needs
+from ``x`` and ``y``, do the point-wise operation, and update its assigned
+element in the output.
 
 .. code-block:: C++
 
@@ -457,15 +427,14 @@ element in the output.
         // Convert linear indices to offsets in array
         auto x_offset = elem_to_loc(index, shape, x_strides, ndim);
         auto y_offset = elem_to_loc(index, shape, y_strides, ndim);
-        
+
         // Do the operation and update the output
-        out[index] = 
+        out[index] =
             static_cast<T>(alpha) * x[x_offset] + static_cast<T>(beta) * y[y_offset];
     }
 
 We then need to instantiate this template for all floating point types and give
-each instantiation a unique host name so we can identify the right kernel for 
-each data type. 
+each instantiation a unique host name so we can identify it.
 
 .. code-block:: C++
 
@@ -488,29 +457,21 @@ each data type.
     instantiate_axpby(bfloat16, bfloat16_t);
     instantiate_axpby(complex64, complex64_t);
 
-This kernel will be compiled into a metal library ``mlx_ext.metallib`` as we 
-will see later in :ref:`Building with CMake`. In the following example, we 
-assume that the library ``mlx_ext.metallib`` will always be co-located with 
-the executable/ shared-library calling the :meth:`register_library` function. 
-The :meth:`register_library` function takes the library's name and potential 
-path (or in this case, a function that can produce the path of the metal 
-library) and tries to load that library if it hasn't already been registered 
-by the relevant static :class:`mlx::core::metal::Device` object. This is why, 
-it is important to package your C++ library with the metal library. We will 
-go over this process in more detail later. 
-
-The logic to determine the kernel, set the inputs, resolve the grid dimensions 
-and dispatch it to the GPU are contained in :meth:`Axpby::eval_gpu` as shown 
+The logic to determine the kernel, set the inputs, resolve the grid dimensions,
+and dispatch to the GPU are contained in :meth:`Axpby::eval_gpu` as shown
 below.
 
 .. code-block:: C++
 
     /** Evaluate primitive on GPU */
-    void Axpby::eval_gpu(const std::vector<array>& inputs, array& out) {
+    void Axpby::eval_gpu(
+      const std::vector<array>& inputs,
+      std::vector<array>& outputs) {
         // Prepare inputs
         assert(inputs.size() == 2);
         auto& x = inputs[0];
         auto& y = inputs[1];
+        auto& out = outputs[0];
 
         // Each primitive carries the stream it should execute on
         // and each stream carries its device identifiers
@@ -518,10 +479,10 @@ below.
         // We get the needed metal device using the stream
         auto& d = metal::device(s.device);
 
-        // Allocate output memory 
+        // Allocate output memory
         out.set_data(allocator::malloc_or_wait(out.nbytes()));
 
-        // Resolve name of kernel (corresponds to axpby.metal)
+        // Resolve name of kernel
         std::ostringstream kname;
         kname << "axpby_" << "general_" << type_to_name(out);
 
@@ -552,7 +513,7 @@ below.
         compute_encoder->setBytes(&alpha_, sizeof(float), 3);
         compute_encoder->setBytes(&beta_, sizeof(float), 4);
 
-        // Encode shape, strides and ndim 
+        // Encode shape, strides and ndim
         compute_encoder->setBytes(x.shape().data(), ndim * sizeof(int), 5);
         compute_encoder->setBytes(x.strides().data(), ndim * sizeof(size_t), 6);
         compute_encoder->setBytes(y.strides().data(), ndim * sizeof(size_t), 7);
@@ -575,28 +536,25 @@ below.
 
 We can now call the :meth:`axpby` operation on both the CPU and the GPU!
 
-A few things to note about MLX and metal before moving on. MLX keeps track 
-of the active ``compute_encoder``. We rely on :meth:`d.get_command_encoder` 
-to give us the active metal compute command encoder instead of building a 
-new one and calling :meth:`compute_encoder->end_encoding` at the end. 
-MLX keeps adding kernels (compute pipelines) to the active command encoder 
-until some specified limit is hit or the compute encoder needs to be flushed 
-for synchronization. MLX also handles enqueuing and committing the associated 
-command buffers as needed. We suggest taking a deeper dive into 
-:class:`metal::Device` if you would like to study this routine further.
+A few things to note about MLX and Metal before moving on. MLX keeps track of
+the active ``command_buffer`` and the ``MTLCommandBuffer`` to which it is
+associated. We rely on :meth:`d.get_command_encoder` to give us the active
+metal compute command encoder instead of building a new one and calling
+:meth:`compute_encoder->end_encoding` at the end. MLX adds kernels (compute
+pipelines) to the active command buffer until some specified limit is hit or
+the command buffer needs to be flushed for synchronization.
 
 Primitive Transforms
 ^^^^^^^^^^^^^^^^^^^^^
 
-Now that we have come this far, let's also learn how to add implementations to 
-transformations in a :class:`Primitive`. These transformations can be built on 
-top of our operations, including the one we just defined now. Which then gives 
-us the following :meth:`Axpby::jvp` and :meth:`Axpby::vjp` implementations.
+Next, let's add implementations for transformations in a :class:`Primitive`.
+These transformations can be built on top of other operations, including the
+one we just defined:
 
 .. code-block:: C++
 
     /** The Jacobian-vector product. */
-    array Axpby::jvp(
+    std::vector<array> Axpby::jvp(
             const std::vector<array>& primals,
             const std::vector<array>& tangents,
             const std::vector<int>& argnums) {
@@ -611,12 +569,12 @@ us the following :meth:`Axpby::jvp` and :meth:`Axpby::vjp` implementations.
         if (argnums.size() > 1) {
             auto scale = argnums[0] == 0 ? alpha_ : beta_;
             auto scale_arr = array(scale, tangents[0].dtype());
-            return multiply(scale_arr, tangents[0], stream());
+            return {multiply(scale_arr, tangents[0], stream())};
         }
         // If, argnums = {0, 1}, we take contributions from both
         // which gives us jvp = tangent_x * alpha + tangent_y * beta
         else {
-            return axpby(tangents[0], tangents[1], alpha_, beta_, stream());
+            return {axpby(tangents[0], tangents[1], alpha_, beta_, stream())};
         }
     }
 
@@ -625,34 +583,35 @@ us the following :meth:`Axpby::jvp` and :meth:`Axpby::vjp` implementations.
     /** The vector-Jacobian product. */
     std::vector<array> Axpby::vjp(
             const std::vector<array>& primals,
-            const array& cotan,
-            const std::vector<int>& argnums) {
+            const std::vector<array>& cotangents,
+            const std::vector<int>& argnums,
+            const std::vector<int>& /* unused */) {
         // Reverse mode diff
         std::vector<array> vjps;
         for (auto arg : argnums) {
             auto scale = arg == 0 ? alpha_ : beta_;
-            auto scale_arr = array(scale, cotan.dtype());
-            vjps.push_back(multiply(scale_arr, cotan, stream()));
+            auto scale_arr = array(scale, cotangents[0].dtype());
+            vjps.push_back(multiply(scale_arr, cotangents[0], stream()));
         }
         return vjps;
     }
 
-Finally, you need not have a transformation fully defined to start using your 
-own :class:`Primitive`.
+Note, a transformation does not need to be fully defined to start using
+the :class:`Primitive`.
 
 .. code-block:: C++
 
     /** Vectorize primitive along given axis */
-    std::pair<array, int> Axpby::vmap(
+    std::pair<std::vector<array>, std::vector<int>> Axpby::vmap(
             const std::vector<array>& inputs,
             const std::vector<int>& axes) {
-        throw std::runtime_error("Axpby has no vmap implementation.");
+        throw std::runtime_error("[Axpby] vmap not implemented.");
     }
 
 Building and Binding
 --------------------
 
-Let's look at the overall directory structure first. 
+Let's look at the overall directory structure first.
 
 | extensions
 | ├── axpby
@@ -666,40 +625,39 @@ Let's look at the overall directory structure first.
 | └── setup.py
 
 * ``extensions/axpby/`` defines the C++ extension library
-* ``extensions/mlx_sample_extensions`` sets out the structure for the 
-  associated python package
-* ``extensions/bindings.cpp`` provides python bindings for our operation
-* ``extensions/CMakeLists.txt`` holds CMake rules to build the library and 
-  python bindings
+* ``extensions/mlx_sample_extensions`` sets out the structure for the
+  associated Python package
+* ``extensions/bindings.cpp`` provides Python bindings for our operation
+* ``extensions/CMakeLists.txt`` holds CMake rules to build the library and
+  Python bindings
 * ``extensions/setup.py`` holds the ``setuptools`` rules to build and install
-  the python package
+  the Python package
 
 Binding to Python
 ^^^^^^^^^^^^^^^^^^
 
-We use PyBind11_ to build a Python API for the C++ library. Since bindings for
+We use nanobind_ to build a Python API for the C++ library. Since bindings for
 components such as :class:`mlx.core.array`, :class:`mlx.core.stream`, etc. are
-already provided, adding our :meth:`axpby` is simple!
+already provided, adding our :meth:`axpby` is simple.
 
 .. code-block:: C++
 
-    PYBIND11_MODULE(mlx_sample_extensions, m) {
-        m.doc() = "Sample C++ and metal extensions for MLX";
+   NB_MODULE(_ext, m) {
+        m.doc() = "Sample extension for MLX";
 
         m.def(
             "axpby",
             &axpby,
             "x"_a,
             "y"_a,
-            py::pos_only(),
             "alpha"_a,
             "beta"_a,
-            py::kw_only(),
-            "stream"_a = py::none(),
-            R"pbdoc(
+            nb::kw_only(),
+            "stream"_a = nb::none(),
+            R"(
                 Scale and sum two vectors element-wise
                 ``z = alpha * x + beta * y``
-                
+
                 Follows numpy style broadcasting between ``x`` and ``y``
                 Inputs are upcasted to floats if needed
 
@@ -711,17 +669,17 @@ already provided, adding our :meth:`axpby` is simple!
 
                 Returns:
                     array: ``alpha * x + beta * y``
-            )pbdoc");
+            )");
     }
 
-Most of the complexity in the above example comes from additional bells and 
+Most of the complexity in the above example comes from additional bells and
 whistles such as the literal names and doc-strings.
 
 .. warning::
 
-    :mod:`mlx.core` needs to be imported before importing 
-    :mod:`mlx_sample_extensions` as defined by the pybind11 module above to 
-    ensure that the casters for :mod:`mlx.core` components like 
+    :mod:`mlx.core` must be imported before importing
+    :mod:`mlx_sample_extensions` as defined by the nanobind module above to
+    ensure that the casters for :mod:`mlx.core` components like
     :class:`mlx.core.array` are available.
 
 .. _Building with CMake:
@@ -729,8 +687,8 @@ whistles such as the literal names and doc-strings.
 Building with CMake
 ^^^^^^^^^^^^^^^^^^^^
 
-Building the C++ extension library itself is simple, it only requires that you 
-``find_package(MLX CONFIG)`` and then link it to your library. 
+Building the C++ extension library only requires that you ``find_package(MLX
+CONFIG)`` and then link it to your library.
 
 .. code-block:: cmake
 
@@ -752,12 +710,12 @@ Building the C++ extension library itself is simple, it only requires that you
     # Link to mlx
     target_link_libraries(mlx_ext PUBLIC mlx)
 
-We also need to build the attached metal library. For convenience, we provide a 
-:meth:`mlx_build_metallib` function that builds a ``.metallib`` target given 
-sources, headers, destinations, etc. (defined in ``cmake/extension.cmake`` and 
-automatically imported with MLX package). 
+We also need to build the attached Metal library. For convenience, we provide a
+:meth:`mlx_build_metallib` function that builds a ``.metallib`` target given
+sources, headers, destinations, etc. (defined in ``cmake/extension.cmake`` and
+automatically imported with MLX package).
 
-Here is what that looks like in practice!
+Here is what that looks like in practice:
 
 .. code-block:: cmake
 
@@ -779,27 +737,29 @@ Here is what that looks like in practice!
 
     endif()
 
-Finally, we build the Pybind11_ bindings
+Finally, we build the nanobind_ bindings
 
 .. code-block:: cmake
 
-    pybind11_add_module(
-        mlx_sample_extensions
-        ${CMAKE_CURRENT_LIST_DIR}/bindings.cpp
+    nanobind_add_module(
+      _ext
+      NB_STATIC STABLE_ABI LTO NOMINSIZE
+      NB_DOMAIN mlx
+      ${CMAKE_CURRENT_LIST_DIR}/bindings.cpp
     )
-    target_link_libraries(mlx_sample_extensions PRIVATE mlx_ext)
+    target_link_libraries(_ext PRIVATE mlx_ext)
 
     if(BUILD_SHARED_LIBS)
-        target_link_options(mlx_sample_extensions PRIVATE -Wl,-rpath,@loader_path)
+      target_link_options(_ext PRIVATE -Wl,-rpath,@loader_path)
     endif()
 
 Building with ``setuptools``
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Once we have set out the CMake build rules as described above, we can use the
-build utilities defined in :mod:`mlx.extension` for a simple build process. 
+build utilities defined in :mod:`mlx.extension`:
 
-.. code-block:: python 
+.. code-block:: python
 
     from mlx import extension
     from setuptools import setup
@@ -809,48 +769,50 @@ build utilities defined in :mod:`mlx.extension` for a simple build process.
             name="mlx_sample_extensions",
             version="0.0.0",
             description="Sample C++ and Metal extensions for MLX primitives.",
-            ext_modules=[extension.CMakeExtension("mlx_sample_extensions")],
+            ext_modules=[extension.CMakeExtension("mlx_sample_extensions._ext")],
             cmdclass={"build_ext": extension.CMakeBuild},
-            packages = ["mlx_sample_extensions"],
-            package_dir = {"": "mlx_sample_extensions"},
-            package_data = {"mlx_sample_extensions" : ["*.so", "*.dylib", "*.metallib"]},
+            packages=["mlx_sample_extensions"],
+            package_data={"mlx_sample_extensions": ["*.so", "*.dylib", "*.metallib"]},
+            extras_require={"dev":[]},
             zip_safe=False,
-            python_requires=">=3.7",
+            python_requires=">=3.8",
         )
 
 .. note::
     We treat ``extensions/mlx_sample_extensions`` as the package directory
     even though it only contains a ``__init__.py`` to ensure the following:
-    
-    * :mod:`mlx.core` is always imported before importing  :mod:`mlx_sample_extensions`
-    * The C++ extension library and the metal library are co-located with the python 
-      bindings and copied together if the package is installed 
 
-You can build inplace for development using
+    * :mod:`mlx.core` must be imported before importing :mod:`_ext`
+    * The C++ extension library and the metal library are co-located with the python
+      bindings and copied together if the package is installed
+
+To build the package, first install the build dependencies with ``pip install
+-r requirements.txt``.  You can then build inplace for development using
 ``python setup.py build_ext -j8 --inplace`` (in ``extensions/``)
 
-This will result in a directory structure as follows:
+This results in the directory structure:
 
 | extensions
 | ├── mlx_sample_extensions
 | │   ├── __init__.py
 | │   ├── libmlx_ext.dylib # C++ extension library
 | │   ├── mlx_ext.metallib # Metal library
-| │   └── mlx_sample_extensions.cpython-3x-darwin.so # Python Binding
+| │   └── _ext.cpython-3x-darwin.so # Python Binding
 | ...
 
-When you try to install using the command ``python -m pip install .`` 
-(in ``extensions/``), the package will be installed with the same structure as 
-``extensions/mlx_sample_extensions`` and the C++ and metal library will be 
-copied along with the python binding since they are specified as ``package_data``.
+When you try to install using the command ``python -m pip install .`` (in
+``extensions/``), the package will be installed with the same structure as
+``extensions/mlx_sample_extensions`` and the C++ and Metal library will be
+copied along with the Python binding since they are specified as
+``package_data``.
 
 Usage
 -----
 
-After installing the extension as described above, you should be able to simply 
-import the python package and play with it as you would any other MLX operation!
+After installing the extension as described above, you should be able to simply
+import the Python package and play with it as you would any other MLX operation.
 
-Let's looks at a simple script and it's results!
+Let's look at a simple script and its results:
 
 .. code-block:: python
 
@@ -874,12 +836,12 @@ Output:
     c correctness: True
 
 Results
-^^^^^^^^^^^^^^^^
+^^^^^^^
 
-Let's run a quick benchmark and see how our new ``axpby`` operation compares 
-with the naive :meth:`simple_axpby` we defined at first on the CPU. 
+Let's run a quick benchmark and see how our new ``axpby`` operation compares
+with the naive :meth:`simple_axpby` we first defined on the CPU.
 
-.. code-block:: python 
+.. code-block:: python
 
     import mlx.core as mx
     from mlx_sample_extensions import axpby
@@ -898,7 +860,7 @@ with the naive :meth:`simple_axpby` we defined at first on the CPU.
     alpha = 4.0
     beta = 2.0
 
-    mx.eval((x, y))
+    mx.eval(x, y)
 
     def bench(f):
         # Warm up
@@ -919,30 +881,23 @@ with the naive :meth:`simple_axpby` we defined at first on the CPU.
 
     print(f"Simple axpby: {simple_time:.3f} s | Custom axpby: {custom_time:.3f} s")
 
-Results:
-
-.. code-block::
-
-    Simple axpby: 0.114 s | Custom axpby: 0.109 s
-
-We see some modest improvements right away! 
+The results are ``Simple axpby: 0.114 s | Custom axpby: 0.109 s``. We see
+modest improvements right away!
 
 This operation is now good to be used to build other operations, in
 :class:`mlx.nn.Module` calls, and also as a part of graph transformations like
-:meth:`grad`!
+:meth:`grad`.
 
 Scripts
 -------
 
 .. admonition:: Download the code
 
-   The full example code is available in `mlx <code>`_.
-
-.. code: `https://github.com/ml-explore/mlx/tree/main/examples/extensions/`_
+   The full example code is available in `mlx <https://github.com/ml-explore/mlx/tree/main/examples/extensions/>`_.
 
 .. _Accelerate: https://developer.apple.com/documentation/accelerate/blas?language=objc
 .. _Metal: https://developer.apple.com/documentation/metal?language=objc
 .. _Metal-cpp: https://developer.apple.com/metal/cpp/
 .. _`Metal Specification`: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf
 .. _`Metal Example`: https://developer.apple.com/documentation/metal/performing_calculations_on_a_gpu?language=objc
-.. _PyBind11: https://pybind11.readthedocs.io/en/stable/
+.. _nanobind: https://nanobind.readthedocs.io/en/latest/

docs/src/dev/metal_debugger.rst

@@ -0,0 +1,69 @@
+Metal Debugger
+==============
+
+.. currentmodule:: mlx.core
+
+Profiling is a key step for performance optimization. You can build MLX with
+the ``MLX_METAL_DEBUG`` option to improve the Metal debugging and
+optimization workflow. The ``MLX_METAL_DEBUG`` debug option:
+
+* Records source during Metal compilation, for later inspection while
+  debugging.
+* Labels Metal objects such as command queues, improving capture readability.
+
+To build with debugging enabled in Python prepend
+``CMAKE_ARGS="-DMLX_METAL_DEBUG=ON"`` to the build call.
+
+The :func:`metal.start_capture` function initiates a capture of all MLX GPU
+work.
+
+.. note::
+
+   To capture a GPU trace you must run the application with
+   ``MTL_CAPTURE_ENABLED=1``.
+
+.. code-block:: python
+
+    import mlx.core as mx
+
+    a = mx.random.uniform(shape=(512, 512))
+    b = mx.random.uniform(shape=(512, 512))
+    mx.eval(a, b)
+
+    trace_file = "mlx_trace.gputrace"
+
+    if not mx.metal.start_capture(trace_file):
+      print("Make sure to run with MTL_CAPTURE_ENABLED=1 and "
+            f"that the path {trace_file} does not already exist.")
+      exit(1)
+
+    for _ in range(10):
+      mx.eval(mx.add(a, b))
+
+    mx.metal.stop_capture()
+
+You can open and replay the GPU trace in Xcode. The ``Dependencies`` view
+has a great overview of all operations. Checkout the `Metal debugger
+documentation`_ for more information.
+
+.. image:: ../_static/metal_debugger/capture.png
+    :class: dark-light
+
+Xcode Workflow
+--------------
+
+You can skip saving to a path by running within Xcode. First, generate an
+Xcode project using CMake.
+
+.. code-block::
+
+    mkdir build && cd build
+    cmake .. -DMLX_METAL_DEBUG=ON -G Xcode
+    open mlx.xcodeproj
+
+Select the ``metal_capture`` example schema and run.
+
+.. image:: ../_static/metal_debugger/schema.png
+    :class: dark-light
+
+.. _`Metal debugger documentation`: https://developer.apple.com/documentation/xcode/metal-debugger

docs/src/index.rst

@@ -58,6 +58,7 @@ are the CPU and GPU.
    :maxdepth: 1
 
    python/array
+   python/data_types
    python/devices_and_streams
    python/ops
    python/random
@@ -81,3 +82,4 @@ are the CPU and GPU.
    :maxdepth: 1
 
    dev/extensions
+   dev/metal_debugger

docs/src/install.rst

@@ -155,6 +155,8 @@ should point to the path to the built metal library.
      - ON
    * - MLX_BUILD_PYTHON_BINDINGS
      - OFF
+   * - MLX_METAL_DEBUG
+     - OFF
 
 
 .. note::

docs/src/python/array.rst

@@ -19,7 +19,6 @@ Array
     array.ndim
     array.shape
     array.size
-    Dtype
     array.abs
     array.all
     array.any
@@ -32,7 +31,6 @@ Array
     array.cumsum
     array.diag
     array.diagonal
-    array.dtype
     array.exp
     array.flatten
     array.log

docs/src/python/data_types.rst

@@ -1,7 +1,5 @@
 .. _data_types:
 
-:orphan:
-
 Data Types
 ==========
 
@@ -56,3 +54,15 @@ The default floating point type is ``float32`` and the default integer type is
    * - ``complex64``
      - 8 
      - 64-bit complex float
+
+
+Data type are aranged in a hierarchy. See the :obj:`DtypeCategory` object
+documentation for more information. Use :func:`issubdtype` to determine if one
+``dtype`` (or category) is a subtype of another category.
+
+.. autosummary::
+   :toctree: _autosummary
+
+   Dtype
+   DtypeCategory
+   issubdtype

docs/src/python/metal.rst

@@ -3,7 +3,7 @@ Metal
 
 .. currentmodule:: mlx.core.metal
 
-.. autosummary:: 
+.. autosummary::
   :toctree: _autosummary
 
   is_available
@@ -12,3 +12,5 @@ Metal
   get_cache_memory
   set_memory_limit
   set_cache_limit
+  start_capture
+  stop_capture

docs/src/python/nn/module.rst

@@ -30,6 +30,7 @@ Module
       Module.named_modules
       Module.parameters
       Module.save_weights
+      Module.set_dtype
       Module.train
       Module.trainable_parameters
       Module.unfreeze

docs/src/python/ops.rst

@@ -5,13 +5,13 @@ Operations
 
 .. currentmodule:: mlx.core
 
-.. autosummary:: 
+.. autosummary::
   :toctree: _autosummary
 
    abs
    add
    all
-   allclose 
+   allclose
    any
    arange
    arccos
@@ -51,6 +51,7 @@ Operations
    erf
    erfinv
    exp
+   expm1
    expand_dims
    eye
    flatten
@@ -62,10 +63,10 @@ Operations
    identity
    inner
    isclose
-   isnan
-   isposinf
-   isneginf
    isinf
+   isnan
+   isneginf
+   isposinf
    less
    less_equal
    linspace
@@ -83,6 +84,7 @@ Operations
    max
    maximum
    mean
+   meshgrid
    min
    minimum
    moveaxis
@@ -117,6 +119,7 @@ Operations
    square
    squeeze
    stack
+   std
    stop_gradient
    subtract
    sum

docs/src/python/random.rst

@@ -38,6 +38,7 @@ we use a splittable version of Threefry, which is a counter-based PRNG.
    gumbel
    key
    normal
+   multivariate_normal
    randint
    seed
    split

docs/src/usage/saving_and_loading.rst

@@ -49,7 +49,7 @@ it will be added. You can load the array with:
 
 .. code-block:: shell
 
-   >>> mx.load("array.npy", a)
+   >>> mx.load("array.npy")
    array([1], dtype=float32)
 
 Here's an example of saving several arrays to a single file:

examples/cpp/CMakeLists.txt

@@ -8,3 +8,4 @@ endfunction(build_example)
 build_example(tutorial.cpp)
 build_example(linear_regression.cpp)
 build_example(logistic_regression.cpp)
+build_example(metal_capture.cpp)

examples/cpp/metal_capture.cpp

@@ -0,0 +1,31 @@
+// Copyright © 2024 Apple Inc.
+
+#include <cassert>
+#include <iostream>
+
+#include "mlx/mlx.h"
+
+using namespace mlx::core;
+
+int main() {
+  // To use Metal debugging and profiling:
+  // 1. Build with the MLX_METAL_DEBUG CMake option (i.e. -DMLX_METAL_DEBUG=ON).
+  // 2. Run with MTL_CAPTURE_ENABLED=1.
+  assert(metal::start_capture("mlx_trace.gputrace"));
+
+  // Start at index two because the default GPU and CPU streams have indices
+  // zero and one, respectively. This naming matches the label assigned to each
+  // stream's command queue.
+  auto s2 = new_stream(Device::gpu);
+  auto s3 = new_stream(Device::gpu);
+
+  auto a = arange(1.f, 10.f, 1.f, float32, s2);
+  auto b = arange(1.f, 10.f, 1.f, float32, s3);
+  auto x = add(a, a, s2);
+  auto y = add(b, b, s3);
+
+  // The multiply will happen on the default stream.
+  std::cout << multiply(x, y) << std::endl;
+
+  metal::stop_capture();
+}

examples/extensions/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.27)
 
-project(mlx_sample_extensions LANGUAGES CXX)
+project(_ext LANGUAGES CXX)
 
 # ----------------------------- Setup -----------------------------
 set(CMAKE_CXX_STANDARD 17)
@@ -11,8 +11,12 @@ option(BUILD_SHARED_LIBS "Build extensions as a shared library" ON)
 
 # ----------------------------- Dependencies -----------------------------
 find_package(MLX CONFIG REQUIRED)
-find_package(Python COMPONENTS Interpreter Development)
-find_package(pybind11 CONFIG REQUIRED)
+find_package(Python 3.8 COMPONENTS Interpreter Development.Module REQUIRED)
+execute_process(
+  COMMAND "${Python_EXECUTABLE}" -m nanobind --cmake_dir
+  OUTPUT_STRIP_TRAILING_WHITESPACE OUTPUT_VARIABLE NB_DIR)
+list(APPEND CMAKE_PREFIX_PATH "${NB_DIR}")
+find_package(nanobind CONFIG REQUIRED)
 
 # ----------------------------- Extensions -----------------------------
 
@@ -38,7 +42,6 @@ target_link_libraries(mlx_ext PUBLIC mlx)
 
 # Build metallib
 if(MLX_BUILD_METAL)
-
   mlx_build_metallib(
     TARGET mlx_ext_metallib
     TITLE mlx_ext
@@ -54,13 +57,15 @@ if(MLX_BUILD_METAL)
 
 endif()
 
-# ----------------------------- Pybind -----------------------------
-pybind11_add_module(
-  mlx_sample_extensions
+# ----------------------------- Python Bindings -----------------------------
+nanobind_add_module(
+  _ext
+  NB_STATIC STABLE_ABI LTO NOMINSIZE
+  NB_DOMAIN mlx 
   ${CMAKE_CURRENT_LIST_DIR}/bindings.cpp
 )
-target_link_libraries(mlx_sample_extensions PRIVATE mlx_ext)
+target_link_libraries(_ext PRIVATE mlx_ext)
 
 if(BUILD_SHARED_LIBS)
-  target_link_options(mlx_sample_extensions PRIVATE -Wl,-rpath,@loader_path)
+  target_link_options(_ext PRIVATE -Wl,-rpath,@loader_path)
 endif()

examples/extensions/README.md

@@ -0,0 +1,18 @@
+
+## Build the extensions
+
+```
+pip install -e .
+```
+
+For faster builds during development, you can also pre-install the requirements:
+
+```
+pip install -r requirements.txt
+```
+
+And then run:
+
+```
+python setup.py build_ext -j8 --inplace
+```

examples/extensions/axpby/axpby.cpp

@@ -1,4 +1,4 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #include <cassert>
 #include <iostream>
@@ -43,7 +43,7 @@ array axpby(
   auto promoted_dtype = promote_types(x.dtype(), y.dtype());
 
   // Upcast to float32 for non-floating point inputs x and y
-  auto out_dtype = is_floating_point(promoted_dtype)
+  auto out_dtype = issubdtype(promoted_dtype, float32)
       ? promoted_dtype
       : promote_types(promoted_dtype, float32);
 
@@ -61,7 +61,7 @@ array axpby(
       /* const std::vector<int>& shape = */ out_shape,
       /* Dtype dtype = */ out_dtype,
       /* std::unique_ptr<Primitive> primitive = */
-      std::make_unique<Axpby>(to_stream(s), alpha, beta),
+      std::make_shared<Axpby>(to_stream(s), alpha, beta),
       /* const std::vector<array>& inputs = */ broadcasted_inputs);
 }
 
@@ -106,12 +106,12 @@ void axpby_impl(
 /** Fall back implementation for evaluation on CPU */
 void Axpby::eval(
     const std::vector<array>& inputs,
-    std::vector<array>& out_arr) {
-  auto out = out_arr[0];
+    std::vector<array>& outputs) {
   // Check the inputs (registered in the op while constructing the out array)
   assert(inputs.size() == 2);
   auto& x = inputs[0];
   auto& y = inputs[1];
+  auto& out = outputs[0];
 
   // Dispatch to the correct dtype
   if (out.dtype() == float32) {
@@ -150,11 +150,7 @@ void axpby_impl_accelerate(
   // The data in the output array is allocated to match the strides in y
   // such that x, y, and out are contiguous in the same mode and
   // no transposition is needed
-  out.set_data(
-      allocator::malloc_or_wait(y.data_size() * out.itemsize()),
-      y.data_size(),
-      y.strides(),
-      y.flags());
+  out.set_data(allocator::malloc_or_wait(out.nbytes()));
 
   // We then copy over the elements using the contiguous vector specialization
   copy_inplace(y, out, CopyType::Vector);
@@ -180,11 +176,11 @@ void axpby_impl_accelerate(
 /** Evaluate primitive on CPU using accelerate specializations */
 void Axpby::eval_cpu(
     const std::vector<array>& inputs,
-    std::vector<array>& outarr) {
-  auto out = outarr[0];
+    std::vector<array>& outputs) {
   assert(inputs.size() == 2);
   auto& x = inputs[0];
   auto& y = inputs[1];
+  auto& out = outputs[0];
 
   // Accelerate specialization for contiguous single precision float arrays
   if (out.dtype() == float32 &&
@@ -195,7 +191,7 @@ void Axpby::eval_cpu(
   }
 
   // Fall back to common backend if specializations are not available
-  eval(inputs, outarr);
+  eval(inputs, outputs);
 }
 
 #else // Accelerate not available
@@ -203,8 +199,8 @@ void Axpby::eval_cpu(
 /** Evaluate primitive on CPU falling back to common backend */
 void Axpby::eval_cpu(
     const std::vector<array>& inputs,
-    std::vector<array>& out) {
-  eval(inputs, out);
+    const std::vector<array>& outputs) {
+  eval(inputs, outputs);
 }
 
 #endif
@@ -218,12 +214,12 @@ void Axpby::eval_cpu(
 /** Evaluate primitive on GPU */
 void Axpby::eval_gpu(
     const std::vector<array>& inputs,
-    std::vector<array>& outarr) {
+    std::vector<array>& outputs) {
   // Prepare inputs
-  auto out = outarr[0];
   assert(inputs.size() == 2);
   auto& x = inputs[0];
   auto& y = inputs[1];
+  auto& out = outputs[0];
 
   // Each primitive carries the stream it should execute on
   // and each stream carries its device identifiers
@@ -372,4 +368,4 @@ bool Axpby::is_equivalent(const Primitive& other) const {
   return alpha_ == r_other.alpha_ && beta_ == r_other.beta_;
 }
 
-} // namespace mlx::core
+} // namespace mlx::core

examples/extensions/axpby/axpby.h

@@ -42,9 +42,9 @@ class Axpby : public Primitive {
    * To avoid unnecessary allocations, the evaluation function
    * is responsible for allocating space for the array.
    */
-  void eval_cpu(const std::vector<array>& inputs, std::vector<array>& out)
+  void eval_cpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override;
-  void eval_gpu(const std::vector<array>& inputs, std::vector<array>& out)
+  void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override;
 
   /** The Jacobian-vector product. */
@@ -83,7 +83,7 @@ class Axpby : public Primitive {
   float beta_;
 
   /** Fall back implementation for evaluation on CPU */
-  void eval(const std::vector<array>& inputs, std::vector<array>& out);
+  void eval(const std::vector<array>& inputs, std::vector<array>& outputs);
 };
 
-} // namespace mlx::core
+} // namespace mlx::core

examples/extensions/bindings.cpp

@@ -1,31 +1,31 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
-#include <pybind11/pybind11.h>
-#include <pybind11/stl.h>
+#include <nanobind/nanobind.h>
+#include <nanobind/stl/variant.h>
 
 #include "axpby/axpby.h"
 
-namespace py = pybind11;
-using namespace py::literals;
+namespace nb = nanobind;
+using namespace nb::literals;
+
 using namespace mlx::core;
 
-PYBIND11_MODULE(mlx_sample_extensions, m) {
-  m.doc() = "Sample C++ and metal extensions for MLX";
+NB_MODULE(_ext, m) {
+  m.doc() = "Sample extension for MLX";
 
   m.def(
       "axpby",
       &axpby,
       "x"_a,
       "y"_a,
-      py::pos_only(),
       "alpha"_a,
       "beta"_a,
-      py::kw_only(),
-      "stream"_a = py::none(),
-      R"pbdoc(
+      nb::kw_only(),
+      "stream"_a = nb::none(),
+      R"(
         Scale and sum two vectors element-wise
         ``z = alpha * x + beta * y``
-        
+
         Follows numpy style broadcasting between ``x`` and ``y``
         Inputs are upcasted to floats if needed
 
@@ -37,5 +37,5 @@ PYBIND11_MODULE(mlx_sample_extensions, m) {
 
         Returns:
             array: ``alpha * x + beta * y``
-      )pbdoc");
-}
+      )");
+}

examples/extensions/pyproject.toml

@@ -1,3 +1,8 @@
 [build-system]
-requires = ["setuptools>=42", "pybind11>=2.10", "cmake>=3.24", "mlx @ git+https://github.com/mlx-explore/mlx@main"]
-build-backend = "setuptools.build_meta"
+requires = [
+  "setuptools>=42",
+  "cmake>=3.24",
+  "mlx>=0.9.0",
+  "nanobind@git+https://github.com/wjakob/nanobind.git#egg=4148debcf91f5ccab0c3b8d67b5c3cabd61f407f",
+]
+build-backend = "setuptools.build_meta"

examples/extensions/requirements.txt

@@ -0,0 +1,4 @@
+setuptools>=42
+cmake>=3.24
+mlx>=0.9.0
+nanobind@git+https://github.com/wjakob/nanobind.git#egg=4148debcf91f5ccab0c3b8d67b5c3cabd61f407f

examples/extensions/setup.py

@@ -1,4 +1,4 @@
-# Copyright © 2023 Apple Inc.
+# Copyright © 2023-2024 Apple Inc.
 
 from setuptools import setup
 
@@ -9,11 +9,11 @@ if __name__ == "__main__":
         name="mlx_sample_extensions",
         version="0.0.0",
         description="Sample C++ and Metal extensions for MLX primitives.",
-        ext_modules=[extension.CMakeExtension("mlx_sample_extensions")],
+        ext_modules=[extension.CMakeExtension("mlx_sample_extensions._ext")],
         cmdclass={"build_ext": extension.CMakeBuild},
         packages=["mlx_sample_extensions"],
-        package_dir={"": "."},
         package_data={"mlx_sample_extensions": ["*.so", "*.dylib", "*.metallib"]},
+        extras_require={"dev": []},
         zip_safe=False,
         python_requires=">=3.8",
     )

mlx/array.cpp

@@ -93,7 +93,9 @@ void array::detach() {
 }
 
 void array::eval() {
-  mlx::core::eval({*this});
+  if (!is_evaled()) {
+    mlx::core::eval({*this});
+  }
 }
 
 bool array::is_tracer() const {
@@ -190,6 +192,36 @@ array::ArrayDesc::ArrayDesc(
   init();
 }
 
+array::ArrayDesc::~ArrayDesc() {
+  // When an array description is destroyed it will delete a bunch of arrays
+  // that may also destory their corresponding descriptions and so on and so
+  // forth.
+  //
+  // This calls recursively the destructor and can result in stack overflow, we
+  // instead put them in a vector and destroy them one at a time resulting in a
+  // max stack depth of 2.
+  std::vector<std::shared_ptr<ArrayDesc>> for_deletion;
+
+  for (array& a : inputs) {
+    if (a.array_desc_.use_count() == 1) {
+      for_deletion.push_back(std::move(a.array_desc_));
+    }
+  }
+
+  while (!for_deletion.empty()) {
+    // top is going to be deleted at the end of the block *after* the arrays
+    // with inputs have been moved into the vector
+    auto top = std::move(for_deletion.back());
+    for_deletion.pop_back();
+
+    for (array& a : top->inputs) {
+      if (a.array_desc_.use_count() == 1) {
+        for_deletion.push_back(std::move(a.array_desc_));
+      }
+    }
+  }
+}
+
 array::ArrayIterator::ArrayIterator(const array& arr, int idx)
     : arr(arr), idx(idx) {
   if (arr.ndim() == 0) {

mlx/array.h

@@ -256,6 +256,17 @@ class array {
     array_desc_->position = position;
   }
 
+  /** The i-th output of the array's primitive. */
+  const array& output(int i) const {
+    if (i == array_desc_->position) {
+      return *this;
+    } else if (i < array_desc_->position) {
+      return siblings()[i];
+    } else {
+      return siblings()[i + 1];
+    }
+  };
+
   /** The outputs of the array's primitive (i.e. this array and
    * its siblings) in the order the primitive expects. */
   std::vector<array> outputs() const {
@@ -393,6 +404,8 @@ class array {
         std::shared_ptr<Primitive> primitive,
         std::vector<array> inputs);
 
+    ~ArrayDesc();
+
    private:
     // Initialize size, strides, and other metadata
     void init();
@@ -510,4 +523,15 @@ void array::init(It src) {
   }
 }
 
+/* Utilities for determining whether a template parameter is array. */
+template <typename T>
+inline constexpr bool is_array_v =
+    std::is_same_v<std::remove_cv_t<std::remove_reference_t<T>>, array>;
+
+template <typename... T>
+inline constexpr bool is_arrays_v = (is_array_v<T> && ...);
+
+template <typename... T>
+using enable_for_arrays_t = typename std::enable_if_t<is_arrays_v<T...>>;
+
 } // namespace mlx::core

mlx/backend/accelerate/primitives.cpp

@@ -301,7 +301,7 @@ void Exp::eval_cpu(const std::vector<array>& inputs, array& out) {
     set_unary_output_data(in, out);
     auto size = in.data_size();
     vvexpf(out.data<float>(), in.data<float>(), reinterpret_cast<int*>(&size));
-  } else if (is_floating_point(out.dtype())) {
+  } else if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, [](auto x) { return std::exp(x); });
   } else {
     throw std::invalid_argument(
@@ -310,6 +310,19 @@ void Exp::eval_cpu(const std::vector<array>& inputs, array& out) {
   }
 }
 
+void Expm1::eval_cpu(const std::vector<array>& inputs, array& out) {
+  assert(inputs.size() == 1);
+  const auto& in = inputs[0];
+  if (out.dtype() == float32 && in.flags().contiguous) {
+    set_unary_output_data(in, out);
+    auto size = in.data_size();
+    vvexpm1f(
+        out.data<float>(), in.data<float>(), reinterpret_cast<int*>(&size));
+  } else {
+    eval(inputs, out);
+  }
+}
+
 void Full::eval_cpu(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   auto& in = inputs[0];
@@ -355,7 +368,7 @@ void Log1p::eval_cpu(const std::vector<array>& inputs, array& out) {
     auto size = in.data_size();
     vvlog1pf(
         out.data<float>(), in.data<float>(), reinterpret_cast<int*>(&size));
-  } else if (is_floating_point(out.dtype())) {
+  } else if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, [](auto x) { return std::log1p(x); });
   } else {
     throw std::invalid_argument(

mlx/backend/accelerate/softmax.cpp

@@ -1,4 +1,4 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #include <cassert>
 #include <limits>
@@ -201,7 +201,7 @@ struct NeonFp16SimdOps {
   }
 };
 
-template <typename T, typename VT, typename Ops, int N>
+template <typename T, typename AccT, typename VT, typename Ops, int N>
 void softmax(const array& in, array& out) {
   Ops ops;
 
@@ -218,13 +218,21 @@ void softmax(const array& in, array& out) {
     VT vmaximum = ops.init(-std::numeric_limits<float>::infinity());
     size_t s = M;
     while (s >= N) {
-      vmaximum = ops.max(ops.load(current_in_ptr), vmaximum);
+      VT vals;
+      if constexpr (std::is_same<T, AccT>::value) {
+        vals = ops.load(current_in_ptr);
+      } else {
+        for (int i = 0; i < N; ++i) {
+          vals[i] = static_cast<AccT>(current_in_ptr[i]);
+        }
+      }
+      vmaximum = ops.max(vals, vmaximum);
       current_in_ptr += N;
       s -= N;
     }
-    T maximum = ops.reduce_max(vmaximum);
+    AccT maximum = ops.reduce_max(vmaximum);
     while (s-- > 0) {
-      maximum = std::max(maximum, *current_in_ptr);
+      maximum = std::max(maximum, static_cast<AccT>(*current_in_ptr));
       current_in_ptr++;
     }
 
@@ -234,18 +242,29 @@ void softmax(const array& in, array& out) {
     current_in_ptr = in_ptr;
     s = M;
     while (s >= N) {
-      VT vexp = ops.exp(ops.sub(*(VT*)current_in_ptr, maximum));
-      ops.store(current_out_ptr, vexp);
-      *(VT*)current_out_ptr = vexp;
+      VT vexp;
+      if constexpr (std::is_same<T, AccT>::value) {
+        vexp = ops.load(current_in_ptr);
+      } else {
+        for (int i = 0; i < N; ++i) {
+          vexp[i] = static_cast<AccT>(current_in_ptr[i]);
+        }
+      }
+      vexp = ops.exp(ops.sub(vexp, maximum));
+      if constexpr (std::is_same<T, AccT>::value) {
+        ops.store(current_out_ptr, vexp);
+      }
       vnormalizer = ops.add(vnormalizer, vexp);
       current_in_ptr += N;
       current_out_ptr += N;
       s -= N;
     }
-    T normalizer = ops.reduce_add(vnormalizer);
+    AccT normalizer = ops.reduce_add(vnormalizer);
     while (s-- > 0) {
-      T _exp = std::exp(*current_in_ptr - maximum);
-      *current_out_ptr = _exp;
+      AccT _exp = std::exp(*current_in_ptr - maximum);
+      if (std::is_same<T, AccT>::value) {
+        *current_out_ptr = _exp;
+      }
       normalizer += _exp;
       current_in_ptr++;
       current_out_ptr++;
@@ -254,14 +273,33 @@ void softmax(const array& in, array& out) {
 
     // Normalize
     current_out_ptr = out_ptr;
+    current_in_ptr = in_ptr;
     s = M;
     while (s >= N) {
-      ops.store(current_out_ptr, ops.mul(*(VT*)current_out_ptr, normalizer));
+      if constexpr (std::is_same<T, AccT>::value) {
+        ops.store(current_out_ptr, ops.mul(*(VT*)current_out_ptr, normalizer));
+      } else {
+        VT vexp;
+        for (int i = 0; i < N; ++i) {
+          vexp[i] = static_cast<AccT>(current_in_ptr[i]);
+        }
+        vexp = ops.mul(ops.exp(ops.sub(vexp, maximum)), normalizer);
+        for (int i = 0; i < N; ++i) {
+          current_out_ptr[i] = vexp[i];
+        }
+        current_in_ptr += N;
+      }
       current_out_ptr += N;
       s -= N;
     }
     while (s-- > 0) {
-      *current_out_ptr *= normalizer;
+      if constexpr (std::is_same<T, AccT>::value) {
+        *current_out_ptr *= normalizer;
+      } else {
+        AccT _exp = std::exp(*current_in_ptr - maximum);
+        *current_out_ptr = static_cast<T>(_exp * normalizer);
+        current_in_ptr++;
+      }
       current_out_ptr++;
     }
   }
@@ -308,15 +346,29 @@ void Softmax::eval_cpu(const std::vector<array>& inputs, array& out) {
           "Softmax is defined only for floating point types");
       break;
     case float32:
-      softmax<float, simd_float16, AccelerateSimdOps<float, simd_float16>, 16>(
-          in, out);
+      softmax<
+          float,
+          float,
+          simd_float16,
+          AccelerateSimdOps<float, simd_float16>,
+          16>(in, out);
       break;
     case float16:
-      softmax<
-          float16_t,
-          float16x8_t,
-          NeonFp16SimdOps<float16_t, float16x8_t>,
-          8>(in, out);
+      if (precise_) {
+        softmax<
+            float16_t,
+            float,
+            simd_float16,
+            AccelerateSimdOps<float, simd_float16>,
+            16>(in, out);
+      } else {
+        softmax<
+            float16_t,
+            float16_t,
+            float16x8_t,
+            NeonFp16SimdOps<float16_t, float16x8_t>,
+            8>(in, out);
+      }
       break;
     case bfloat16:
       eval(inputs, out);

mlx/backend/common/binary.cpp

@@ -179,18 +179,16 @@ void LogAddExp::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 2);
   auto& a = inputs[0];
   auto& b = inputs[1];
-  if (is_floating_point(out.dtype())) {
-    if (out.dtype() == float32) {
-      binary_op<float>(a, b, out, detail::LogAddExp());
-    } else if (out.dtype() == float16) {
-      binary_op<float16_t>(a, b, out, detail::LogAddExp());
-    } else if (out.dtype() == bfloat16) {
-      binary_op<bfloat16_t>(a, b, out, detail::LogAddExp());
-    } else {
-      std::ostringstream err;
-      err << "[logaddexp] Does not support " << out.dtype();
-      throw std::invalid_argument(err.str());
-    }
+  if (out.dtype() == float32) {
+    binary_op<float>(a, b, out, detail::LogAddExp());
+  } else if (out.dtype() == float16) {
+    binary_op<float16_t>(a, b, out, detail::LogAddExp());
+  } else if (out.dtype() == bfloat16) {
+    binary_op<bfloat16_t>(a, b, out, detail::LogAddExp());
+  } else if (issubdtype(out.dtype(), inexact)) {
+    std::ostringstream err;
+    err << "[logaddexp] Does not support " << out.dtype();
+    throw std::invalid_argument(err.str());
   } else {
     throw std::invalid_argument(
         "[logaddexp] Cannot compute logaddexp for arrays with"

mlx/backend/common/compiled.cpp

@@ -126,4 +126,102 @@ std::string build_lib_name(
   return os.str();
 }
 
+bool compiled_check_contiguity(
+    const std::vector<array>& inputs,
+    const std::vector<int>& shape) {
+  bool contiguous = true;
+  bool all_contig = true;
+  bool all_row_contig = true;
+  bool all_col_contig = true;
+  int non_scalar_inputs = 0;
+  for (const auto& x : inputs) {
+    if (is_scalar(x)) {
+      continue;
+    }
+    non_scalar_inputs++;
+    bool shape_eq = x.shape() == shape;
+    all_contig &= (x.flags().contiguous && shape_eq);
+    all_row_contig &= (x.flags().row_contiguous && shape_eq);
+    all_col_contig &= (x.flags().col_contiguous && shape_eq);
+  }
+  if (non_scalar_inputs > 1 && !all_row_contig && !all_col_contig) {
+    contiguous = false;
+  } else if (non_scalar_inputs == 1 && !all_contig) {
+    contiguous = false;
+  } else if (non_scalar_inputs == 0 && !shape.empty()) {
+    contiguous = false;
+  }
+  return contiguous;
+}
+
+void compiled_allocate_outputs(
+    const std::vector<array>& inputs,
+    std::vector<array>& outputs,
+    const std::vector<array>& inputs_,
+    const std::unordered_set<uintptr_t>& constant_ids_,
+    bool contiguous,
+    bool move_buffers /* = false */) {
+  if (contiguous) {
+    int o = 0;
+    std::vector<size_t> strides;
+    size_t data_size;
+    array::Flags flags;
+    for (int i = 0; i < inputs.size() && o < outputs.size(); ++i) {
+      auto& in = inputs[i];
+      // Conditions for donation
+      // - Correct size
+      // - Not a scalar
+      // - Donatable
+      // - Not a constant
+      if (in.itemsize() == outputs[o].itemsize() && !is_scalar(in) &&
+          in.is_donatable() &&
+          constant_ids_.find(inputs_[i].id()) == constant_ids_.end()) {
+        if (move_buffers) {
+          outputs[o++].move_shared_buffer(in);
+        } else {
+          outputs[o++].copy_shared_buffer(in);
+        }
+      }
+      // Get representative input flags to properly set non-donated outputs
+      if (strides.empty() && in.size() == outputs[0].size()) {
+        strides = in.strides();
+        flags = in.flags();
+        data_size = in.data_size();
+      }
+    }
+    for (; o < outputs.size(); ++o) {
+      outputs[o].set_data(
+          allocator::malloc_or_wait(data_size * outputs[o].itemsize()),
+          data_size,
+          strides,
+          flags);
+    }
+  } else {
+    int o = 0;
+    for (int i = 0; i < inputs.size() && o < outputs.size(); ++i) {
+      auto& in = inputs[i];
+      // Conditions for donation
+      // - Row contiguous
+      // - Donatable
+      // - Correct size
+      // - Not a constant
+      if (in.flags().row_contiguous && in.nbytes() == outputs[o].nbytes() &&
+          in.is_donatable() &&
+          constant_ids_.find(inputs_[i].id()) == constant_ids_.end()) {
+        if (move_buffers) {
+          outputs[o].move_shared_buffer(
+              in, outputs[o].strides(), in.flags(), in.data_size());
+        } else {
+          outputs[o].copy_shared_buffer(
+              in, outputs[o].strides(), in.flags(), in.data_size());
+        }
+        o++;
+      }
+    }
+    for (; o < outputs.size(); ++o) {
+      outputs[o].set_data(allocator::malloc_or_wait(outputs[o].nbytes()));
+    }
+  }
+}
+
 } // namespace mlx::core

mlx/backend/common/compiled.h

@@ -53,4 +53,18 @@ inline bool is_scalar(const array& x) {
   return x.ndim() == 0;
 }
 
+// Check if we can use a contiguous operation given inputs and the output shape
+bool compiled_check_contiguity(
+    const std::vector<array>& inputs,
+    const std::vector<int>& shape);
+
+// Allocate space for the outputs possibly with input donation
+void compiled_allocate_outputs(
+    const std::vector<array>& inputs,
+    std::vector<array>& outputs,
+    const std::vector<array>& inputs_,
+    const std::unordered_set<uintptr_t>& constant_ids_,
+    bool contiguous,
+    bool move_buffers = false);
+
 } // namespace mlx::core

mlx/backend/common/compiled_cpu.cpp

@@ -52,8 +52,25 @@ void* compile(
     return nullptr;
   }
 
+  std::string kernel_file_name;
+
+  // Deal with long kernel names. Maximum length for files on macOS is 255
+  // characters. Clip file name with a little extra room and append a 16
+  // character hash.
+  constexpr int max_file_name_length = 245;
+  if (kernel_name.size() > max_file_name_length) {
+    std::ostringstream file_name;
+    file_name
+        << std::string_view(kernel_name).substr(0, max_file_name_length - 16);
+    auto file_id = std::hash<std::string>{}(kernel_name);
+    file_name << "_" << std::hex << std::setw(16) << file_id << std::dec;
+    kernel_file_name = file_name.str();
+  } else {
+    kernel_file_name = kernel_name;
+  }
+
   std::ostringstream shared_lib_name;
-  shared_lib_name << "lib" << kernel_name << ".so";
+  shared_lib_name << "lib" << kernel_file_name << ".so";
   auto shared_lib_path = get_temp_file(shared_lib_name.str());
   bool lib_exists = false;
   {
@@ -64,7 +81,7 @@ void* compile(
   if (!lib_exists) {
     // Open source file and write source code to it
     std::ostringstream source_file_name;
-    source_file_name << kernel_name << ".cpp";
+    source_file_name << kernel_file_name << ".cpp";
     auto source_file_path = get_temp_file(source_file_name.str());
 
     std::ofstream source_file(source_file_path);
@@ -248,28 +265,7 @@ void Compiled::eval_cpu(
 
   // Figure out which kernel we are using
   auto& shape = outputs[0].shape();
-  bool contiguous = true;
-  {
-    bool all_contig = true;
-    bool all_row_contig = true;
-    bool all_col_contig = true;
-    int non_scalar_inputs = 0;
-    for (auto& x : inputs) {
-      if (is_scalar(x)) {
-        continue;
-      }
-      non_scalar_inputs++;
-      bool shape_eq = x.shape() == shape;
-      all_contig &= (x.flags().contiguous && shape_eq);
-      all_row_contig &= (x.flags().row_contiguous && shape_eq);
-      all_col_contig &= (x.flags().col_contiguous && shape_eq);
-    }
-    if (non_scalar_inputs > 1 && !all_row_contig && !all_col_contig) {
-      contiguous = false;
-    } else if (non_scalar_inputs == 1 && !all_contig) {
-      contiguous = false;
-    }
-  }
+  bool contiguous = compiled_check_contiguity(inputs, shape);
 
   // Handle all broadcasting and collect function input arguments
   std::vector<void*> args;
@@ -342,58 +338,8 @@ void Compiled::eval_cpu(
     fn_ptr = compile(kernel_name, kernel.str());
   }
 
-  // Allocate space for the outputs possibly with input donation
-  if (contiguous) {
-    int o = 0;
-    std::vector<size_t> strides;
-    size_t data_size;
-    array::Flags flags;
-    for (int i = 0; i < inputs.size() && o < outputs.size(); ++i) {
-      auto& in = inputs[i];
-      // Conditions for donation
-      // - Contiguous
-      // - Donatable
-      // - Correct size
-      // - Not a constant
-      if (in.flags().contiguous && !is_scalar(in) && in.is_donatable() &&
-          constant_ids_.find(inputs_[i].id()) == constant_ids_.end()) {
-        outputs[o++].copy_shared_buffer(in);
-      }
-      // Get representative input flags to properly set non-donated outputs
-      if (strides.empty() && in.size() == outputs[0].size()) {
-        strides = in.strides();
-        flags = in.flags();
-        data_size = in.data_size();
-      }
-    }
-    for (; o < outputs.size(); ++o) {
-      outputs[o].set_data(
-          allocator::malloc_or_wait(data_size * outputs[o].itemsize()),
-          data_size,
-          strides,
-          flags);
-    }
-  } else {
-    int o = 0;
-    for (int i = 0; i < inputs.size() && o < outputs.size(); ++i) {
-      auto& in = inputs[i];
-      // Conditions for donation
-      // - Row contiguous
-      // - Donatable
-      // - Correct size
-      // - Not a constant
-      if (in.flags().row_contiguous && in.nbytes() == outputs[o].nbytes() &&
-          in.is_donatable() &&
-          constant_ids_.find(inputs_[i].id()) == constant_ids_.end()) {
-        outputs[o].copy_shared_buffer(
-            in, outputs[o].strides(), in.flags(), in.data_size());
-        o++;
-      }
-    }
-    for (; o < outputs.size(); ++o) {
-      outputs[o].set_data(allocator::malloc_or_wait(outputs[o].nbytes()));
-    }
-  }
+  compiled_allocate_outputs(
+      inputs, outputs, inputs_, constant_ids_, contiguous, false);
 
   for (auto& x : outputs) {
     args.push_back(x.data<void>());

mlx/backend/common/copy.cpp

@@ -272,7 +272,7 @@ inline void copy_general_general(const array& src, array& dst) {
 }
 
 template <typename SrcT, typename DstT, typename... Args>
-void copy(const array& src, array& dst, CopyType ctype, Args... args) {
+void copy(const array& src, array& dst, CopyType ctype, Args&&... args) {
   switch (ctype) {
     case CopyType::Scalar:
       copy_single<SrcT, DstT>(src, dst);
@@ -281,54 +281,54 @@ void copy(const array& src, array& dst, CopyType ctype, Args... args) {
       copy_vector<SrcT, DstT>(src, dst);
       return;
     case CopyType::General:
-      copy_general<SrcT, DstT>(src, dst, args...);
+      copy_general<SrcT, DstT>(src, dst, std::forward<Args>(args)...);
       return;
     case CopyType::GeneralGeneral:
-      copy_general_general<SrcT, DstT>(src, dst, args...);
+      copy_general_general<SrcT, DstT>(src, dst, std::forward<Args>(args)...);
   }
 }
 
 template <typename SrcT, typename... Args>
-void copy(const array& src, array& dst, CopyType ctype, Args... args) {
+void copy(const array& src, array& dst, CopyType ctype, Args&&... args) {
   switch (dst.dtype()) {
     case bool_:
-      copy<SrcT, bool>(src, dst, ctype, args...);
+      copy<SrcT, bool>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint8:
-      copy<SrcT, uint8_t>(src, dst, ctype, args...);
+      copy<SrcT, uint8_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint16:
-      copy<SrcT, uint16_t>(src, dst, ctype, args...);
+      copy<SrcT, uint16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint32:
-      copy<SrcT, uint32_t>(src, dst, ctype, args...);
+      copy<SrcT, uint32_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint64:
-      copy<SrcT, uint64_t>(src, dst, ctype, args...);
+      copy<SrcT, uint64_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int8:
-      copy<SrcT, int8_t>(src, dst, ctype, args...);
+      copy<SrcT, int8_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int16:
-      copy<SrcT, int16_t>(src, dst, ctype, args...);
+      copy<SrcT, int16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int32:
-      copy<SrcT, int32_t>(src, dst, ctype, args...);
+      copy<SrcT, int32_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int64:
-      copy<SrcT, int64_t>(src, dst, ctype, args...);
+      copy<SrcT, int64_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case float16:
-      copy<SrcT, float16_t>(src, dst, ctype, args...);
+      copy<SrcT, float16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case float32:
-      copy<SrcT, float>(src, dst, ctype, args...);
+      copy<SrcT, float>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case bfloat16:
-      copy<SrcT, bfloat16_t>(src, dst, ctype, args...);
+      copy<SrcT, bfloat16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case complex64:
-      copy<SrcT, complex64_t>(src, dst, ctype, args...);
+      copy<SrcT, complex64_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
   }
 }
@@ -338,46 +338,46 @@ inline void copy_inplace_dispatch(
     const array& src,
     array& dst,
     CopyType ctype,
-    Args... args) {
+    Args&&... args) {
   switch (src.dtype()) {
     case bool_:
-      copy<bool>(src, dst, ctype, args...);
+      copy<bool>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint8:
-      copy<uint8_t>(src, dst, ctype, args...);
+      copy<uint8_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint16:
-      copy<uint16_t>(src, dst, ctype, args...);
+      copy<uint16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint32:
-      copy<uint32_t>(src, dst, ctype, args...);
+      copy<uint32_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case uint64:
-      copy<uint64_t>(src, dst, ctype, args...);
+      copy<uint64_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int8:
-      copy<int8_t>(src, dst, ctype, args...);
+      copy<int8_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int16:
-      copy<int16_t>(src, dst, ctype, args...);
+      copy<int16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int32:
-      copy<int32_t>(src, dst, ctype, args...);
+      copy<int32_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case int64:
-      copy<int64_t>(src, dst, ctype, args...);
+      copy<int64_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case float16:
-      copy<float16_t>(src, dst, ctype, args...);
+      copy<float16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case float32:
-      copy<float>(src, dst, ctype, args...);
+      copy<float>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case bfloat16:
-      copy<bfloat16_t>(src, dst, ctype, args...);
+      copy<bfloat16_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
     case complex64:
-      copy<complex64_t>(src, dst, ctype, args...);
+      copy<complex64_t>(src, dst, ctype, std::forward<Args>(args)...);
       break;
   }
 }

mlx/backend/common/default_primitives.cpp

@@ -57,6 +57,7 @@ DEFAULT(Equal)
 DEFAULT(Erf)
 DEFAULT(ErfInv)
 DEFAULT(Exp)
+DEFAULT(Expm1)
 DEFAULT(FFT)
 DEFAULT(Floor)
 DEFAULT(Full)

mlx/backend/common/ops.h

@@ -241,6 +241,13 @@ struct Exp {
   }
 };
 
+struct Expm1 {
+  template <typename T>
+  T operator()(T x) {
+    return expm1(x);
+  };
+};
+
 struct Floor {
   template <typename T>
   T operator()(T x) {

mlx/backend/common/primitives.cpp

@@ -22,7 +22,7 @@ namespace mlx::core {
 void Abs::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   auto& in = inputs[0];
-  if (is_unsigned(in.dtype())) {
+  if (issubdtype(in.dtype(), unsignedinteger)) {
     // No-op for unsigned types
     out.copy_shared_buffer(in);
   } else {
@@ -37,7 +37,7 @@ void Arange::eval(const std::vector<array>& inputs, array& out) {
 void ArcCos::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::ArcCos());
   } else {
     throw std::invalid_argument(
@@ -49,7 +49,7 @@ void ArcCos::eval(const std::vector<array>& inputs, array& out) {
 void ArcCosh::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::ArcCosh());
   } else {
     throw std::invalid_argument(
@@ -61,7 +61,7 @@ void ArcCosh::eval(const std::vector<array>& inputs, array& out) {
 void ArcSin::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::ArcSin());
   } else {
     throw std::invalid_argument(
@@ -73,7 +73,7 @@ void ArcSin::eval(const std::vector<array>& inputs, array& out) {
 void ArcSinh::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::ArcSinh());
   } else {
     throw std::invalid_argument(
@@ -85,7 +85,7 @@ void ArcSinh::eval(const std::vector<array>& inputs, array& out) {
 void ArcTan::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::ArcTan());
   } else {
     throw std::invalid_argument(
@@ -97,7 +97,7 @@ void ArcTan::eval(const std::vector<array>& inputs, array& out) {
 void ArcTanh::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::ArcTanh());
   } else {
     throw std::invalid_argument(
@@ -171,7 +171,7 @@ void Broadcast::eval(const std::vector<array>& inputs, array& out) {
 void Ceil::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   auto& in = inputs[0];
-  if (not is_integral(in.dtype())) {
+  if (issubdtype(in.dtype(), inexact)) {
     unary_fp(in, out, detail::Ceil());
   } else {
     // No-op integer types
@@ -211,7 +211,7 @@ void Copy::eval(const std::vector<array>& inputs, array& out) {
 void Cos::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Cos());
   } else {
     throw std::invalid_argument(
@@ -223,7 +223,7 @@ void Cos::eval(const std::vector<array>& inputs, array& out) {
 void Cosh::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Cosh());
   } else {
     throw std::invalid_argument(
@@ -350,7 +350,7 @@ void ErfInv::eval(const std::vector<array>& inputs, array& out) {
 void Exp::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Exp());
   } else {
     throw std::invalid_argument(
@@ -359,10 +359,22 @@ void Exp::eval(const std::vector<array>& inputs, array& out) {
   }
 }
 
+void Expm1::eval(const std::vector<array>& inputs, array& out) {
+  assert(inputs.size() == 1);
+  const auto& in = inputs[0];
+  if (issubdtype(out.dtype(), inexact)) {
+    unary_fp(in, out, detail::Expm1());
+  } else {
+    throw std::invalid_argument(
+        "[expm1] Cannot exponentiate elements in array"
+        " with non floating point type.");
+  }
+}
+
 void Floor::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   auto& in = inputs[0];
-  if (not is_integral(in.dtype())) {
+  if (issubdtype(in.dtype(), inexact)) {
     unary_fp(in, out, detail::Floor());
   } else {
     // No-op integer types
@@ -388,7 +400,7 @@ void Full::eval(const std::vector<array>& inputs, array& out) {
 void Log::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     switch (base_) {
       case Base::e:
         unary_fp(in, out, detail::Log());
@@ -410,7 +422,7 @@ void Log::eval(const std::vector<array>& inputs, array& out) {
 void Log1p::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Log1p());
   } else {
     throw std::invalid_argument(
@@ -597,7 +609,7 @@ void Reshape::eval(const std::vector<array>& inputs, array& out) {
 void Round::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   auto& in = inputs[0];
-  if (not is_integral(in.dtype())) {
+  if (issubdtype(in.dtype(), inexact)) {
     unary_fp(in, out, detail::Round());
   } else {
     // No-op integer types
@@ -608,7 +620,7 @@ void Round::eval(const std::vector<array>& inputs, array& out) {
 void Sigmoid::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Sigmoid());
   } else {
     throw std::invalid_argument(
@@ -630,7 +642,7 @@ void Sign::eval(const std::vector<array>& inputs, array& out) {
 void Sin::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Sin());
   } else {
     throw std::invalid_argument(
@@ -642,7 +654,7 @@ void Sin::eval(const std::vector<array>& inputs, array& out) {
 void Sinh::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Sinh());
   } else {
     throw std::invalid_argument(
@@ -850,7 +862,7 @@ void StopGradient::eval(const std::vector<array>& inputs, array& out) {
 void Tan::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Tan());
   } else {
     throw std::invalid_argument(
@@ -862,7 +874,7 @@ void Tan::eval(const std::vector<array>& inputs, array& out) {
 void Tanh::eval(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (is_floating_point(out.dtype())) {
+  if (issubdtype(out.dtype(), inexact)) {
     unary_fp(in, out, detail::Tanh());
   } else {
     throw std::invalid_argument(

mlx/backend/common/reduce.h

@@ -6,8 +6,6 @@
 
 namespace mlx::core {
 
-namespace {
-
 enum ReductionOpType {
   // Self-explanatory. Read everything and produce 1 output.
   ContiguousAllReduce,
@@ -38,6 +36,21 @@ enum ReductionOpType {
   GeneralReduce
 };
 
+struct ReductionPlan {
+  ReductionOpType type;
+  std::vector<int> shape;
+  std::vector<size_t> strides;
+
+  ReductionPlan(
+      ReductionOpType type_,
+      std::vector<int> shape_,
+      std::vector<size_t> strides_)
+      : type(type_), shape(std::move(shape_)), strides(std::move(strides_)) {}
+  ReductionPlan(ReductionOpType type_) : type(type_) {}
+};
+
+namespace {
+
 // Helper for the ndimensional strided loop
 // Should this be in utils?
 inline void nd_loop(
@@ -110,19 +123,6 @@ struct DefaultContiguousReduce {
   }
 };
 
-struct ReductionPlan {
-  ReductionOpType type;
-  std::vector<int> shape;
-  std::vector<size_t> strides;
-
-  ReductionPlan(
-      ReductionOpType type_,
-      std::vector<int> shape_,
-      std::vector<size_t> strides_)
-      : type(type_), shape(std::move(shape_)), strides(std::move(strides_)) {}
-  ReductionPlan(ReductionOpType type_) : type(type_) {}
-};
-
 ReductionPlan get_reduction_plan(const array& x, const std::vector<int> axes) {
   // The data is all there and we are reducing over everything
   if (x.size() == x.data_size() && axes.size() == x.ndim() &&

mlx/backend/common/scan.cpp

@@ -222,7 +222,7 @@ void scan_dispatch(
     }
     case Scan::Min: {
       auto op = [](U* o, const U* y, const T* x) { *o = (*x < *y) ? *x : *y; };
-      auto init = (is_floating_point(input.dtype()))
+      auto init = (issubdtype(input.dtype(), floating))
           ? static_cast<U>(std::numeric_limits<float>::infinity())
           : std::numeric_limits<U>::max();
       auto opcs = DefaultContiguousScan<T, U, decltype(op)>(op, init);
@@ -232,7 +232,7 @@ void scan_dispatch(
     }
     case Scan::Max: {
       auto op = [](U* o, const U* y, const T* x) { *o = (*x < *y) ? *y : *x; };
-      auto init = (is_floating_point(input.dtype()))
+      auto init = (issubdtype(input.dtype(), floating))
           ? static_cast<U>(-std::numeric_limits<float>::infinity())
           : std::numeric_limits<U>::max();
       auto opcs = DefaultContiguousScan<T, U, decltype(op)>(op, init);

mlx/backend/common/softmax.cpp

@@ -1,4 +1,4 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #include <cassert>
 #include <cmath>
@@ -10,7 +10,7 @@ namespace mlx::core {
 
 namespace {
 
-template <typename T>
+template <typename T, typename AccT>
 void softmax(const array& in, array& out) {
   const T* in_ptr = in.data<T>();
   T* out_ptr = out.data<T>();
@@ -22,26 +22,36 @@ void softmax(const array& in, array& out) {
   for (int i = 0; i < M; i++, in_ptr += N, out_ptr += N) {
     // Find the maximum
     current_in_ptr = in_ptr;
-    T maximum = *current_in_ptr;
+    AccT maximum = *current_in_ptr;
     for (int j = 0; j < N; j++, current_in_ptr++) {
-      maximum = (maximum < *current_in_ptr) ? *current_in_ptr : maximum;
+      maximum = (maximum < *current_in_ptr) ? static_cast<AccT>(*current_in_ptr)
+                                            : maximum;
     }
 
     // Compute the normalizer and the exponentials
-    T normalizer = 0;
+    AccT normalizer = 0;
     current_out_ptr = out_ptr;
     current_in_ptr = in_ptr;
     for (int j = 0; j < N; j++, current_out_ptr++, current_in_ptr++) {
-      T expv = std::exp(*current_in_ptr - maximum);
+      AccT expv = std::exp(*current_in_ptr - maximum);
       normalizer += expv;
-      *current_out_ptr = expv;
+      if constexpr (std::is_same<T, AccT>::value) {
+        *current_out_ptr = expv;
+      }
     }
     normalizer = 1 / normalizer;
 
     // Normalize
+    current_in_ptr = in_ptr;
     current_out_ptr = out_ptr;
     for (int j = 0; j < N; j++, current_out_ptr++) {
-      *current_out_ptr *= normalizer;
+      if constexpr (std::is_same<T, AccT>::value) {
+        *current_out_ptr *= normalizer;
+      } else {
+        auto v = std::exp(*current_in_ptr - maximum);
+        *current_out_ptr = static_cast<T>(v * normalizer);
+        current_in_ptr++;
+      }
     }
   }
 }
@@ -91,13 +101,21 @@ void Softmax::eval(const std::vector<array>& inputs, array& out) {
           "Softmax is defined only for floating point types");
       break;
     case float32:
-      softmax<float>(in, out);
+      softmax<float, float>(in, out);
       break;
     case float16:
-      softmax<float16_t>(in, out);
+      if (precise_) {
+        softmax<float16_t, float>(in, out);
+      } else {
+        softmax<float16_t, float16_t>(in, out);
+      }
       break;
     case bfloat16:
-      softmax<bfloat16_t>(in, out);
+      if (precise_) {
+        softmax<bfloat16_t, float>(in, out);
+      } else {
+        softmax<bfloat16_t, bfloat16_t>(in, out);
+      }
       break;
     case complex64:
       throw std::invalid_argument(

mlx/backend/common/utils.h

@@ -89,9 +89,8 @@ collapse_contiguous_dims(const std::vector<array>& xs) {
   return collapse_contiguous_dims(xs[0].shape(), strides);
 }
 
-template <typename... Arrays>
-inline std::tuple<std::vector<int>, std::vector<std::vector<size_t>>>
-collapse_contiguous_dims(Arrays... xs) {
+template <typename... Arrays, typename = enable_for_arrays_t<Arrays...>>
+inline auto collapse_contiguous_dims(Arrays&&... xs) {
   return collapse_contiguous_dims(
       std::vector<array>{std::forward<Arrays>(xs)...});
 }

mlx/backend/metal/allocator.cpp

@@ -1,6 +1,7 @@
 // Copyright © 2023-2024 Apple Inc.
 #include "mlx/backend/metal/allocator.h"
 #include "mlx/backend/metal/metal.h"
+#include "mlx/backend/metal/metal_impl.h"
 
 #include <mach/vm_page_size.h>
 #include <unistd.h>

mlx/backend/metal/compiled.cpp

@@ -229,14 +229,7 @@ void Compiled::eval_gpu(
 
   // Figure out which kernel we are using
   auto& output_shape = outputs[0].shape();
-  bool contiguous = true;
-  for (auto& x : inputs) {
-    if ((!x.flags().row_contiguous || x.shape() != output_shape) &&
-        !is_scalar(x)) {
-      contiguous = false;
-      break;
-    }
-  }
+  bool contiguous = compiled_check_contiguity(inputs, output_shape);
 
   // Collapse contiguous dims to route to a faster kernel if possible. Also
   // handle all broadcasting.
@@ -296,7 +289,7 @@ void Compiled::eval_gpu(
     }
   }
   auto kernel = d.get_kernel(kernel_name, lib);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
 
   // Put the inputs in
@@ -307,7 +300,7 @@ void Compiled::eval_gpu(
       continue;
     }
     auto& x = inputs[i];
-    set_array_buffer(compute_encoder, x, cnt++);
+    compute_encoder.set_input_array(x, cnt++);
     if (!contiguous && !is_scalar(x)) {
       compute_encoder->setBytes(
           strides[stride_idx].data(),
@@ -317,32 +310,12 @@ void Compiled::eval_gpu(
     }
   }
 
-  // Allocate space for the outputs possibly with input donation
-  {
-    int o = 0;
-    for (int i = 0; i < inputs.size() && o < outputs.size(); ++i) {
-      auto& in = inputs[i];
-      // Conditions for donation
-      // - Row contiguous
-      // - Donatable
-      // - Correct size
-      // - Not a constant
-      if (in.flags().row_contiguous && in.nbytes() == outputs[o].nbytes() &&
-          in.is_donatable() &&
-          constant_ids_.find(inputs_[i].id()) == constant_ids_.end()) {
-        outputs[o].move_shared_buffer(
-            in, outputs[o].strides(), in.flags(), in.data_size());
-        o++;
-      }
-    }
-    for (; o < outputs.size(); ++o) {
-      outputs[o].set_data(allocator::malloc_or_wait(outputs[o].nbytes()));
-    }
-  }
+  compiled_allocate_outputs(
+      inputs, outputs, inputs_, constant_ids_, contiguous, true);
 
   // Put the outputs in
   for (auto& x : outputs) {
-    set_array_buffer(compute_encoder, x, cnt++);
+    compute_encoder.set_output_array(x, cnt++);
   }
 
   // Put the output shape and strides in

mlx/backend/metal/conv.cpp

@@ -41,12 +41,12 @@ void explicit_gemm_conv_ND_gpu(
   // Prepare unfolding kernel
   std::ostringstream kname;
   kname << "naive_unfold_nd_" << type_to_name(in_unfolded) << "_" << N;
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
-  set_array_buffer(compute_encoder, in, 0);
-  set_array_buffer(compute_encoder, in_unfolded, 1);
+  compute_encoder.set_input_array(in, 0);
+  compute_encoder.set_output_array(in_unfolded, 1);
 
   compute_encoder->setBytes(&conv_params, sizeof(conv_params), 2);
 
@@ -140,7 +140,7 @@ void slow_conv_2D_gpu(
         << "_tm" << tm << "_tn" << tn;
 
   // Encode and dispatch kernel
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -153,9 +153,9 @@ void slow_conv_2D_gpu(
   MTL::Size group_dims = MTL::Size(bm, bn, 1);
   MTL::Size grid_dims = MTL::Size(grid_dim_x, grid_dim_y, grid_dim_z);
 
-  set_array_buffer(compute_encoder, in, 0);
-  set_array_buffer(compute_encoder, wt, 1);
-  set_array_buffer(compute_encoder, out, 2);
+  compute_encoder.set_input_array(in, 0);
+  compute_encoder.set_input_array(wt, 1);
+  compute_encoder.set_output_array(out, 2);
 
   compute_encoder->setBytes(&conv_params, sizeof(MLXConvParams<2>), 3);
   compute_encoder->dispatchThreadgroups(grid_dims, group_dims);
@@ -241,7 +241,7 @@ void implicit_gemm_conv_2D_gpu(
         << "_filter_" << (small_filter ? 's' : 'l');
 
   // Encode and dispatch kernel
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -254,9 +254,9 @@ void implicit_gemm_conv_2D_gpu(
   MTL::Size grid_dims = MTL::Size(grid_dim_x, grid_dim_y, 1);
 
   // Encode arrays
-  set_array_buffer(compute_encoder, in, 0);
-  set_array_buffer(compute_encoder, wt, 1);
-  set_array_buffer(compute_encoder, out, 2);
+  compute_encoder.set_input_array(in, 0);
+  compute_encoder.set_input_array(wt, 1);
+  compute_encoder.set_output_array(out, 2);
 
   // Encode params
   compute_encoder->setBytes(&conv_params, sizeof(MLXConvParams<2>), 3);
@@ -394,7 +394,7 @@ void implicit_gemm_conv_2D_general_gpu(
         << "_bn" << bn << "_bk" << bk << "_wm" << wm << "_wn" << wn;
 
   // Encode and dispatch kernel
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -408,9 +408,9 @@ void implicit_gemm_conv_2D_general_gpu(
   MTL::Size grid_dims = MTL::Size(grid_dim_x, grid_dim_y, grid_dim_z);
 
   // Encode arrays
-  set_array_buffer(compute_encoder, in, 0);
-  set_array_buffer(compute_encoder, wt, 1);
-  set_array_buffer(compute_encoder, out, 2);
+  compute_encoder.set_input_array(in, 0);
+  compute_encoder.set_input_array(wt, 1);
+  compute_encoder.set_output_array(out, 2);
 
   // Encode params
   compute_encoder->setBytes(&conv_params, sizeof(MLXConvParams<2>), 3);
@@ -511,12 +511,12 @@ void winograd_conv_2D_gpu(
     std::ostringstream kname;
     kname << "winograd_conv_2d_weight_transform_" << type_to_name(out) << "_bc"
           << bc;
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
-    set_array_buffer(compute_encoder, wt, 0);
-    set_array_buffer(compute_encoder, filt_wg, 1);
+    compute_encoder.set_input_array(wt, 0);
+    compute_encoder.set_output_array(filt_wg, 1);
 
     compute_encoder->setBytes(&C_c, sizeof(int), 2);
     compute_encoder->setBytes(&O_c, sizeof(int), 3);
@@ -539,12 +539,12 @@ void winograd_conv_2D_gpu(
     std::ostringstream kname;
     kname << "winograd_conv_2d_input_transform_" << type_to_name(out) << "_bc"
           << bc;
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
-    set_array_buffer(compute_encoder, in_padded, 0);
-    set_array_buffer(compute_encoder, inp_wg, 1);
+    compute_encoder.set_input_array(in_padded, 0);
+    compute_encoder.set_output_array(inp_wg, 1);
 
     compute_encoder->setBytes(
         &conv_params_updated, sizeof(MLXConvParams<2>), 2);
@@ -587,12 +587,12 @@ void winograd_conv_2D_gpu(
     std::ostringstream kname;
     kname << "winograd_conv_2d_output_transform_" << type_to_name(out) << "_bo"
           << bc;
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
-    set_array_buffer(compute_encoder, out_wg, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(out_wg, 0);
+    compute_encoder.set_output_array(out, 1);
 
     compute_encoder->setBytes(
         &conv_params_updated, sizeof(MLXConvParams<2>), 2);

mlx/backend/metal/copy.cpp

@@ -83,15 +83,15 @@ void copy_gpu_inplace(
     kname << "_" << shape.size();
   }
   auto kernel = d.get_kernel(kname.str());
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
   bool donate_in = in.data_shared_ptr() == nullptr;
 
   inp_offset *= size_of(in.dtype());
   out_offset *= size_of(out.dtype());
 
-  set_array_buffer(compute_encoder, donate_in ? out : in, inp_offset, 0);
-  set_array_buffer(compute_encoder, out, out_offset, 1);
+  compute_encoder.set_input_array(donate_in ? out : in, 0, inp_offset);
+  compute_encoder.set_output_array(out, 1, out_offset);
 
   if (ctype == CopyType::General || ctype == CopyType::GeneralGeneral) {
     int ndim = shape.size();

mlx/backend/metal/device.cpp

@@ -1,4 +1,4 @@
-// Copyright © 2023-24 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #include <dlfcn.h>
 #include <cstdlib>
@@ -11,7 +11,9 @@
 
 #include "mlx/backend/metal/device.h"
 #include "mlx/backend/metal/metal.h"
+#include "mlx/backend/metal/metal_impl.h"
 #include "mlx/backend/metal/mps/gemm.h"
+#include "mlx/backend/metal/utils.h"
 
 namespace fs = std::filesystem;
 
@@ -145,6 +147,7 @@ void Device::new_queue(int index) {
   // We lock this as a critical section for safety
   const std::lock_guard<std::mutex> lock(mtx_);
   auto q = device_->newCommandQueue(MAX_BUFFERS_PER_QUEUE);
+  debug_set_stream_queue_label(q, index);
   if (!q) {
     throw std::runtime_error(
         "[metal::Device] Failed to make new command queue.");
@@ -203,14 +206,15 @@ void Device::end_encoding(int index) {
   }
 }
 
-MTL::ComputeCommandEncoder* Device::get_command_encoder(int index) {
+CommandEncoder& Device::get_command_encoder(int index) {
   auto eit = encoder_map_.find(index);
   if (eit == encoder_map_.end()) {
     auto cb = get_command_buffer(index);
-    auto compute_encoder = cb->computeCommandEncoder();
+    auto compute_encoder =
+        cb->computeCommandEncoder(MTL::DispatchTypeConcurrent);
     // Increment ref count so the buffer is not garbage collected
     compute_encoder->retain();
-    eit = encoder_map_.insert({index, compute_encoder}).first;
+    eit = encoder_map_.emplace(index, CommandEncoder{compute_encoder}).first;
   }
   return eit->second;
 }

mlx/backend/metal/device.h

@@ -1,4 +1,4 @@
-// Copyright © 2023-24 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #pragma once
 
@@ -7,10 +7,12 @@
 #include <mutex>
 #include <string>
 #include <unordered_map>
+#include <unordered_set>
 
 #include <dlfcn.h>
 #include <filesystem>
 
+#include "mlx/array.h"
 #include "mlx/device.h"
 
 namespace fs = std::filesystem;
@@ -34,6 +36,69 @@ inline std::string get_colocated_mtllib_path(const std::string& lib_name) {
 using MTLFCList =
     std::vector<std::tuple<const void*, MTL::DataType, NS::UInteger>>;
 
+struct CommandEncoder {
+  CommandEncoder(MTL::ComputeCommandEncoder* enc)
+      : enc(enc), concurrent(false){};
+  CommandEncoder& operator=(const CommandEncoder&) = delete;
+
+  struct ConcurrentContext {
+    ConcurrentContext(CommandEncoder& enc) : enc(enc) {
+      enc.concurrent = true;
+    }
+    ~ConcurrentContext() {
+      enc.concurrent = false;
+      enc.outputs.insert(
+          enc.concurrent_outputs.begin(), enc.concurrent_outputs.end());
+      enc.concurrent_outputs.clear();
+    }
+
+   private:
+    CommandEncoder& enc;
+  };
+
+  MTL::ComputeCommandEncoder* operator->() {
+    return enc;
+  }
+
+  void set_input_array(const array& a, int idx, int offset = 0) {
+    auto r_buf =
+        static_cast<MTL::Resource*>(const_cast<void*>(a.buffer().ptr()));
+    if (auto it = outputs.find(r_buf); it != outputs.end()) {
+      // Insert a barrier
+      enc->memoryBarrier(&r_buf, 1);
+
+      // Remove the output
+      outputs.erase(it);
+    }
+    auto a_buf = static_cast<const MTL::Buffer*>(a.buffer().ptr());
+    auto base_offset = a.data<char>() -
+        static_cast<char*>(const_cast<MTL::Buffer*>(a_buf)->contents());
+    base_offset += offset;
+    enc->setBuffer(a_buf, base_offset, idx);
+  }
+
+  void set_output_array(array& a, int idx, int offset = 0) {
+    // Add barriers before adding the output to the output set
+    set_input_array(a, idx, offset);
+    auto buf = static_cast<MTL::Resource*>(a.buffer().ptr());
+    if (concurrent) {
+      concurrent_outputs.insert(buf);
+    } else {
+      outputs.insert(buf);
+    }
+  }
+
+  ConcurrentContext start_concurrent() {
+    return ConcurrentContext(*this);
+  }
+
+ private:
+  MTL::ComputeCommandEncoder* enc;
+  bool concurrent;
+  std::unordered_set<MTL::Resource*> outputs;
+  std::unordered_set<MTL::Resource*> concurrent_outputs;
+};
+
 class Device {
  public:
   Device();
@@ -51,7 +116,7 @@ class Device {
   int get_command_buffer_ops(int index);
   void increment_command_buffer_ops(int index);
   void commit_command_buffer(int index);
-  MTL::ComputeCommandEncoder* get_command_encoder(int index);
+  CommandEncoder& get_command_encoder(int index);
   void end_encoding(int index);
 
   void register_library(
@@ -132,7 +197,7 @@ class Device {
   MTL::Device* device_;
   std::unordered_map<int32_t, MTL::CommandQueue*> queue_map_;
   std::unordered_map<int32_t, std::pair<int, MTL::CommandBuffer*>> buffer_map_;
-  std::unordered_map<int32_t, MTL::ComputeCommandEncoder*> encoder_map_;
+  std::unordered_map<int32_t, CommandEncoder> encoder_map_;
   std::unordered_map<std::string, MTL::ComputePipelineState*> kernel_map_;
   std::unordered_map<std::string, MTL::Library*> library_map_;
   std::mutex mtx_;

mlx/backend/metal/indexing.cpp

@@ -49,7 +49,7 @@ void Gather::eval_gpu(const std::vector<array>& inputs, array& out) {
     kname << "_" << idx_ndim;
   }
 
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -81,8 +81,8 @@ void Gather::eval_gpu(const std::vector<array>& inputs, array& out) {
   }
 
   // Set all the buffers
-  set_array_buffer(compute_encoder, src, 0);
-  set_array_buffer(compute_encoder, out, 1);
+  compute_encoder.set_input_array(src, 0);
+  compute_encoder.set_output_array(out, 1);
 
   // Set source info
   compute_encoder->setBytes(src.shape().data(), ndim * sizeof(int), 2);
@@ -103,7 +103,7 @@ void Gather::eval_gpu(const std::vector<array>& inputs, array& out) {
 
   // Set index buffers
   for (int i = 1; i < nidx + 1; ++i) {
-    set_array_buffer(compute_encoder, inputs[i], 20 + i);
+    compute_encoder.set_input_array(inputs[i], 20 + i);
   }
 
   // Launch grid
@@ -183,7 +183,7 @@ void Scatter::eval_gpu(const std::vector<array>& inputs, array& out) {
   }
   kname << "_" << nidx;
 
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
 
   auto& upd = inputs.back();
@@ -192,8 +192,8 @@ void Scatter::eval_gpu(const std::vector<array>& inputs, array& out) {
   compute_encoder->setComputePipelineState(kernel);
 
   // Set all the buffers
-  set_array_buffer(compute_encoder, upd, 1);
-  set_array_buffer(compute_encoder, out, 2);
+  compute_encoder.set_input_array(upd, 1);
+  compute_encoder.set_output_array(out, 2);
 
   // Set update info
   uint upd_ndim = upd.ndim();
@@ -210,7 +210,7 @@ void Scatter::eval_gpu(const std::vector<array>& inputs, array& out) {
 
     // Set index buffers
     for (int i = 1; i < nidx + 1; ++i) {
-      set_array_buffer(compute_encoder, inputs[i], 20 + i);
+      compute_encoder.set_input_array(inputs[i], 20 + i);
     }
 
     // Launch grid
@@ -280,7 +280,7 @@ void Scatter::eval_gpu(const std::vector<array>& inputs, array& out) {
 
     // Set index buffers
     for (int i = 1; i < nidx + 1; ++i) {
-      set_array_buffer(compute_encoder, inputs[i], 20 + i);
+      compute_encoder.set_input_array(inputs[i], 20 + i);
     }
 
     // Launch grid

mlx/backend/metal/kernels/CMakeLists.txt

@@ -7,6 +7,7 @@ set(
   ${CMAKE_CURRENT_SOURCE_DIR}/complex.h
   ${CMAKE_CURRENT_SOURCE_DIR}/defines.h
   ${CMAKE_CURRENT_SOURCE_DIR}/erf.h
+  ${CMAKE_CURRENT_SOURCE_DIR}/expm1f.h
   ${CMAKE_CURRENT_SOURCE_DIR}/indexing.h
   ${CMAKE_CURRENT_SOURCE_DIR}/unary.h
   ${CMAKE_CURRENT_SOURCE_DIR}/utils.h
@@ -37,11 +38,17 @@ set(
 )
 
 function(build_kernel_base TARGET SRCFILE DEPS)
+  set(METAL_FLAGS -Wall -Wextra -fno-fast-math)
+  if(MLX_METAL_DEBUG)
+    set(METAL_FLAGS ${METAL_FLAGS}
+        -gline-tables-only
+        -frecord-sources)
+  endif()
   add_custom_command(
-    COMMAND xcrun -sdk macosx metal -Wall -Wextra
-                  -fno-fast-math
-                  -c ${SRCFILE} 
-                  -I${PROJECT_SOURCE_DIR} 
+    COMMAND xcrun -sdk macosx metal
+                  ${METAL_FLAGS}
+                  -c ${SRCFILE}
+                  -I${PROJECT_SOURCE_DIR}
                   -o ${TARGET}.air
     DEPENDS ${SRCFILE} ${DEPS}
     OUTPUT ${TARGET}.air

mlx/backend/metal/kernels/expm1f.h

@@ -0,0 +1,89 @@
+// Copyright © 2023 Apple Inc.
+
+#pragma once
+
+#include <metal_math>
+
+// Original license copied below:
+//  Copyright (c) 2015-2023 Norbert Juffa
+//  All rights reserved.
+//
+//  Redistribution and use in source and binary forms, with or without
+//  modification, are permitted provided that the following conditions
+//  are met:
+//
+//  1. Redistributions of source code must retain the above copyright
+//     notice, this list of conditions and the following disclaimer.
+//
+//  2. Redistributions in binary form must reproduce the above copyright
+//     notice, this list of conditions and the following disclaimer in the
+//     documentation and/or other materials provided with the distribution.
+//
+//  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+//  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+//  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+//  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+//  HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+//  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+//  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+//  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+//  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+//  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+//  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+/* Compute exponential base e minus 1. Maximum ulp error = 0.997458
+
+   i = rint(a/log(2)), f = a-i*log(2). Then expm1(a) = 2**i * (expm1(f)+1) - 1.
+   Compute r = expm1(f). Then expm1(a)= 2 * (0.5 * 2**i * r + 0.5 * 2**i - 0.5).
+   With t = 0.5*2**i, expm1(a) = 2*(r * t + t-0.5). However, for best accuracy,
+   when i == 1, expm1(a)= 2*(r + 0.5), and when i == 0, expm1(a) = r.
+
+   NOTE: Scale factor b is only applied if i < 0 or i > 1 (should be power of 2)
+*/
+float expm1f_scaled_unchecked(float a, float b) {
+  float f, j, r, s, t, u, v, x, y;
+  int i;
+
+  // exp(a) = 2**i * exp(f); i = rintf (a / log(2))
+  j = fma(1.442695f, a, 12582912.f); // 0x1.715476p0, 0x1.8p23
+  j = j - 12582912.0f; // 0x1.8p23
+  i = (int)j;
+  f = fma(j, -6.93145752e-1f, a);
+
+  // approximate r = exp(f)-1 on interval [-log(2)/2, +log(2)/2]
+  s = f * f;
+  if (a == 0.0f)
+    s = a; // ensure -0 is passed through
+  // err = 0.997458  ulp1 = 11081805
+  r = 1.97350979e-4f; // 0x1.9de000p-13
+  r = fma(r, f, 1.39309070e-3f); // 0x1.6d30bcp-10
+  r = fma(r, f, 8.33343994e-3f); // 0x1.1111f6p-7
+  r = fma(r, f, 4.16668020e-2f); // 0x1.55559ep-5
+  r = fma(r, f, 1.66666716e-1f); // 0x1.55555cp-3
+  r = fma(r, f, 4.99999970e-1f); // 0x1.fffffep-2
+  u = (j == 1) ? (f + 0.5f) : f;
+  v = fma(r, s, u);
+  s = 0.5f * b;
+  t = ldexp(s, i);
+  y = t - s;
+  x = (t - y) - s; // double-float canonicalization of difference
+  r = fma(v, t, x) + y;
+  r = r + r;
+  if (j == 0)
+    r = v;
+  if (j == 1)
+    r = v + v;
+  return r;
+}
+
+/* Compute exponential base e minus 1. max ulp err = 0.99746 */
+float expm1f(float a) {
+  float r;
+
+  r = expm1f_scaled_unchecked(a, 1.0f);
+  /* handle severe overflow and underflow */
+  if (abs(a - 1.0f) > 88.0f) {
+    r = fma(r, r, -1.0f);
+  }
+  return r;
+}

mlx/backend/metal/kernels/layer_norm.metal

@@ -205,39 +205,341 @@ template <typename T, int N_READS = RMS_N_READS>
   }
 }
 
+template <typename T, int N_READS = RMS_N_READS>
+[[kernel]] void vjp_layer_norm_single_row(
+    const device T* x,
+    const device T* w,
+    const device T* g,
+    device T* gx,
+    device T* gw,
+    constant float& eps,
+    constant uint& axis_size,
+    constant uint& w_stride,
+    uint gid [[threadgroup_position_in_grid]],
+    uint lid [[thread_position_in_threadgroup]],
+    uint simd_lane_id [[thread_index_in_simdgroup]],
+    uint simd_group_id [[simdgroup_index_in_threadgroup]]) {
+  // Advance the input pointers
+  x += gid * axis_size + lid * N_READS;
+  g += gid * axis_size + lid * N_READS;
+  w += w_stride * lid * N_READS;
+
+  // Allocate registers for the computation and accumulators
+  float thread_x[N_READS];
+  float thread_w[N_READS];
+  float thread_g[N_READS];
+  float sumx = 0;
+  float sumx2 = 0;
+  float sumwg = 0;
+  float sumwgx = 0;
+
+  constexpr int SIMD_SIZE = 32;
+
+  threadgroup float local_sumx[SIMD_SIZE];
+  threadgroup float local_sumx2[SIMD_SIZE];
+  threadgroup float local_sumwg[SIMD_SIZE];
+  threadgroup float local_sumwgx[SIMD_SIZE];
+  threadgroup float local_mean[1];
+  threadgroup float local_normalizer[1];
+  threadgroup float local_meanwg[1];
+  threadgroup float local_meanwgx[1];
+
+  if (lid * N_READS + N_READS <= axis_size) {
+    for (int i = 0; i < N_READS; i++) {
+      thread_x[i] = x[i];
+      thread_w[i] = w[i * w_stride];
+      thread_g[i] = g[i];
+      float wg = thread_w[i] * thread_g[i];
+      sumx += thread_x[i];
+      sumx2 += thread_x[i] * thread_x[i];
+      sumwg += wg;
+      sumwgx += wg * thread_x[i];
+    }
+  } else {
+    for (int i = 0; i < N_READS; i++) {
+      if ((lid * N_READS + i) < axis_size) {
+        thread_x[i] = x[i];
+        thread_w[i] = w[i * w_stride];
+        thread_g[i] = g[i];
+        float wg = thread_w[i] * thread_g[i];
+        sumx += thread_x[i];
+        sumx2 += thread_x[i] * thread_x[i];
+        sumwg += wg;
+        sumwgx += wg * thread_x[i];
+      }
+    }
+  }
+
+  sumx = simd_sum(sumx);
+  sumx2 = simd_sum(sumx2);
+  sumwg = simd_sum(sumwg);
+  sumwgx = simd_sum(sumwgx);
+
+  //  Initialize shared memory
+  if (simd_group_id == 0) {
+    local_sumx[simd_lane_id] = 0;
+    local_sumx2[simd_lane_id] = 0;
+    local_sumwg[simd_lane_id] = 0;
+    local_sumwgx[simd_lane_id] = 0;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+
+  // Write simd accumulations into shared memory
+  if (simd_lane_id == 0) {
+    local_sumx[simd_group_id] = sumx;
+    local_sumx2[simd_group_id] = sumx2;
+    local_sumwg[simd_group_id] = sumwg;
+    local_sumwgx[simd_group_id] = sumwgx;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+
+  // Accumulate over simd groups
+  if (simd_group_id == 0) {
+    sumx = simd_sum(local_sumx[simd_lane_id]);
+    sumx2 = simd_sum(local_sumx2[simd_lane_id]);
+    sumwg = simd_sum(local_sumwg[simd_lane_id]);
+    sumwgx = simd_sum(local_sumwgx[simd_lane_id]);
+    if (simd_lane_id == 0) {
+      float mean = sumx / axis_size;
+      float variance = sumx2 / axis_size - mean * mean;
+
+      local_mean[0] = mean;
+      local_normalizer[0] = metal::precise::rsqrt(variance + eps);
+      local_meanwg[0] = sumwg / axis_size;
+      local_meanwgx[0] = sumwgx / axis_size;
+    }
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+
+  float mean = local_mean[0];
+  float normalizer = local_normalizer[0];
+  float meanwg = local_meanwg[0];
+  float meanwgxc = local_meanwgx[0] - meanwg * mean;
+  float normalizer2 = normalizer * normalizer;
+
+  // Write the outputs
+  gx += gid * axis_size + lid * N_READS;
+  gw += gid * axis_size + lid * N_READS;
+  if (lid * N_READS + N_READS <= axis_size) {
+    for (int i = 0; i < N_READS; i++) {
+      thread_x[i] = (thread_x[i] - mean) * normalizer;
+      gx[i] = static_cast<T>(normalizer * (thread_w[i] * thread_g[i] - meanwg) -
+                             thread_x[i] * meanwgxc * normalizer2);
+      gw[i] = static_cast<T>(thread_g[i] * thread_x[i]);
+    }
+  } else {
+    for (int i = 0; i < N_READS; i++) {
+      if ((lid * N_READS + i) < axis_size) {
+        thread_x[i] = (thread_x[i] - mean) * normalizer;
+        gx[i] = static_cast<T>(normalizer * (thread_w[i] * thread_g[i] - meanwg) -
+                               thread_x[i] * meanwgxc * normalizer2);
+        gw[i] = static_cast<T>(thread_g[i] * thread_x[i]);
+      }
+    }
+  }
+}
+
+template <typename T, int N_READS = RMS_N_READS>
+[[kernel]] void vjp_layer_norm_looped(
+    const device T* x,
+    const device T* w,
+    const device T* g,
+    device T* gx,
+    device T* gw,
+    constant float& eps,
+    constant uint& axis_size,
+    constant uint& w_stride,
+    uint gid [[threadgroup_position_in_grid]],
+    uint lid [[thread_position_in_threadgroup]],
+    uint lsize [[threads_per_threadgroup]],
+    uint simd_lane_id [[thread_index_in_simdgroup]],
+    uint simd_group_id [[simdgroup_index_in_threadgroup]]) {
+  // Advance the input pointers
+  x += gid * axis_size + lid * N_READS;
+  g += gid * axis_size + lid * N_READS;
+  w += w_stride * lid * N_READS;
+
+  // Allocate registers for the accumulators
+  float sumx = 0;
+  float sumx2 = 0;
+  float sumwg = 0;
+  float sumwgx = 0;
+
+  constexpr int SIMD_SIZE = 32;
+
+  threadgroup float local_sumx[SIMD_SIZE];
+  threadgroup float local_sumx2[SIMD_SIZE];
+  threadgroup float local_sumwg[SIMD_SIZE];
+  threadgroup float local_sumwgx[SIMD_SIZE];
+  threadgroup float local_mean[1];
+  threadgroup float local_normalizer[1];
+  threadgroup float local_meanwg[1];
+  threadgroup float local_meanwgx[1];
+
+  for (uint r = 0; r < axis_size; r += lsize * N_READS) {
+    if (r + lid * N_READS + N_READS <= axis_size) {
+      for (int i = 0; i < N_READS; i++) {
+        float xi = x[i + r];
+        float wi = w[(i + r) * w_stride];
+        float gi = g[i + r];
+        float wg = wi * gi;
+        sumx += xi;
+        sumx2 += xi * xi;
+        sumwg += wg;
+        sumwgx += wg * xi;
+      }
+    } else {
+      for (int i = 0; i < N_READS; i++) {
+        if ((r + lid * N_READS + i) < axis_size) {
+          float xi = x[i + r];
+          float wi = w[(i + r) * w_stride];
+          float gi = g[i + r];
+          float wg = wi * gi;
+          sumx += xi;
+          sumx2 += xi * xi;
+          sumwg += wg;
+          sumwgx += wg * xi;
+        }
+      }
+    }
+  }
+
+  sumx = simd_sum(sumx);
+  sumx2 = simd_sum(sumx2);
+  sumwg = simd_sum(sumwg);
+  sumwgx = simd_sum(sumwgx);
+
+  //  Initialize shared memory
+  if (simd_group_id == 0) {
+    local_sumx[simd_lane_id] = 0;
+    local_sumx2[simd_lane_id] = 0;
+    local_sumwg[simd_lane_id] = 0;
+    local_sumwgx[simd_lane_id] = 0;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+
+  // Write simd accumulations into shared memory
+  if (simd_lane_id == 0) {
+    local_sumx[simd_group_id] = sumx;
+    local_sumx2[simd_group_id] = sumx2;
+    local_sumwg[simd_group_id] = sumwg;
+    local_sumwgx[simd_group_id] = sumwgx;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+
+  // Accumulate over simd groups
+  if (simd_group_id == 0) {
+    sumx = simd_sum(local_sumx[simd_lane_id]);
+    sumx2 = simd_sum(local_sumx2[simd_lane_id]);
+    sumwg = simd_sum(local_sumwg[simd_lane_id]);
+    sumwgx = simd_sum(local_sumwgx[simd_lane_id]);
+    if (simd_lane_id == 0) {
+      float mean = sumx / axis_size;
+      float variance = sumx2 / axis_size - mean * mean;
+
+      local_mean[0] = mean;
+      local_normalizer[0] = metal::precise::rsqrt(variance + eps);
+      local_meanwg[0] = sumwg / axis_size;
+      local_meanwgx[0] = sumwgx / axis_size;
+    }
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+
+  float mean = local_mean[0];
+  float normalizer = local_normalizer[0];
+  float meanwg = local_meanwg[0];
+  float meanwgxc = local_meanwgx[0] - meanwg * mean;
+  float normalizer2 = normalizer * normalizer;
+
+  // Write the outputs
+  gx += gid * axis_size + lid * N_READS;
+  gw += gid * axis_size + lid * N_READS;
+  for (uint r = 0; r < axis_size; r += lsize * N_READS) {
+    if (r + lid * N_READS + N_READS <= axis_size) {
+      for (int i = 0; i < N_READS; i++) {
+        float xi = (x[i + r] - mean) * normalizer;
+        float wi = w[(i + r) * w_stride];
+        float gi = g[i + r];
+        gx[i + r] = static_cast<T>(normalizer * (wi * gi - meanwg) -
+                                   xi * meanwgxc * normalizer2);
+        gw[i + r] = static_cast<T>(gi * xi);
+      }
+    } else {
+      for (int i = 0; i < N_READS; i++) {
+        if ((r + lid * N_READS + i) < axis_size) {
+          float xi = (x[i + r] - mean) * normalizer;
+          float wi = w[(i + r) * w_stride];
+          float gi = g[i + r];
+          gx[i + r] = static_cast<T>(normalizer * (wi * gi - meanwg) -
+                                     xi * meanwgxc * normalizer2);
+          gw[i + r] = static_cast<T>(gi * xi);
+        }
+      }
+    }
+  }
+}
 
 // clang-format off
-#define instantiate_layer_norm_single_row(name, itype)        \
-  template [[host_name("layer_norm" #name)]] [[kernel]] void  \
-  layer_norm_single_row<itype>(                               \
-      const device itype* x,                                  \
-      const device itype* w,                                  \
-      const device itype* b,                                  \
-      device itype* out,                                      \
-      constant float& eps,                                    \
-      constant uint& axis_size,                               \
-      constant uint& w_stride,                                \
-      constant uint& b_stride,                                \
-      uint gid [[thread_position_in_grid]],                   \
-      uint lid [[thread_position_in_threadgroup]],            \
-      uint simd_lane_id [[thread_index_in_simdgroup]],        \
+#define instantiate_layer_norm_single_row(name, itype)            \
+  template [[host_name("layer_norm" #name)]] [[kernel]] void      \
+  layer_norm_single_row<itype>(                                   \
+      const device itype* x,                                      \
+      const device itype* w,                                      \
+      const device itype* b,                                      \
+      device itype* out,                                          \
+      constant float& eps,                                        \
+      constant uint& axis_size,                                   \
+      constant uint& w_stride,                                    \
+      constant uint& b_stride,                                    \
+      uint gid [[thread_position_in_grid]],                       \
+      uint lid [[thread_position_in_threadgroup]],                \
+      uint simd_lane_id [[thread_index_in_simdgroup]],            \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]);     \
+  template [[host_name("vjp_layer_norm" #name)]] [[kernel]] void  \
+  vjp_layer_norm_single_row<itype>(                               \
+      const device itype* x,                                      \
+      const device itype* w,                                      \
+      const device itype* g,                                      \
+      device itype* gx,                                           \
+      device itype* gw,                                           \
+      constant float& eps,                                        \
+      constant uint& axis_size,                                   \
+      constant uint& w_stride,                                    \
+      uint gid [[thread_position_in_grid]],                       \
+      uint lid [[thread_position_in_threadgroup]],                \
+      uint simd_lane_id [[thread_index_in_simdgroup]],            \
       uint simd_group_id [[simdgroup_index_in_threadgroup]]);
 
-#define instantiate_layer_norm_looped(name, itype)                   \
-  template [[host_name("layer_norm_looped" #name)]] [[kernel]] void  \
-  layer_norm_looped<itype>(                                                 \
-      const device itype* x,                                         \
-      const device itype* w,                                         \
-      const device itype* b,                                         \
-      device itype* out,                                             \
-      constant float& eps,                                           \
-      constant uint& axis_size,                                      \
-      constant uint& w_stride,                                       \
-      constant uint& b_stride,                                       \
-      uint gid [[thread_position_in_grid]],                          \
-      uint lid [[thread_position_in_threadgroup]],                   \
-      uint lsize [[threads_per_threadgroup]],                        \
-      uint simd_lane_id [[thread_index_in_simdgroup]],               \
+#define instantiate_layer_norm_looped(name, itype)                       \
+  template [[host_name("layer_norm_looped" #name)]] [[kernel]] void      \
+  layer_norm_looped<itype>(                                              \
+      const device itype* x,                                             \
+      const device itype* w,                                             \
+      const device itype* b,                                             \
+      device itype* out,                                                 \
+      constant float& eps,                                               \
+      constant uint& axis_size,                                          \
+      constant uint& w_stride,                                           \
+      constant uint& b_stride,                                           \
+      uint gid [[thread_position_in_grid]],                              \
+      uint lid [[thread_position_in_threadgroup]],                       \
+      uint lsize [[threads_per_threadgroup]],                            \
+      uint simd_lane_id [[thread_index_in_simdgroup]],                   \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]);            \
+  template [[host_name("vjp_layer_norm_looped" #name)]] [[kernel]] void  \
+  vjp_layer_norm_looped<itype>(                                          \
+      const device itype* x,                                             \
+      const device itype* w,                                             \
+      const device itype* g,                                             \
+      device itype* gx,                                                  \
+      device itype* gb,                                                  \
+      constant float& eps,                                               \
+      constant uint& axis_size,                                          \
+      constant uint& w_stride,                                           \
+      uint gid [[thread_position_in_grid]],                              \
+      uint lid [[thread_position_in_threadgroup]],                       \
+      uint lsize [[threads_per_threadgroup]],                            \
+      uint simd_lane_id [[thread_index_in_simdgroup]],                   \
       uint simd_group_id [[simdgroup_index_in_threadgroup]]);
 
 #define instantiate_layer_norm(name, itype)      \

mlx/backend/metal/kernels/quantized.metal

@@ -15,14 +15,6 @@ using namespace metal;
 
 MLX_MTL_CONST int SIMD_SIZE = 32;
 
-template <typename T> struct AccT {
-  typedef T acc_t;
-};
-
-template <> struct AccT<bfloat16_t> {
-  typedef float acc_t;
-};
-
 
 template <typename T, typename U, int values_per_thread, int bits>
 inline U load_vector(const device T *x, thread U *x_thread) {
@@ -60,6 +52,51 @@ inline U load_vector(const device T *x, thread U *x_thread) {
   return sum;
 }
 
+template <typename T, typename U, int values_per_thread, int bits>
+inline U load_vector_safe(const device T *x, thread U *x_thread, int N) {
+  static_assert(bits == 2 || bits == 4 || bits == 8, "Template undefined for bits not in {2, 4, 8}");
+
+  U sum = 0;
+
+  if (bits == 2) {
+    for (int i = 0; i < N; i += 4) {
+      sum += x[i] + x[i+1] + x[i+2] + x[i+3];
+      x_thread[i] = x[i];
+      x_thread[i+1] = x[i+1] / 4.0f;
+      x_thread[i+2] = x[i+2] / 16.0f;
+      x_thread[i+3] = x[i+3] / 64.0f;
+    }
+    for (int i=N; i<values_per_thread; i++) {
+      x_thread[i] = 0;
+    }
+  }
+
+  else if (bits == 4) {
+    for (int i = 0; i < N; i += 4) {
+      sum += x[i] + x[i+1] + x[i+2] + x[i+3];
+      x_thread[i] = x[i];
+      x_thread[i+1] = x[i+1] / 16.0f;
+      x_thread[i+2] = x[i+2] / 256.0f;
+      x_thread[i+3] = x[i+3] / 4096.0f;
+    }
+    for (int i=N; i<values_per_thread; i++) {
+      x_thread[i] = 0;
+    }
+  }
+
+  else if (bits == 8) {
+    for (int i = 0; i < N; i++) {
+      sum += x[i];
+      x_thread[i] = x[i];
+    }
+    for (int i=N; i<values_per_thread; i++) {
+      x_thread[i] = 0;
+    }
+  }
+
+  return sum;
+}
+
 template <typename U, int values_per_thread, int bits>
 inline U qdot(const device uint8_t* w, const thread U *x_thread, U scale, U bias, U sum) {
   static_assert(bits == 2 || bits == 4 || bits == 8, "Template undefined for bits not in {2, 4, 8}");
@@ -96,6 +133,74 @@ inline U qdot(const device uint8_t* w, const thread U *x_thread, U scale, U bias
   return scale * accum + sum * bias;
 }
 
+template <typename U, int values_per_thread, int bits>
+inline U qdot_safe(const device uint8_t* w, const thread U *x_thread, U scale, U bias, U sum, int N) {
+  static_assert(bits == 2 || bits == 4 || bits == 8, "Template undefined for bits not in {2, 4, 8}");
+
+  U accum = 0;
+
+  if (bits == 2) {
+    for (int i = 0; i < (N / 4); i++) {
+      accum += (
+          x_thread[4*i] * (w[i] & 0x03)
+          + x_thread[4*i+1] * (w[i] & 0x0c)
+          + x_thread[4*i+2] * (w[i] & 0x30)
+          + x_thread[4*i+3] * (w[i] & 0xc0));
+    }
+  }
+
+  else if (bits == 4) {
+    const device uint16_t* ws = (const device uint16_t*)w;
+    for (int i = 0; i < (N / 4); i++) {
+      accum += (
+          x_thread[4*i] * (ws[i] & 0x000f)
+          + x_thread[4*i+1] * (ws[i] & 0x00f0)
+          + x_thread[4*i+2] * (ws[i] & 0x0f00)
+          + x_thread[4*i+3] * (ws[i] & 0xf000));
+    }
+  }
+
+  else if (bits == 8) {
+    for (int i = 0; i < N; i++) {
+      accum += x_thread[i] * w[i];
+    }
+  }
+
+  return scale * accum + sum * bias;
+}
+
+template <typename U, int values_per_thread, int bits>
+inline void qouter(const thread uint8_t* w, U x, U scale, U bias, thread U* result) {
+  static_assert(bits == 2 || bits == 4 || bits == 8, "Template undefined for bits not in {2, 4, 8}");
+
+  if (bits == 2) {
+    U s[4] = {scale, scale / 4.0f, scale / 16.0f, scale / 64.0f};
+    for (int i = 0; i < (values_per_thread / 4); i++) {
+      result[4*i] += x * (s[0] * (w[i] & 0x03) + bias);
+      result[4*i+1] += x * (s[1] * (w[i] & 0x0c) + bias);
+      result[4*i+2] += x * (s[2] * (w[i] & 0x30) + bias);
+      result[4*i+3] += x * (s[3] * (w[i] & 0xc0) + bias);
+    }
+  }
+
+  else if (bits == 4) {
+    const thread uint16_t* ws = (const thread uint16_t*)w;
+    U s[4] = {scale, scale / 16.0f, scale / 256.0f, scale / 4096.0f};
+    for (int i = 0; i < (values_per_thread / 4); i++) {
+      result[4*i] += x * (s[0] * (ws[i] & 0x000f) + bias);
+      result[4*i+1] += x * (s[1] * (ws[i] & 0x00f0) + bias);
+      result[4*i+2] += x * (s[2] * (ws[i] & 0x0f00) + bias);
+      result[4*i+3] += x * (s[3] * (ws[i] & 0xf000) + bias);
+    }
+  }
+
+  else if (bits == 8) {
+    for (int i = 0; i < values_per_thread; i++) {
+      result[i] += x * (scale * w[i] + bias);
+    }
+  }
+}
+
 template <typename T, int group_size, int bits, int packs_per_thread>
 [[kernel]] void qmv_fast(
     const device uint32_t* w [[buffer(0)]],
@@ -204,7 +309,8 @@ template <typename T, const int group_size, const int bits>
     x += tid.z * in_vec_size + simd_lid * values_per_thread;
     y += tid.z * out_vec_size + out_row;
 
-    for (int k = 0; k < in_vec_size; k += block_size) {
+    int k = 0;
+    for (; k < in_vec_size-block_size; k += block_size) {
       U sum = load_vector<T, U, values_per_thread, bits>(x, x_thread);
 
       for (int row = 0; out_row + row < out_vec_size; row++) {
@@ -222,6 +328,18 @@ template <typename T, const int group_size, const int bits>
       biases += block_size / group_size;
       x += block_size;
     }
+    const int remaining = clamp(static_cast<int>(in_vec_size - k - simd_lid * values_per_thread), 0, values_per_thread);
+    U sum = load_vector_safe<T, U, values_per_thread, bits>(x, x_thread, remaining);
+
+    for (int row = 0; out_row + row < out_vec_size; row++) {
+      const device uint8_t* wl = (const device uint8_t *)(w + row * in_vec_size_w);
+      const device T* sl = scales + row * in_vec_size_g;
+      const device T* bl = biases + row * in_vec_size_g;
+
+      U s = sl[0];
+      U b = bl[0];
+      result[row] += qdot<U, values_per_thread, bits>(wl, x_thread, s, b, sum);
+    }
 
     for (int row = 0; out_row + row < out_vec_size; row++) {
       result[row] = simd_sum(result[row]);
@@ -239,7 +357,8 @@ template <typename T, const int group_size, const int bits>
     x += tid.z * in_vec_size + simd_lid * values_per_thread;
     y += tid.z * out_vec_size + used_out_row;
 
-    for (int k = 0; k < in_vec_size; k += block_size) {
+    int k = 0;
+    for (; k < in_vec_size-block_size; k += block_size) {
       U sum = load_vector<T, U, values_per_thread, bits>(x, x_thread);
 
       for (int row = 0; row < results_per_simdgroup; row++) {
@@ -257,6 +376,18 @@ template <typename T, const int group_size, const int bits>
       biases += block_size / group_size;
       x += block_size;
     }
+    const int remaining = clamp(static_cast<int>(in_vec_size - k - simd_lid * values_per_thread), 0, values_per_thread);
+    U sum = load_vector_safe<T, U, values_per_thread, bits>(x, x_thread, remaining);
+
+    for (int row = 0; row < results_per_simdgroup; row++) {
+      const device uint8_t* wl = (const device uint8_t *)(w + row * in_vec_size_w);
+      const device T* sl = scales + row * in_vec_size_g;
+      const device T* bl = biases + row * in_vec_size_g;
+
+      U s = sl[0];
+      U b = bl[0];
+      result[row] += qdot_safe<U, values_per_thread, bits>(wl, x_thread, s, b, sum, remaining);
+    }
 
     for (int row = 0; row < results_per_simdgroup; row++) {
       result[row] = simd_sum(result[row]);
@@ -268,7 +399,7 @@ template <typename T, const int group_size, const int bits>
 }
 
 
-template <typename T, const int BM, const int BN, const int group_size, const int bits>
+template <typename T, const int group_size, const int bits>
 [[kernel]] void qvm(
     const device T* x [[buffer(0)]],
     const device uint32_t* w [[buffer(1)]],
@@ -278,39 +409,28 @@ template <typename T, const int BM, const int BN, const int group_size, const in
     const constant int& in_vec_size [[buffer(5)]],
     const constant int& out_vec_size [[buffer(6)]],
     uint3 tid [[threadgroup_position_in_grid]],
-    uint lid [[thread_index_in_threadgroup]],
     uint simd_gid [[simdgroup_index_in_threadgroup]],
     uint simd_lid [[thread_index_in_simdgroup]]) {
 
-  static_assert(BM == SIMD_SIZE, "qvm expects BM to be equal to SIMD_SIZE");
-  static_assert(BN == BM, "qvm expects a block size of 32x32");
+  constexpr int num_simdgroups = 8;
+  constexpr int pack_factor = 32 / bits;
+  constexpr int blocksize = SIMD_SIZE;
 
-  (void)lid;
-
-  constexpr int bitmask = (1 << bits) - 1;
-  constexpr int el_per_int = 32 / bits;
-  constexpr int colgroup = BN * el_per_int;
-  constexpr int groups_per_block = colgroup / group_size;
-
-  typedef typename AccT<T>::acc_t U;
-  threadgroup U scales_block[BM * groups_per_block];
-  threadgroup U biases_block[BM * groups_per_block];
-  threadgroup U x_block[BM];
+  typedef float U;
 
   thread uint32_t w_local;
-  thread U result[el_per_int] = {0};
+  thread U result[pack_factor] = {0};
   thread U scale = 1;
   thread U bias = 0;
   thread U x_local = 0;
 
   // Adjust positions
-  const int out_vec_size_w = out_vec_size / el_per_int;
+  const int out_vec_size_w = out_vec_size / pack_factor;
   const int out_vec_size_g = out_vec_size / group_size;
-  int out_col_start = tid.y * (BN * el_per_int);
-  int out_col = out_col_start + simd_gid * el_per_int;
-  w += out_col / el_per_int;
-  scales += out_col_start / group_size;
-  biases += out_col_start / group_size;
+  int out_col = tid.y * (num_simdgroups * pack_factor) + simd_gid * pack_factor;
+  w += out_col / pack_factor;
+  scales += out_col / group_size;
+  biases += out_col / group_size;
   x += tid.z * in_vec_size;
   y += tid.z * out_vec_size + out_col;
 
@@ -318,53 +438,39 @@ template <typename T, const int BM, const int BN, const int group_size, const in
     return;
   }
 
-  // Loop over in_vec in blocks of colgroup
-  for (int i=0; i<in_vec_size; i+=BM) {
-    int offset_lid = simd_lid + i;
-    int offset_gid = simd_gid + i;
-    bool thread_in_bounds = offset_lid < in_vec_size;
-    bool group_in_bounds = offset_gid < in_vec_size;
+  // Loop over in_vec in blocks of blocksize
+  int i = 0;
+  for (; i + blocksize <= in_vec_size; i += blocksize) {
+    x_local = x[i + simd_lid];
+    scale = scales[(i + simd_lid) * out_vec_size_g];
+    bias = biases[(i + simd_lid) * out_vec_size_g];
+    w_local = w[(i + simd_lid) * out_vec_size_w];
 
-    // Load the vec to shared memory
-    threadgroup_barrier(mem_flags::mem_threadgroup);
-    if (simd_gid == 0) {
-      x_block[simd_lid] = (thread_in_bounds) ? x[offset_lid] : 0;
-    }
-
-    // Load the scales and biases to shared memory
-    threadgroup_barrier(mem_flags::mem_threadgroup);
-    if (simd_lid < groups_per_block && group_in_bounds) {
-      scales_block[simd_gid * groups_per_block + simd_lid] = scales[offset_gid * out_vec_size_g + simd_lid];
-      biases_block[simd_gid * groups_per_block + simd_lid] = biases[offset_gid * out_vec_size_g + simd_lid];
-    }
-    threadgroup_barrier(mem_flags::mem_threadgroup);
-
-    // Load in_vec, scale, bias to registers
-    x_local = x_block[simd_lid];
-    scale = scales_block[simd_lid * groups_per_block + (simd_gid * el_per_int) / group_size];
-    bias = biases_block[simd_lid * groups_per_block + (simd_gid * el_per_int) / group_size];
-
-    // Load the matrix elements
-    w_local = (thread_in_bounds) ? w[offset_lid * out_vec_size_w] : 0;
-
-    // Do all the work.
-    #pragma clang loop unroll(full)
-    for (int k=0; k<el_per_int; k++) {
-      result[k] += (scale * static_cast<U>(w_local & bitmask) + bias) * x_local;
-      w_local >>= bits;
-    }
+    qouter<U, pack_factor, bits>((thread uint8_t *)&w_local, x_local, scale, bias, result);
   }
+  if (static_cast<int>(i + simd_lid) < in_vec_size) {
+    x_local = x[i + simd_lid];
+    scale = scales[(i + simd_lid) * out_vec_size_g];
+    bias = biases[(i + simd_lid) * out_vec_size_g];
+    w_local = w[(i + simd_lid) * out_vec_size_w];
+  } else {
+    x_local = 0;
+    scale = 0;
+    bias = 0;
+    w_local = 0;
+  }
+  qouter<U, pack_factor, bits>((thread uint8_t *)&w_local, x_local, scale, bias, result);
 
   // Accumulate in the simdgroup
   #pragma clang loop unroll(full)
-  for (int k=0; k<el_per_int; k++) {
+  for (int k=0; k<pack_factor; k++) {
     result[k] = simd_sum(result[k]);
   }
 
   // Store the result
   if (simd_lid == 0) {
     #pragma clang loop unroll(full)
-    for (int k=0; k<el_per_int; k++) {
+    for (int k=0; k<pack_factor; k++) {
       y[k] = static_cast<T>(result[k]);
     }
   }
@@ -414,6 +520,7 @@ template <typename T, const int BM, const int BK, const int BN, const int group_
   const int K_g = K / group_size;
   const int y_row = tid.y * BM;
   const int y_col = tid.x * BN;
+
   x += y_row * K;
   w += y_col * K_w;
   scales += y_col * K_g;
@@ -466,7 +573,10 @@ template <typename T, const int BM, const int BK, const int BN, const int group_
           const device uint32_t * w_local = w + offset_row * K_w + offset_col;
           threadgroup T * Ws_local = Ws + offset_row * BK + offset_col * el_per_int;
 
-          if (y_row + offset_row < N) {
+          // y_col corresponds to the row of the weight matrix and added to
+          // offset_row it should be less than the total number of rows
+          // otherwise skip.
+          if (y_col + offset_row < N) {
             uint32_t wi = *w_local;
             T scale = scales_block[offset_row * groups_per_block + offset_col / (group_size / el_per_int)];
             T bias = biases_block[offset_row * groups_per_block + offset_col / (group_size / el_per_int)];
@@ -619,7 +729,7 @@ template <typename T, const int BM, const int BK, const int BN, const int group_
           const device uint32_t * w_local = w + offset_row * N_w + offset_col;
           threadgroup T * Ws_local = Ws + offset_row * BN + offset_col * el_per_int;
 
-          if (y_row + offset_row < K) {
+          if (k + offset_row < K) {
             uint32_t wi = *w_local;
             T scale = scales_block[offset_row * groups_per_block + offset_col / (group_size / el_per_int)];
             T bias = biases_block[offset_row * groups_per_block + offset_col / (group_size / el_per_int)];
@@ -738,7 +848,7 @@ instantiate_qmv_types( 32, 8)
 
 #define instantiate_qvm(name, itype, group_size, bits) \
   template [[host_name("qvm_" #name "_gs_" #group_size "_b_" #bits)]] \
-  [[kernel]] void qvm<itype, 32, 32, group_size, bits>( \
+  [[kernel]] void qvm<itype, group_size, bits>( \
     const device itype* x [[buffer(0)]], \
     const device uint32_t* w [[buffer(1)]], \
     const device itype* scales [[buffer(2)]], \
@@ -747,7 +857,6 @@ instantiate_qmv_types( 32, 8)
     const constant int& in_vec_size [[buffer(5)]], \
     const constant int& out_vec_size [[buffer(6)]], \
     uint3 tid [[threadgroup_position_in_grid]], \
-    uint lid [[thread_index_in_threadgroup]], \
     uint simd_gid [[simdgroup_index_in_threadgroup]], \
     uint simd_lid [[thread_index_in_simdgroup]]);
 

mlx/backend/metal/kernels/rms_norm.metal

@@ -150,6 +150,216 @@ template <typename T, int N_READS = RMS_N_READS>
   }
 }
 
+template <typename T, int N_READS = RMS_N_READS>
+[[kernel]] void vjp_rms_single_row(
+    const device T* x,
+    const device T* w,
+    const device T* g,
+    device T* gx,
+    device T* gw,
+    constant float& eps,
+    constant uint& axis_size,
+    constant uint& w_stride,
+    uint gid [[threadgroup_position_in_grid]],
+    uint lid [[thread_position_in_threadgroup]],
+    uint simd_lane_id [[thread_index_in_simdgroup]],
+    uint simd_group_id [[simdgroup_index_in_threadgroup]]) {
+  // Advance the input pointers
+  x += gid * axis_size + lid * N_READS;
+  g += gid * axis_size + lid * N_READS;
+  w += w_stride * lid * N_READS;
+
+  // Allocate registers for the computation and accumulators
+  float thread_x[N_READS];
+  float thread_w[N_READS];
+  float thread_g[N_READS];
+  float sumx2 = 0;
+  float sumgwx = 0;
+
+  // Allocate shared memory to implement the reduction
+  constexpr int SIMD_SIZE = 32;
+  threadgroup float local_sumx2[SIMD_SIZE];
+  threadgroup float local_sumgwx[SIMD_SIZE];
+  threadgroup float local_normalizer[1];
+  threadgroup float local_meangwx[1];
+
+  // Read and accumulate locally
+  if (lid * N_READS + N_READS <= axis_size) {
+    for (int i = 0; i < N_READS; i++) {
+      thread_x[i] = x[i];
+      thread_w[i] = w[w_stride * i];
+      thread_g[i] = g[i];
+
+      sumx2 += thread_x[i] * thread_x[i];
+      sumgwx += thread_x[i] * thread_w[i] * thread_g[i];
+    }
+  } else {
+    for (int i = 0; i < N_READS; i++) {
+      if ((lid * N_READS + i) < axis_size) {
+        thread_x[i] = x[i];
+        thread_w[i] = w[w_stride * i];
+        thread_g[i] = g[i];
+
+        sumx2 += thread_x[i] * thread_x[i];
+        sumgwx += thread_x[i] * thread_w[i] * thread_g[i];
+      }
+    }
+  }
+
+  // Accumulate across threads
+  sumx2 = simd_sum(sumx2);
+  sumgwx = simd_sum(sumgwx);
+  if (simd_group_id == 0) {
+    local_sumx2[simd_lane_id] = 0;
+    local_sumgwx[simd_lane_id] = 0;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+  if (simd_lane_id == 0) {
+    local_sumx2[simd_group_id] = sumx2;
+    local_sumgwx[simd_group_id] = sumgwx;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+  if (simd_group_id == 0) {
+    sumx2 = simd_sum(local_sumx2[simd_lane_id]);
+    sumgwx = simd_sum(local_sumgwx[simd_lane_id]);
+    if (simd_lane_id == 0) {
+      local_meangwx[0] = sumgwx / axis_size;
+      local_normalizer[0] = metal::precise::rsqrt(sumx2 / axis_size + eps);
+    }
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+  float meangwx = local_meangwx[0];
+  float normalizer = local_normalizer[0];
+  float normalizer3 = normalizer * normalizer * normalizer;
+
+  // Write the outputs
+  gx += gid * axis_size + lid * N_READS;
+  gw += gid * axis_size + lid * N_READS;
+  if (lid * N_READS + N_READS <= axis_size) {
+    for (int i = 0; i < N_READS; i++) {
+      gx[i] = static_cast<T>(thread_g[i] * thread_w[i] * normalizer - thread_x[i] * meangwx * normalizer3);
+      gw[i] = static_cast<T>(thread_g[i] * thread_x[i] * normalizer);
+    }
+  } else {
+    for (int i = 0; i < N_READS; i++) {
+      if ((lid * N_READS + i) < axis_size) {
+        gx[i] = static_cast<T>(thread_g[i] * thread_w[i] * normalizer - thread_x[i] * meangwx * normalizer3);
+        gw[i] = static_cast<T>(thread_g[i] * thread_x[i] * normalizer);
+      }
+    }
+  }
+}
+
+template <typename T, int N_READS = RMS_N_READS>
+[[kernel]] void vjp_rms_looped(
+    const device T* x,
+    const device T* w,
+    const device T* g,
+    device T* gx,
+    device T* gw,
+    constant float& eps,
+    constant uint& axis_size,
+    constant uint& w_stride,
+    uint gid [[threadgroup_position_in_grid]],
+    uint lid [[thread_position_in_threadgroup]],
+    uint lsize [[threads_per_threadgroup]],
+    uint simd_lane_id [[thread_index_in_simdgroup]],
+    uint simd_group_id [[simdgroup_index_in_threadgroup]]) {
+  // Advance the input pointers
+  x += gid * axis_size + lid * N_READS;
+  g += gid * axis_size + lid * N_READS;
+  w += w_stride * lid * N_READS;
+
+  // Allocate registers for the accumulators
+  float sumx2 = 0;
+  float sumgwx = 0;
+
+  // Allocate shared memory to implement the reduction
+  constexpr int SIMD_SIZE = 32;
+  threadgroup float local_sumx2[SIMD_SIZE];
+  threadgroup float local_sumgwx[SIMD_SIZE];
+  threadgroup float local_normalizer[1];
+  threadgroup float local_meangwx[1];
+
+  // Read and accumulate locally
+  for (uint r = 0; r < axis_size; r += lsize * N_READS) {
+    if (r + lid * N_READS + N_READS <= axis_size) {
+      for (int i = 0; i < N_READS; i++) {
+        float xi = x[i + r];
+        float wi = w[w_stride * (i + r)];
+        float gi = g[i + r];
+
+        sumx2 += xi * xi;
+        sumgwx += xi * wi * gi;
+      }
+    } else {
+      for (int i = 0; i < N_READS; i++) {
+        if ((r + lid * N_READS + i) < axis_size) {
+          float xi = x[i + r];
+          float wi = w[w_stride * (i + r)];
+          float gi = g[i + r];
+
+          sumx2 += xi * xi;
+          sumgwx += xi * wi * gi;
+        }
+      }
+    }
+  }
+
+  // Accumulate across threads
+  sumx2 = simd_sum(sumx2);
+  sumgwx = simd_sum(sumgwx);
+  if (simd_group_id == 0) {
+    local_sumx2[simd_lane_id] = 0;
+    local_sumgwx[simd_lane_id] = 0;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+  if (simd_lane_id == 0) {
+    local_sumx2[simd_group_id] = sumx2;
+    local_sumgwx[simd_group_id] = sumgwx;
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+  if (simd_group_id == 0) {
+    sumx2 = simd_sum(local_sumx2[simd_lane_id]);
+    sumgwx = simd_sum(local_sumgwx[simd_lane_id]);
+    if (simd_lane_id == 0) {
+      local_meangwx[0] = sumgwx / axis_size;
+      local_normalizer[0] = metal::precise::rsqrt(sumx2 / axis_size + eps);
+    }
+  }
+  threadgroup_barrier(mem_flags::mem_threadgroup);
+  float meangwx = local_meangwx[0];
+  float normalizer = local_normalizer[0];
+  float normalizer3 = normalizer * normalizer * normalizer;
+
+  // Write the outputs
+  gx += gid * axis_size + lid * N_READS;
+  gw += gid * axis_size + lid * N_READS;
+  for (uint r = 0; r < axis_size; r += lsize * N_READS) {
+    if (r + lid * N_READS + N_READS <= axis_size) {
+      for (int i = 0; i < N_READS; i++) {
+        float xi = x[i + r];
+        float wi = w[w_stride * (i + r)];
+        float gi = g[i + r];
+
+        gx[i + r] = static_cast<T>(gi * wi * normalizer - xi * meangwx * normalizer3);
+        gw[i + r] = static_cast<T>(gi * xi * normalizer);
+      }
+    } else {
+      for (int i = 0; i < N_READS; i++) {
+        if ((r + lid * N_READS + i) < axis_size) {
+          float xi = x[i + r];
+          float wi = w[w_stride * (i + r)];
+          float gi = g[i + r];
+
+          gx[i + r] = static_cast<T>(gi * wi * normalizer - xi * meangwx * normalizer3);
+          gw[i + r] = static_cast<T>(gi * xi * normalizer);
+        }
+      }
+    }
+  }
+}
+
 // clang-format off
 #define instantiate_rms_single_row(name, itype)               \
   template [[host_name("rms" #name)]] [[kernel]] void         \
@@ -165,25 +375,56 @@ template <typename T, int N_READS = RMS_N_READS>
       uint gid [[thread_position_in_grid]],                   \
       uint lid [[thread_position_in_threadgroup]],            \
       uint simd_lane_id [[thread_index_in_simdgroup]],        \
-      uint simd_group_id [[simdgroup_index_in_threadgroup]]);
-
-#define instantiate_rms_looped(name, itype)                   \
-  template [[host_name("rms_looped" #name)]] [[kernel]] void  \
-  rms_looped<itype>(                                          \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]); \
+                                                              \
+  template [[host_name("vjp_rms" #name)]] [[kernel]] void     \
+  vjp_rms_single_row<itype>(                                  \
       const device itype* x,                                  \
       const device itype* w,                                  \
-      device itype* out,                                      \
+      const device itype* g,                                  \
+      device itype* gx,                                       \
+      device itype* gw,                                       \
       constant float& eps,                                    \
       constant uint& axis_size,                               \
       constant uint& w_stride,                                \
-      threadgroup float* local_inv_mean [[threadgroup(0)]],   \
-      threadgroup float* local_sums [[threadgroup(1)]],       \
       uint gid [[thread_position_in_grid]],                   \
       uint lid [[thread_position_in_threadgroup]],            \
-      uint lsize [[threads_per_threadgroup]],                 \
       uint simd_lane_id [[thread_index_in_simdgroup]],        \
       uint simd_group_id [[simdgroup_index_in_threadgroup]]);
 
+#define instantiate_rms_looped(name, itype)                      \
+  template [[host_name("rms_looped" #name)]] [[kernel]] void     \
+  rms_looped<itype>(                                             \
+      const device itype* x,                                     \
+      const device itype* w,                                     \
+      device itype* out,                                         \
+      constant float& eps,                                       \
+      constant uint& axis_size,                                  \
+      constant uint& w_stride,                                   \
+      threadgroup float* local_inv_mean [[threadgroup(0)]],      \
+      threadgroup float* local_sums [[threadgroup(1)]],          \
+      uint gid [[thread_position_in_grid]],                      \
+      uint lid [[thread_position_in_threadgroup]],               \
+      uint lsize [[threads_per_threadgroup]],                    \
+      uint simd_lane_id [[thread_index_in_simdgroup]],           \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]);    \
+                                                                 \
+  template [[host_name("vjp_rms_looped" #name)]] [[kernel]] void \
+  vjp_rms_looped<itype>(                                         \
+      const device itype* x,                                     \
+      const device itype* w,                                     \
+      const device itype* g,                                     \
+      device itype* gx,                                          \
+      device itype* gw,                                          \
+      constant float& eps,                                       \
+      constant uint& axis_size,                                  \
+      constant uint& w_stride,                                   \
+      uint gid [[thread_position_in_grid]],                      \
+      uint lid [[thread_position_in_threadgroup]],               \
+      uint lsize [[threads_per_threadgroup]],                    \
+      uint simd_lane_id [[thread_index_in_simdgroup]],           \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]);
+
 #define instantiate_rms(name, itype)      \
   instantiate_rms_single_row(name, itype) \
   instantiate_rms_looped(name, itype)

mlx/backend/metal/kernels/rope.metal

@@ -5,7 +5,7 @@
 #include "mlx/backend/metal/kernels/bf16.h"
 #include "mlx/backend/metal/kernels/utils.h"
 
-template <typename T, bool traditional>
+template <typename T, bool traditional, bool forward>
 [[kernel]] void rope(
     const device T *in [[buffer(0)]],
     device T * out [[buffer(1)]],
@@ -43,15 +43,22 @@ template <typename T, bool traditional>
   // Read and write the output
   float x1 = static_cast<float>(in[in_index_1]);
   float x2 = static_cast<float>(in[in_index_2]);
-  float rx1 = x1 * costheta - x2 * sintheta;
-  float rx2 = x1 * sintheta + x2 * costheta;
+  float rx1;
+  float rx2;
+  if (forward) {
+    rx1 = x1 * costheta - x2 * sintheta;
+    rx2 = x1 * sintheta + x2 * costheta;
+  } else {
+    rx1 = x2 * sintheta + x1 * costheta;
+    rx2 = x2 * costheta - x1 * sintheta;
+  }
   out[out_index_1] = static_cast<T>(rx1);
   out[out_index_2] = static_cast<T>(rx2);
 }
 
-#define instantiate_rope(name, type, traditional) \
+#define instantiate_rope(name, type, traditional, forward) \
   template [[host_name("rope_" #name)]] \
-  [[kernel]] void rope<type, traditional>( \
+  [[kernel]] void rope<type, traditional, forward>( \
       const device type* in [[buffer(0)]], \
       device type* out [[buffer(1)]], \
     constant const size_t strides[3], \
@@ -62,9 +69,15 @@ template <typename T, bool traditional>
     uint3 pos [[thread_position_in_grid]], \
     uint3 grid [[threads_per_grid]]);
 
-instantiate_rope(traditional_float16, half, true)
-instantiate_rope(traditional_bfloat16, bfloat16_t, true)
-instantiate_rope(traditional_float32, float, true)
-instantiate_rope(float16, half, false)
-instantiate_rope(bfloat16, bfloat16_t, false)
-instantiate_rope(float32, float, false)
+instantiate_rope(traditional_float16, half, true, true)
+instantiate_rope(traditional_bfloat16, bfloat16_t, true, true)
+instantiate_rope(traditional_float32, float, true, true)
+instantiate_rope(float16, half, false, true)
+instantiate_rope(bfloat16, bfloat16_t, false, true)
+instantiate_rope(float32, float, false, true)
+instantiate_rope(vjp_traditional_float16, half, true, false)
+instantiate_rope(vjp_traditional_bfloat16, bfloat16_t, true, false)
+instantiate_rope(vjp_traditional_float32, float, true, false)
+instantiate_rope(vjp_float16, half, false, false)
+instantiate_rope(vjp_bfloat16, bfloat16_t, false, false)
+instantiate_rope(vjp_float32, float, false, false)

mlx/backend/metal/kernels/scan.metal

@@ -451,7 +451,7 @@ instantiate_scan_helper(sum_int32_int32,         int32_t,     int32_t,     CumSu
 //instantiate_scan_helper(sum_int64_int64,         int64_t,     int64_t,     CumSum, 2)
 instantiate_scan_helper(sum_float16_float16,     half,        half,        CumSum, 4)
 instantiate_scan_helper(sum_float32_float32,     float,       float,       CumSum, 4)
-//instantiate_scan_helper(sum_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumSum, 4)
+instantiate_scan_helper(sum_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumSum, 4)
 //instantiate_scan_helper(sum_complex64_complex64, complex64_t, complex64_t, CumSum)
 //instantiate_scan_helper(prod_bool__bool_,         bool,        bool,        CumProd, 4)
 instantiate_scan_helper(prod_uint8_uint8,         uint8_t,     uint8_t,     CumProd, 4)
@@ -464,7 +464,7 @@ instantiate_scan_helper(prod_int32_int32,         int32_t,     int32_t,     CumP
 //instantiate_scan_helper(prod_int64_int64,         int64_t,     int64_t,     CumProd, 2)
 instantiate_scan_helper(prod_float16_float16,     half,        half,        CumProd, 4)
 instantiate_scan_helper(prod_float32_float32,     float,       float,       CumProd, 4)
-//instantiate_scan_helper(prod_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumProd, 4)
+instantiate_scan_helper(prod_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumProd, 4)
 //instantiate_scan_helper(prod_complex64_complex64, complex64_t, complex64_t, CumProd)
 //instantiate_scan_helper(max_bool__bool_,         bool,        bool,        CumMax, 4)
 instantiate_scan_helper(max_uint8_uint8,         uint8_t,     uint8_t,     CumMax, 4)
@@ -477,7 +477,7 @@ instantiate_scan_helper(max_int32_int32,         int32_t,     int32_t,     CumMa
 //instantiate_scan_helper(max_int64_int64,         int64_t,     int64_t,     CumMax, 2)
 instantiate_scan_helper(max_float16_float16,     half,        half,        CumMax, 4)
 instantiate_scan_helper(max_float32_float32,     float,       float,       CumMax, 4)
-//instantiate_scan_helper(max_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumMax, 4)
+instantiate_scan_helper(max_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumMax, 4)
 //instantiate_scan_helper(max_complex64_complex64, complex64_t, complex64_t, CumMax)
 //instantiate_scan_helper(min_bool__bool_,         bool,        bool,        CumMin, 4)
 instantiate_scan_helper(min_uint8_uint8,         uint8_t,     uint8_t,     CumMin, 4)
@@ -490,5 +490,5 @@ instantiate_scan_helper(min_int32_int32,         int32_t,     int32_t,     CumMi
 //instantiate_scan_helper(min_int64_int64,         int64_t,     int64_t,     CumMin, 2)
 instantiate_scan_helper(min_float16_float16,     half,        half,        CumMin, 4)
 instantiate_scan_helper(min_float32_float32,     float,       float,       CumMin, 4)
-//instantiate_scan_helper(min_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumMin, 4)
+instantiate_scan_helper(min_bfloat16_bfloat16,   bfloat16_t,  bfloat16_t,  CumMin, 4)
 //instantiate_scan_helper(min_complex64_complex64, complex64_t, complex64_t, CumMin)

mlx/backend/metal/kernels/softmax.metal

@@ -11,46 +11,48 @@ using namespace metal;
 
 template <typename T>
 inline T softmax_exp(T x) {
-  // Softmax doesn't need high precision exponential cause it is gonna be x
-  // will be in (-oo, 0] anyway and subsequently it will be divided by
-  // sum(exp(x_i)).
+  // Softmax doesn't need high precision exponential cause x is gonna be in
+  // (-oo, 0] anyway and subsequently it will be divided by sum(exp(x_i)).
   return fast::exp(x);
 }
 
-template <typename T, int N_READS = SOFTMAX_N_READS>
+template <typename T, typename AccT = T, int N_READS = SOFTMAX_N_READS>
 [[kernel]] void softmax_single_row(
     const device T* in,
     device T* out,
     constant int& axis_size,
-    threadgroup T* local_max [[threadgroup(0)]],
-    threadgroup T* local_normalizer [[threadgroup(1)]],
     uint gid [[threadgroup_position_in_grid]],
     uint _lid [[thread_position_in_threadgroup]],
     uint simd_lane_id [[thread_index_in_simdgroup]],
     uint simd_group_id [[simdgroup_index_in_threadgroup]]) {
   int lid = _lid;
 
-  T ld[N_READS];
+  constexpr int SIMD_SIZE = 32;
+
+  threadgroup AccT local_max[SIMD_SIZE];
+  threadgroup AccT local_normalizer[SIMD_SIZE];
+
+  AccT ld[N_READS];
 
   in += gid * axis_size + lid * N_READS;
   if (lid * N_READS + N_READS <= axis_size) {
-    for (int i=0; i<N_READS; i++) {
-        ld[i] = in[i];
+    for (int i = 0; i < N_READS; i++) {
+      ld[i] = AccT(in[i]);
     }
   } else {
-      for (int i = 0; i < N_READS; i++) {
-        ld[i] =
-            ((lid * N_READS + i) < axis_size) ? in[i] : T(Limits<T>::finite_min);
-      }
+    for (int i = 0; i < N_READS; i++) {
+      ld[i] = ((lid * N_READS + i) < axis_size) ? AccT(in[i])
+                                                : Limits<AccT>::finite_min;
+    }
   }
   if (simd_group_id == 0) {
-    local_max[simd_lane_id] = Limits<T>::finite_min;
+    local_max[simd_lane_id] = Limits<AccT>::finite_min;
     local_normalizer[simd_lane_id] = 0;
   }
   threadgroup_barrier(mem_flags::mem_threadgroup);
 
   // Get the max
-  T maxval = Limits<T>::finite_min;
+  AccT maxval = Limits<AccT>::finite_min;
   for (int i = 0; i < N_READS; i++) {
     maxval = (maxval < ld[i]) ? ld[i] : maxval;
   }
@@ -69,9 +71,9 @@ template <typename T, int N_READS = SOFTMAX_N_READS>
   maxval = local_max[0];
 
   // Compute exp(x_i - maxval) and store the partial sums in local_normalizer
-  T normalizer = 0;
+  AccT normalizer = 0;
   for (int i = 0; i < N_READS; i++) {
-    T exp_x = softmax_exp(ld[i] - maxval);
+    AccT exp_x = softmax_exp(ld[i] - maxval);
     ld[i] = exp_x;
     normalizer += exp_x;
   }
@@ -92,25 +94,23 @@ template <typename T, int N_READS = SOFTMAX_N_READS>
   // Normalize and write to the output
   out += gid * axis_size + lid * N_READS;
   if (lid * N_READS + N_READS <= axis_size) {
-    for (int i=0; i<N_READS; i++) {
-        out[i] = ld[i] * normalizer;
+    for (int i = 0; i < N_READS; i++) {
+      out[i] = T(ld[i] * normalizer);
     }
   } else {
-      for (int i = 0; i < N_READS; i++) {
-        if ((lid * N_READS + i) < axis_size) {
-          out[i] = ld[i] * normalizer;
-        }
+    for (int i = 0; i < N_READS; i++) {
+      if ((lid * N_READS + i) < axis_size) {
+        out[i] = T(ld[i] * normalizer);
       }
+    }
   }
 }
 
-template <typename T, int N_READS = SOFTMAX_N_READS>
+template <typename T, typename AccT = T, int N_READS = SOFTMAX_N_READS>
 [[kernel]] void softmax_looped(
     const device T* in,
     device T* out,
     constant int& axis_size,
-    threadgroup T* local_max [[threadgroup(0)]],
-    threadgroup T* local_normalizer [[threadgroup(1)]],
     uint gid [[threadgroup_position_in_grid]],
     uint lid [[thread_position_in_threadgroup]],
     uint lsize [[threads_per_threadgroup]],
@@ -118,22 +118,27 @@ template <typename T, int N_READS = SOFTMAX_N_READS>
     uint simd_group_id [[simdgroup_index_in_threadgroup]]) {
   in += gid * axis_size;
 
+  constexpr int SIMD_SIZE = 32;
+
+  threadgroup AccT local_max[SIMD_SIZE];
+  threadgroup AccT local_normalizer[SIMD_SIZE];
+
   // Get the max and the normalizer in one go
-  T prevmax;
-  T maxval = Limits<T>::finite_min;
-  T normalizer = 0;
+  AccT prevmax;
+  AccT maxval = Limits<AccT>::finite_min;
+  AccT normalizer = 0;
   for (int r = 0; r < static_cast<int>(ceildiv(axis_size, N_READS * lsize));
        r++) {
     int offset = r * lsize * N_READS + lid * N_READS;
-    T vals[N_READS];
+    AccT vals[N_READS];
     if (offset + N_READS <= axis_size) {
       for (int i = 0; i < N_READS; i++) {
-        vals[i] = in[offset + i];
+        vals[i] = AccT(in[offset + i]);
       }
     } else {
       for (int i = 0; i < N_READS; i++) {
-        vals[i] =
-            (offset + i < axis_size) ? in[offset + i] : T(Limits<T>::finite_min);
+        vals[i] = (offset + i < axis_size) ? AccT(in[offset + i])
+                                           : Limits<AccT>::finite_min;
       }
     }
     prevmax = maxval;
@@ -179,50 +184,66 @@ template <typename T, int N_READS = SOFTMAX_N_READS>
        r++) {
     int offset = r * lsize * N_READS + lid * N_READS;
     if (offset + N_READS <= axis_size) {
-      for (int i=0; i<N_READS; i++) {
-        out[offset + i] = softmax_exp(in[offset + i] - maxval) * normalizer;
+      for (int i = 0; i < N_READS; i++) {
+        out[offset + i] = T(softmax_exp(in[offset + i] - maxval) * normalizer);
       }
     } else {
       for (int i = 0; i < N_READS; i++) {
         if (offset + i < axis_size) {
-          out[offset + i] = softmax_exp(in[offset + i] - maxval) * normalizer;
+          out[offset + i] =
+              T(softmax_exp(in[offset + i] - maxval) * normalizer);
         }
       }
     }
   }
 }
 
-#define instantiate_softmax_single_row(name, itype)           \
+// clang-format off
+#define instantiate_softmax(name, itype)  \
   template [[host_name("softmax_" #name)]] [[kernel]] void    \
   softmax_single_row<itype>(                                  \
       const device itype* in,                                 \
       device itype* out,                                      \
       constant int& axis_size,                                \
-      threadgroup itype* local_max [[threadgroup(0)]],        \
-      threadgroup itype* local_normalizer [[threadgroup(1)]], \
       uint gid [[thread_position_in_grid]],                   \
       uint _lid [[thread_position_in_threadgroup]],           \
       uint simd_lane_id [[thread_index_in_simdgroup]],        \
-      uint simd_group_id [[simdgroup_index_in_threadgroup]]);
-
-#define instantiate_softmax_looped(name, itype)                   \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]); \
   template [[host_name("softmax_looped_" #name)]] [[kernel]] void \
   softmax_looped<itype>(                                          \
       const device itype* in,                                     \
       device itype* out,                                          \
       constant int& axis_size,                                    \
-      threadgroup itype* local_max [[threadgroup(0)]],            \
-      threadgroup itype* local_normalizer [[threadgroup(1)]],     \
       uint gid [[threadgroup_position_in_grid]],                  \
       uint lid [[thread_position_in_threadgroup]],                \
       uint lsize [[threads_per_threadgroup]],                     \
       uint simd_lane_id [[thread_index_in_simdgroup]],            \
       uint simd_group_id [[simdgroup_index_in_threadgroup]]);
 
-#define instantiate_softmax(name, itype)      \
-  instantiate_softmax_single_row(name, itype) \
-      instantiate_softmax_looped(name, itype)
+#define instantiate_softmax_precise(name, itype)                   \
+  template [[host_name("softmax_precise_" #name)]] [[kernel]] void \
+  softmax_single_row<itype, float>(                                \
+      const device itype* in,                                      \
+      device itype* out,                                           \
+      constant int& axis_size,                                     \
+      uint gid [[thread_position_in_grid]],                        \
+      uint _lid [[thread_position_in_threadgroup]],                \
+      uint simd_lane_id [[thread_index_in_simdgroup]],             \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]);      \
+  template [[host_name("softmax_looped_precise_" #name)]] [[kernel]] void \
+  softmax_looped<itype, float>(                                           \
+      const device itype* in,                                             \
+      device itype* out,                                                  \
+      constant int& axis_size,                                            \
+      uint gid [[threadgroup_position_in_grid]],                          \
+      uint lid [[thread_position_in_threadgroup]],                        \
+      uint lsize [[threads_per_threadgroup]],                             \
+      uint simd_lane_id [[thread_index_in_simdgroup]],                    \
+      uint simd_group_id [[simdgroup_index_in_threadgroup]]);
 
 instantiate_softmax(float32, float)
 instantiate_softmax(float16, half)
 instantiate_softmax(bfloat16, bfloat16_t)
+instantiate_softmax_precise(float16, half)
+instantiate_softmax_precise(bfloat16, bfloat16_t)
+// clang-format on

mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_l.h

@@ -394,7 +394,7 @@ struct Conv2DWeightBlockLoader {
       const constant ImplicitGemmConv2DParams* gemm_params_,
       uint simd_group_id [[simdgroup_index_in_threadgroup]],
       uint simd_lane_id [[thread_index_in_simdgroup]])
-      : src_ld(params_->wt_strides[0]),
+      : src_ld(params_ -> wt_strides[0]),
         thread_idx(simd_group_id * 32 + simd_lane_id),
         bi(thread_idx / TCOLS),
         bj(vec_size * (thread_idx % TCOLS)),

mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_n.h

@@ -244,7 +244,7 @@ struct Conv2DWeightBlockLoaderSmallChannels {
       const constant ImplicitGemmConv2DParams* gemm_params_,
       uint simd_group_id [[simdgroup_index_in_threadgroup]],
       uint simd_lane_id [[thread_index_in_simdgroup]])
-      : src_ld(params_->wt_strides[0]),
+      : src_ld(params_ -> wt_strides[0]),
         thread_idx(simd_group_id * 32 + simd_lane_id),
         bi(thread_idx / TCOLS),
         bj(vec_size * (thread_idx % TCOLS)),

mlx/backend/metal/kernels/steel/conv/loaders/loader_general.h

@@ -220,7 +220,7 @@ struct Conv2DWeightBlockLoaderGeneral {
       const short base_ww_,
       uint simd_group_id [[simdgroup_index_in_threadgroup]],
       uint simd_lane_id [[thread_index_in_simdgroup]])
-      : src_ld(params_->wt_strides[0]),
+      : src_ld(params_ -> wt_strides[0]),
         thread_idx(simd_group_id * 32 + simd_lane_id),
         bi(thread_idx / TCOLS),
         bj(vec_size * (thread_idx % TCOLS)),

mlx/backend/metal/kernels/unary.h

@@ -7,6 +7,7 @@
 
 #include "mlx/backend/metal/kernels/bf16.h"
 #include "mlx/backend/metal/kernels/erf.h"
+#include "mlx/backend/metal/kernels/expm1f.h"
 #include "mlx/backend/metal/kernels/utils.h"
 
 namespace {
@@ -183,6 +184,13 @@ struct Exp {
   }
 };
 
+struct Expm1 {
+  template <typename T>
+  T operator()(T x) {
+    return static_cast<T>(expm1f(static_cast<float>(x)));
+  };
+};
+
 struct Floor {
   template <typename T>
   T operator()(T x) {

mlx/backend/metal/kernels/unary.metal

@@ -71,6 +71,7 @@ instantiate_unary_types(ceil, Ceil)
 instantiate_unary_float(cos, Cos)
 instantiate_unary_float(cosh, Cosh)
 instantiate_unary_float(exp, Exp)
+instantiate_unary_float(expm1, Expm1)
 instantiate_unary_types(floor, Floor)
 instantiate_unary_float(log, Log)
 instantiate_unary_float(log2, Log2)

mlx/backend/metal/matmul.cpp

@@ -197,8 +197,8 @@ inline auto collapse_batches(const array& a, const array& b) {
   std::vector<int> B_bshape{b.shape().begin(), b.shape().end() - 2};
   if (A_bshape != B_bshape) {
     std::ostringstream msg;
-    msg << "[matmul] Got matrices with incorrectly broadcasted shapes: "
-        << "A " << a.shape() << ", B " << b.shape() << ".";
+    msg << "[matmul] Got matrices with incorrectly broadcasted shapes: " << "A "
+        << a.shape() << ", B " << b.shape() << ".";
     throw std::runtime_error(msg.str());
   }
 
@@ -227,9 +227,8 @@ inline auto collapse_batches(const array& a, const array& b, const array& c) {
   std::vector<int> C_bshape{c.shape().begin(), c.shape().end() - 2};
   if (A_bshape != B_bshape || A_bshape != C_bshape) {
     std::ostringstream msg;
-    msg << "[addmm] Got matrices with incorrectly broadcasted shapes: "
-        << "A " << a.shape() << ", B " << b.shape() << ", B " << c.shape()
-        << ".";
+    msg << "[addmm] Got matrices with incorrectly broadcasted shapes: " << "A "
+        << a.shape() << ", B " << b.shape() << ", B " << c.shape() << ".";
     throw std::runtime_error(msg.str());
   }
 
@@ -332,11 +331,11 @@ void steel_matmul(
           << (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
           << type_to_name(C_split) << "_bm" << bm << "_bn" << bn << "_bk" << bk
           << "_wm" << wm << "_wn" << wn << "_MN_"
-          << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
-          << "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned";
+          << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
+          << ((K % bk == 0) ? "t" : "n") << "aligned";
 
     // Encode and dispatch gemm kernel
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
@@ -360,9 +359,9 @@ void steel_matmul(
     MTL::Size group_dims = MTL::Size(32, wn, wm);
     MTL::Size grid_dims = MTL::Size(tn, tm, split_k_partitions);
 
-    set_array_buffer(compute_encoder, a, 0);
-    set_array_buffer(compute_encoder, b, 1);
-    set_array_buffer(compute_encoder, C_split, 2);
+    compute_encoder.set_input_array(a, 0);
+    compute_encoder.set_input_array(b, 1);
+    compute_encoder.set_output_array(C_split, 2);
 
     compute_encoder->setBytes(&params, sizeof(GEMMSpiltKParams), 3);
     compute_encoder->dispatchThreadgroups(grid_dims, group_dims);
@@ -380,8 +379,8 @@ void steel_matmul(
       compute_encoder->setComputePipelineState(kernel);
 
       // Set the arguments for the kernel
-      set_array_buffer(compute_encoder, C_split, 0);
-      set_array_buffer(compute_encoder, out, 1);
+      compute_encoder.set_input_array(C_split, 0);
+      compute_encoder.set_output_array(out, 1);
       compute_encoder->setBytes(&split_k_partitions, sizeof(int), 2);
       compute_encoder->setBytes(&split_k_partition_stride, sizeof(int), 3);
       compute_encoder->setBytes(&N, sizeof(int), 4);
@@ -422,11 +421,11 @@ void steel_matmul(
         << (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
         << type_to_name(out) << "_bm" << bm << "_bn" << bn << "_bk" << bk
         << "_wm" << wm << "_wn" << wn << "_MN_"
-        << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
-        << "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned";
+        << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
+        << ((K % bk == 0) ? "t" : "n") << "aligned";
 
   // Encode and dispatch kernel
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -467,9 +466,9 @@ void steel_matmul(
       batch_strides.end(), B_batch_stride.begin(), B_batch_stride.end());
 
   // Launch kernel
-  set_array_buffer(compute_encoder, a, 0);
-  set_array_buffer(compute_encoder, b, 1);
-  set_array_buffer(compute_encoder, out, 3);
+  compute_encoder.set_input_array(a, 0);
+  compute_encoder.set_input_array(b, 1);
+  compute_encoder.set_output_array(out, 3);
 
   compute_encoder->setBytes(&params, sizeof(GEMMParams), 4);
 
@@ -488,7 +487,7 @@ void steel_matmul(
 
 void Matmul::eval_gpu(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 2);
-  if (!is_floating_point(out.dtype())) {
+  if (!issubdtype(out.dtype(), floating)) {
     throw std::runtime_error(
         "[matmul] Does not yet support non-floating point types.");
   }
@@ -622,7 +621,7 @@ void Matmul::eval_gpu(const std::vector<array>& inputs, array& out) {
     kname << "_nc" << !contiguous_kernel << "_axpby0";
 
     // Encode and dispatch kernel
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
@@ -630,9 +629,9 @@ void Matmul::eval_gpu(const std::vector<array>& inputs, array& out) {
     MTL::Size group_dims = MTL::Size(bn, bm, 1);
     MTL::Size grid_dims = MTL::Size(n_tgp, 1, batch_size_out);
 
-    set_array_buffer(compute_encoder, mat, 0);
-    set_array_buffer(compute_encoder, vec, 1);
-    set_array_buffer(compute_encoder, out, 3);
+    compute_encoder.set_input_array(mat, 0);
+    compute_encoder.set_input_array(vec, 1);
+    compute_encoder.set_output_array(out, 3);
 
     compute_encoder->setBytes(&in_vector_len, sizeof(int), 4);
     compute_encoder->setBytes(&out_vector_len, sizeof(int), 5);
@@ -696,7 +695,7 @@ void Matmul::eval_gpu(const std::vector<array>& inputs, array& out) {
 
 void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 3);
-  if (!is_floating_point(out.dtype())) {
+  if (!issubdtype(out.dtype(), floating)) {
     throw std::runtime_error(
         "[matmul] Does not yet support non-floating point types.");
   }
@@ -834,7 +833,7 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
     kname << "_nc" << !contiguous_kernel << "_axpby1";
 
     // Encode and dispatch kernel
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
@@ -842,10 +841,10 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
     MTL::Size group_dims = MTL::Size(bn, bm, 1);
     MTL::Size grid_dims = MTL::Size(n_tgp, 1, batch_size_out);
 
-    set_array_buffer(compute_encoder, mat, 0);
-    set_array_buffer(compute_encoder, vec, 1);
-    set_array_buffer(compute_encoder, c, 2);
-    set_array_buffer(compute_encoder, out, 3);
+    compute_encoder.set_input_array(mat, 0);
+    compute_encoder.set_input_array(vec, 1);
+    compute_encoder.set_input_array(c, 2);
+    compute_encoder.set_output_array(out, 3);
 
     compute_encoder->setBytes(&in_vector_len, sizeof(int), 4);
     compute_encoder->setBytes(&out_vector_len, sizeof(int), 5);
@@ -903,11 +902,11 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
           << (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
           << type_to_name(C_split) << "_bm" << bm << "_bn" << bn << "_bk" << bk
           << "_wm" << wm << "_wn" << wn << "_MN_"
-          << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
-          << "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned";
+          << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
+          << ((K % bk == 0) ? "t" : "n") << "aligned";
 
     // Encode and dispatch gemm kernel
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
@@ -931,9 +930,9 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
     MTL::Size group_dims = MTL::Size(32, wn, wm);
     MTL::Size grid_dims = MTL::Size(tn, tm, split_k_partitions);
 
-    set_array_buffer(compute_encoder, a, 0);
-    set_array_buffer(compute_encoder, b, 1);
-    set_array_buffer(compute_encoder, C_split, 2);
+    compute_encoder.set_input_array(a, 0);
+    compute_encoder.set_input_array(b, 1);
+    compute_encoder.set_output_array(C_split, 2);
 
     compute_encoder->setBytes(&params, sizeof(GEMMSpiltKParams), 3);
     compute_encoder->dispatchThreadgroups(grid_dims, group_dims);
@@ -946,12 +945,12 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
       compute_encoder->setComputePipelineState(kernel);
 
       // Set the arguments for the kernel
-      set_array_buffer(compute_encoder, C_split, 0);
-      set_array_buffer(compute_encoder, out, 1);
+      compute_encoder.set_input_array(C_split, 0);
+      compute_encoder.set_output_array(out, 1);
       compute_encoder->setBytes(&split_k_partitions, sizeof(int), 2);
       compute_encoder->setBytes(&split_k_partition_stride, sizeof(int), 3);
       compute_encoder->setBytes(&N, sizeof(int), 4);
-      set_array_buffer(compute_encoder, c, 5);
+      compute_encoder.set_input_array(c, 5);
       compute_encoder->setBytes(&ldc, sizeof(int), 6);
       compute_encoder->setBytes(&fdc, sizeof(int), 7);
       compute_encoder->setBytes(&alpha_, sizeof(float), 8);
@@ -992,12 +991,12 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
         << (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
         << type_to_name(out) << "_bm" << bm << "_bn" << bn << "_bk" << bk
         << "_wm" << wm << "_wn" << wn << "_MN_"
-        << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
-        << "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned"
+        << ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
+        << ((K % bk == 0) ? "t" : "n") << "aligned"
         << ((alpha_ == 1. && beta_ == 1.) ? "_add" : "_axpby");
 
   // Encode and dispatch kernel
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -1045,10 +1044,10 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
       batch_strides.end(), C_batch_stride.begin(), C_batch_stride.end());
 
   // Launch kernel
-  set_array_buffer(compute_encoder, a, 0);
-  set_array_buffer(compute_encoder, b, 1);
-  set_array_buffer(compute_encoder, c, 2);
-  set_array_buffer(compute_encoder, out, 3);
+  compute_encoder.set_input_array(a, 0);
+  compute_encoder.set_input_array(b, 1);
+  compute_encoder.set_input_array(c, 2);
+  compute_encoder.set_output_array(out, 3);
 
   compute_encoder->setBytes(&gemm_params, sizeof(GEMMParams), 4);
   compute_encoder->setBytes(&params, sizeof(GEMMAddMMParams), 5);

mlx/backend/metal/metal.cpp

@@ -1,10 +1,10 @@
 // Copyright © 2023-2024 Apple Inc.
-
 #include <cstdlib>
 #include <future>
 #include <memory>
 
 #include "mlx/backend/metal/device.h"
+#include "mlx/backend/metal/utils.h"
 #include "mlx/primitives.h"
 #include "mlx/scheduler.h"
 
@@ -74,6 +74,8 @@ std::function<void()> make_task(
       if (arr.is_tracer()) {
         inputs = arr.inputs();
       }
+
+      debug_set_primitive_buffer_label(command_buffer, arr.primitive());
       arr.primitive().eval_gpu(arr.inputs(), outputs);
     }
     std::vector<std::shared_ptr<array::Data>> buffers;
@@ -86,7 +88,6 @@ std::function<void()> make_task(
     if (!arr.is_tracer()) {
       arr.detach();
     }
-
     if (p) {
       metal::device(s.device).end_encoding(s.index);
       scheduler::notify_new_task(s);
@@ -108,4 +109,31 @@ std::function<void()> make_task(
   return task;
 }
 
+bool start_capture(std::string path, id object) {
+  auto pool = new_scoped_memory_pool();
+
+  auto descriptor = MTL::CaptureDescriptor::alloc()->init();
+  descriptor->setCaptureObject(object);
+
+  if (path.length() > 0) {
+    auto string = NS::String::string(path.c_str(), NS::UTF8StringEncoding);
+    auto url = NS::URL::fileURLWithPath(string);
+    descriptor->setDestination(MTL::CaptureDestinationGPUTraceDocument);
+    descriptor->setOutputURL(url);
+  }
+
+  auto manager = MTL::CaptureManager::sharedCaptureManager();
+  return manager->startCapture(descriptor, nullptr);
+}
+
+bool start_capture(std::string path) {
+  auto& device = metal::device(mlx::core::Device::gpu);
+  return start_capture(path, device.mtl_device());
+}
+
+void stop_capture() {
+  auto manager = MTL::CaptureManager::sharedCaptureManager();
+  manager->stopCapture();
+}
+
 } // namespace mlx::core::metal

mlx/backend/metal/metal.h

@@ -2,15 +2,11 @@
 
 #pragma once
 
-#include <future>
-#include <memory>
-#include <vector>
-
 #include "mlx/array.h"
-#include "mlx/stream.h"
 
 namespace mlx::core::metal {
 
+/* Check if the Metal backend is available. */
 bool is_available();
 
 /* Get the actively used memory in bytes.
@@ -58,12 +54,8 @@ size_t set_memory_limit(size_t limit, bool relaxed = true);
  * */
 size_t set_cache_limit(size_t limit);
 
-void new_stream(Stream stream);
-std::shared_ptr<void> new_scoped_memory_pool();
-
-std::function<void()> make_task(
-    array& arr,
-    std::vector<std::shared_future<void>> deps,
-    std::shared_ptr<std::promise<void>> p);
+/** Capture a GPU trace, saving it to an absolute file `path` */
+bool start_capture(std::string path = "");
+void stop_capture();
 
 } // namespace mlx::core::metal

mlx/backend/metal/metal_impl.h

@@ -0,0 +1,22 @@
+// Copyright © 2023-2024 Apple Inc.
+
+#pragma once
+
+#include <future>
+#include <memory>
+#include <vector>
+
+#include "mlx/array.h"
+#include "mlx/stream.h"
+
+namespace mlx::core::metal {
+
+void new_stream(Stream stream);
+std::shared_ptr<void> new_scoped_memory_pool();
+
+std::function<void()> make_task(
+    array& arr,
+    std::vector<std::shared_future<void>> deps,
+    std::shared_ptr<std::promise<void>> p);
+
+} // namespace mlx::core::metal

mlx/backend/metal/normalization.cpp

@@ -4,6 +4,7 @@
 #include "mlx/backend/metal/copy.h"
 #include "mlx/backend/metal/device.h"
 #include "mlx/backend/metal/kernels/defines.h"
+#include "mlx/backend/metal/reduce.h"
 #include "mlx/backend/metal/utils.h"
 #include "mlx/fast_primitives.h"
 
@@ -18,7 +19,7 @@ void RMSNorm::eval_gpu(
 
   // Make sure that the last dimension is contiguous
   std::vector<array> copies;
-  auto check_input = [&copies, &s](const array& x) {
+  auto check_input = [&copies, &s](const array& x) -> const array& {
     bool no_copy = x.strides()[x.ndim() - 1] == 1;
     if (x.ndim() > 1) {
       auto s = x.strides()[x.ndim() - 2];
@@ -27,10 +28,9 @@ void RMSNorm::eval_gpu(
     if (no_copy) {
       return x;
     } else {
-      array x_copy(x.shape(), x.dtype(), nullptr, {});
-      copy_gpu(x, x_copy, CopyType::General, s);
-      copies.push_back(x_copy);
-      return x_copy;
+      copies.push_back(array(x.shape(), x.dtype(), nullptr, {}));
+      copy_gpu(x, copies.back(), CopyType::General, s);
+      return copies.back();
     }
   };
   const array& x = check_input(inputs[0]);
@@ -57,7 +57,7 @@ void RMSNorm::eval_gpu(
     op_name += "_looped";
   }
   op_name += type_to_name(out);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   {
     auto kernel = d.get_kernel(op_name);
 
@@ -79,10 +79,10 @@ void RMSNorm::eval_gpu(
 
     uint32_t w_stride = w.strides()[0];
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(
-        compute_encoder, x.data_shared_ptr() == nullptr ? out : x, 0);
-    set_array_buffer(compute_encoder, w, 1);
-    set_array_buffer(compute_encoder, out, 2);
+    compute_encoder.set_input_array(
+        x.data_shared_ptr() == nullptr ? out : x, 0);
+    compute_encoder.set_input_array(w, 1);
+    compute_encoder.set_output_array(out, 2);
     compute_encoder->setBytes(&eps_, sizeof(float), 3);
     compute_encoder->setBytes(&axis_size, sizeof(int), 4);
     compute_encoder->setBytes(&w_stride, sizeof(uint32_t), 5);
@@ -95,6 +95,113 @@ void RMSNorm::eval_gpu(
       [copies](MTL::CommandBuffer*) mutable { copies.clear(); });
 }
 
+void RMSNormVJP::eval_gpu(
+    const std::vector<array>& inputs,
+    std::vector<array>& outputs) {
+  auto& s = stream();
+  auto& d = metal::device(s.device);
+
+  // Ensure row contiguity. We could relax this step by checking that the array
+  // is contiguous (no broadcasts or holes) and that the input strides are the
+  // same as the cotangent strides but for now this is simpler.
+  std::vector<array> copies;
+  auto check_input = [&copies, &s](const array& x) -> const array& {
+    if (x.flags().row_contiguous) {
+      return x;
+    }
+    copies.push_back(array(x.shape(), x.dtype(), nullptr, {}));
+    copy_gpu(x, copies.back(), CopyType::General, s);
+    return copies.back();
+  };
+  const array& x = check_input(inputs[0]);
+  const array& w = inputs[1];
+  const array& g = check_input(inputs[2]);
+  array& gx = outputs[0];
+  array& gw = outputs[1];
+
+  // Allocate space for the outputs
+  bool x_in_gx = false;
+  bool g_in_gx = false;
+  if (x.is_donatable()) {
+    gx.move_shared_buffer(x);
+    x_in_gx = true;
+  } else if (g.is_donatable()) {
+    gx.move_shared_buffer(g);
+    g_in_gx = true;
+  } else {
+    gx.set_data(allocator::malloc_or_wait(gx.nbytes()));
+  }
+
+  auto axis_size = static_cast<uint32_t>(x.shape().back());
+  int n_rows = x.data_size() / axis_size;
+
+  // Allocate a temporary to store the gradients for w and initialize the
+  // gradient accumulator to 0.
+  array gw_temp({n_rows, x.shape().back()}, gw.dtype(), nullptr, {});
+  bool g_in_gw = false;
+  if (!g_in_gx && g.is_donatable()) {
+    gw_temp.move_shared_buffer(g);
+    g_in_gw = true;
+  } else {
+    gw_temp.set_data(allocator::malloc_or_wait(gw_temp.nbytes()));
+  }
+  copies.push_back(gw_temp);
+  {
+    array zero(0, gw.dtype());
+    copy_gpu(zero, gw, CopyType::Scalar, s);
+    copies.push_back(std::move(zero));
+  }
+
+  const int simd_size = 32;
+  const int n_reads = RMS_N_READS;
+  const int looped_limit = RMS_LOOPED_LIMIT;
+  std::string op_name = "vjp_rms";
+  if (axis_size > looped_limit) {
+    op_name += "_looped";
+  }
+  op_name += type_to_name(gx);
+  auto& compute_encoder = d.get_command_encoder(s.index);
+  {
+    auto kernel = d.get_kernel(op_name);
+
+    MTL::Size grid_dims, group_dims;
+    if (axis_size <= looped_limit) {
+      size_t threadgroup_needed = (axis_size + n_reads - 1) / n_reads;
+      size_t simds_needed = (threadgroup_needed + simd_size - 1) / simd_size;
+      size_t threadgroup_size = simd_size * simds_needed;
+      assert(threadgroup_size <= kernel->maxTotalThreadsPerThreadgroup());
+      size_t n_threads = n_rows * threadgroup_size;
+      grid_dims = MTL::Size(n_threads, 1, 1);
+      group_dims = MTL::Size(threadgroup_size, 1, 1);
+    } else {
+      size_t threadgroup_size = kernel->maxTotalThreadsPerThreadgroup();
+      size_t n_threads = n_rows * threadgroup_size;
+      grid_dims = MTL::Size(n_threads, 1, 1);
+      group_dims = MTL::Size(threadgroup_size, 1, 1);
+    }
+
+    uint32_t w_stride = w.strides()[0];
+    compute_encoder->setComputePipelineState(kernel);
+    compute_encoder.set_input_array(x_in_gx ? gx : x, 0);
+    compute_encoder.set_input_array(w, 1);
+    compute_encoder.set_input_array(g_in_gx ? gx : (g_in_gw ? gw_temp : g), 2);
+    compute_encoder.set_output_array(gx, 3);
+    compute_encoder.set_output_array(gw_temp, 4);
+    compute_encoder->setBytes(&eps_, sizeof(float), 5);
+    compute_encoder->setBytes(&axis_size, sizeof(int), 6);
+    compute_encoder->setBytes(&w_stride, sizeof(uint32_t), 7);
+    compute_encoder->dispatchThreads(grid_dims, group_dims);
+  }
+
+  ReductionPlan plan(
+      ReductionOpType::ContiguousStridedReduce, {n_rows}, {axis_size});
+  strided_reduce_general_dispatch(
+      gw_temp, gw, "sum", plan, {0}, compute_encoder, d, s);
+
+  d.get_command_buffer(s.index)->addCompletedHandler(
+      [copies](MTL::CommandBuffer*) mutable { copies.clear(); });
+}
+
 void LayerNorm::eval_gpu(
     const std::vector<array>& inputs,
     std::vector<array>& outputs) {
@@ -104,7 +211,7 @@ void LayerNorm::eval_gpu(
 
   // Make sure that the last dimension is contiguous
   std::vector<array> copies;
-  auto check_input = [&copies, &s](const array& x) {
+  auto check_input = [&copies, &s](const array& x) -> const array& {
     bool no_copy = x.strides()[x.ndim() - 1] == 1;
     if (x.ndim() > 1) {
       auto s = x.strides()[x.ndim() - 2];
@@ -113,10 +220,9 @@ void LayerNorm::eval_gpu(
     if (no_copy) {
       return x;
     } else {
-      array x_copy(x.shape(), x.dtype(), nullptr, {});
-      copy_gpu(x, x_copy, CopyType::General, s);
-      copies.push_back(x_copy);
-      return x_copy;
+      copies.push_back(array(x.shape(), x.dtype(), nullptr, {}));
+      copy_gpu(x, copies.back(), CopyType::General, s);
+      return copies.back();
     }
   };
   const array& x = check_input(inputs[0]);
@@ -144,7 +250,7 @@ void LayerNorm::eval_gpu(
     op_name += "_looped";
   }
   op_name += type_to_name(out);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   {
     auto kernel = d.get_kernel(op_name);
 
@@ -167,11 +273,11 @@ void LayerNorm::eval_gpu(
     uint32_t w_stride = (w.ndim() == 1) ? w.strides()[0] : 0;
     uint32_t b_stride = (b.ndim() == 1) ? b.strides()[0] : 0;
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(
-        compute_encoder, x.data_shared_ptr() == nullptr ? out : x, 0);
-    set_array_buffer(compute_encoder, w, 1);
-    set_array_buffer(compute_encoder, b, 2);
-    set_array_buffer(compute_encoder, out, 3);
+    compute_encoder.set_input_array(
+        x.data_shared_ptr() == nullptr ? out : x, 0);
+    compute_encoder.set_input_array(w, 1);
+    compute_encoder.set_input_array(b, 2);
+    compute_encoder.set_output_array(out, 3);
     compute_encoder->setBytes(&eps_, sizeof(float), 4);
     compute_encoder->setBytes(&axis_size, sizeof(int), 5);
     compute_encoder->setBytes(&w_stride, sizeof(uint32_t), 6);
@@ -182,4 +288,131 @@ void LayerNorm::eval_gpu(
       [copies](MTL::CommandBuffer*) mutable { copies.clear(); });
 }
 
+void LayerNormVJP::eval_gpu(
+    const std::vector<array>& inputs,
+    std::vector<array>& outputs) {
+  auto& s = stream();
+  auto& d = metal::device(s.device);
+
+  // Ensure row contiguity. We could relax this step by checking that the array
+  // is contiguous (no broadcasts or holes) and that the input strides are the
+  // same as the cotangent strides but for now this is simpler.
+  std::vector<array> copies;
+  auto check_input = [&copies, &s](const array& x) -> const array& {
+    if (x.flags().row_contiguous) {
+      return x;
+    }
+    copies.push_back(array(x.shape(), x.dtype(), nullptr, {}));
+    copy_gpu(x, copies.back(), CopyType::General, s);
+    return copies.back();
+  };
+  const array& x = check_input(inputs[0]);
+  const array& w = inputs[1];
+  const array& b = inputs[2];
+  const array& g = check_input(inputs[3]);
+  array& gx = outputs[0];
+  array& gw = outputs[1];
+  array& gb = outputs[2];
+
+  // Allocate space for the outputs
+  bool x_in_gx = false;
+  bool g_in_gx = false;
+  if (x.is_donatable()) {
+    gx.move_shared_buffer(x);
+    x_in_gx = true;
+  } else if (g.is_donatable()) {
+    gx.move_shared_buffer(g);
+    g_in_gx = true;
+  } else {
+    gx.set_data(allocator::malloc_or_wait(gx.nbytes()));
+  }
+
+  auto axis_size = static_cast<uint32_t>(x.shape().back());
+  int n_rows = x.data_size() / axis_size;
+
+  // Allocate a temporary to store the gradients for w and initialize the
+  // gradient accumulator to 0.
+  array gw_temp({n_rows, x.shape().back()}, gw.dtype(), nullptr, {});
+  bool g_in_gw = false;
+  if (!g_in_gx && g.is_donatable()) {
+    gw_temp.move_shared_buffer(g);
+    g_in_gw = true;
+  } else {
+    gw_temp.set_data(allocator::malloc_or_wait(gw_temp.nbytes()));
+  }
+  copies.push_back(gw_temp);
+  {
+    array zero(0, gw.dtype());
+    copy_gpu(zero, gw, CopyType::Scalar, s);
+    copy_gpu(zero, gb, CopyType::Scalar, s);
+    copies.push_back(std::move(zero));
+  }
+
+  // Finish with the gradient for b in case we had a b
+  auto& compute_encoder = d.get_command_encoder(s.index);
+  if (gb.ndim() == 1 && gb.size() == axis_size) {
+    ReductionPlan plan(
+        ReductionOpType::ContiguousStridedReduce, {n_rows}, {axis_size});
+    strided_reduce_general_dispatch(
+        g_in_gx ? gx : (g_in_gw ? gw_temp : g),
+        gb,
+        "sum",
+        plan,
+        {0},
+        compute_encoder,
+        d,
+        s);
+  }
+
+  const int simd_size = 32;
+  const int n_reads = RMS_N_READS;
+  const int looped_limit = RMS_LOOPED_LIMIT;
+  std::string op_name = "vjp_layer_norm";
+  if (axis_size > looped_limit) {
+    op_name += "_looped";
+  }
+  op_name += type_to_name(gx);
+  {
+    auto kernel = d.get_kernel(op_name);
+
+    MTL::Size grid_dims, group_dims;
+    if (axis_size <= looped_limit) {
+      size_t threadgroup_needed = (axis_size + n_reads - 1) / n_reads;
+      size_t simds_needed = (threadgroup_needed + simd_size - 1) / simd_size;
+      size_t threadgroup_size = simd_size * simds_needed;
+      assert(threadgroup_size <= kernel->maxTotalThreadsPerThreadgroup());
+      size_t n_threads = n_rows * threadgroup_size;
+      grid_dims = MTL::Size(n_threads, 1, 1);
+      group_dims = MTL::Size(threadgroup_size, 1, 1);
+    } else {
+      size_t threadgroup_size = kernel->maxTotalThreadsPerThreadgroup();
+      size_t n_threads = n_rows * threadgroup_size;
+      grid_dims = MTL::Size(n_threads, 1, 1);
+      group_dims = MTL::Size(threadgroup_size, 1, 1);
+    }
+
+    uint32_t w_stride = (w.ndim() == 1) ? w.strides()[0] : 0;
+    compute_encoder->setComputePipelineState(kernel);
+    compute_encoder.set_input_array(x_in_gx ? gx : x, 0);
+    compute_encoder.set_input_array(w, 1);
+    compute_encoder.set_input_array(g_in_gx ? gx : (g_in_gw ? gw_temp : g), 2);
+    compute_encoder.set_output_array(gx, 3);
+    compute_encoder.set_output_array(gw_temp, 4);
+    compute_encoder->setBytes(&eps_, sizeof(float), 5);
+    compute_encoder->setBytes(&axis_size, sizeof(int), 6);
+    compute_encoder->setBytes(&w_stride, sizeof(uint32_t), 7);
+    compute_encoder->dispatchThreads(grid_dims, group_dims);
+  }
+
+  if (gw.ndim() == 1 && gw.size() == axis_size) {
+    ReductionPlan plan(
+        ReductionOpType::ContiguousStridedReduce, {n_rows}, {axis_size});
+    strided_reduce_general_dispatch(
+        gw_temp, gw, "sum", plan, {0}, compute_encoder, d, s);
+  }
+
+  d.get_command_buffer(s.index)->addCompletedHandler(
+      [copies](MTL::CommandBuffer*) mutable { copies.clear(); });
+}
+
 } // namespace mlx::core::fast

mlx/backend/metal/primitives.cpp

@@ -68,18 +68,18 @@ void binary_op(
   auto& s = out.primitive().stream();
   auto& d = metal::device(s.device);
   auto kernel = d.get_kernel(kname.str());
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
   // - If a is donated it goes to the first output
   // - If b is donated it goes to the first output if a was not donated
   //   otherwise it goes to the second output
   bool donate_a = a.data_shared_ptr() == nullptr;
   bool donate_b = b.data_shared_ptr() == nullptr;
-  set_array_buffer(compute_encoder, donate_a ? outputs[0] : a, 0);
-  set_array_buffer(
-      compute_encoder, donate_b ? (donate_a ? outputs[1] : outputs[0]) : b, 1);
-  set_array_buffer(compute_encoder, outputs[0], 2);
-  set_array_buffer(compute_encoder, outputs[1], 3);
+  compute_encoder.set_input_array(donate_a ? outputs[0] : a, 0);
+  compute_encoder.set_input_array(
+      donate_b ? (donate_a ? outputs[1] : outputs[0]) : b, 1);
+  compute_encoder.set_output_array(outputs[0], 2);
+  compute_encoder.set_output_array(outputs[1], 3);
 
   if (bopt == BinaryOpType::General) {
     auto ndim = shape.size();
@@ -167,13 +167,13 @@ void binary_op(
   auto& s = out.primitive().stream();
   auto& d = metal::device(s.device);
   auto kernel = d.get_kernel(kname.str());
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
   bool donate_a = a.data_shared_ptr() == nullptr;
   bool donate_b = b.data_shared_ptr() == nullptr;
-  set_array_buffer(compute_encoder, donate_a ? out : a, 0);
-  set_array_buffer(compute_encoder, donate_b ? out : b, 1);
-  set_array_buffer(compute_encoder, out, 2);
+  compute_encoder.set_input_array(donate_a ? out : a, 0);
+  compute_encoder.set_input_array(donate_b ? out : b, 1);
+  compute_encoder.set_output_array(out, 2);
 
   if (bopt == BinaryOpType::General) {
     auto ndim = shape.size();
@@ -253,12 +253,12 @@ void ternary_op(
   auto& s = out.primitive().stream();
   auto& d = metal::device(s.device);
   auto kernel = d.get_kernel(kname.str());
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
-  set_array_buffer(compute_encoder, a, 0);
-  set_array_buffer(compute_encoder, b, 1);
-  set_array_buffer(compute_encoder, c, 2);
-  set_array_buffer(compute_encoder, out, 3);
+  compute_encoder.set_input_array(a, 0);
+  compute_encoder.set_input_array(b, 1);
+  compute_encoder.set_input_array(c, 2);
+  compute_encoder.set_output_array(out, 3);
 
   if (topt == TernaryOpType::General) {
     auto ndim = shape.size();
@@ -339,11 +339,11 @@ void unary_op(
   }
   MTL::Size group_dims = MTL::Size(thread_group_size, 1, 1);
 
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
-  set_array_buffer(
-      compute_encoder, in.data_shared_ptr() == nullptr ? out : in, 0);
-  set_array_buffer(compute_encoder, out, 1);
+  compute_encoder.set_input_array(
+      in.data_shared_ptr() == nullptr ? out : in, 0);
+  compute_encoder.set_output_array(out, 1);
   if (!contig) {
     compute_encoder->setBytes(in.shape().data(), in.ndim() * sizeof(int), 2);
     compute_encoder->setBytes(
@@ -365,7 +365,7 @@ void Add::eval_gpu(const std::vector<array>& inputs, array& out) {
 }
 
 template <typename T>
-void arange_set_scalars(T start, T next, MTL::ComputeCommandEncoder* enc) {
+void arange_set_scalars(T start, T next, CommandEncoder& enc) {
   enc->setBytes(&start, sizeof(T), 0);
   T step = next - start;
   enc->setBytes(&step, sizeof(T), 1);
@@ -384,7 +384,7 @@ void Arange::eval_gpu(const std::vector<array>& inputs, array& out) {
   MTL::Size grid_dims = MTL::Size(nthreads, 1, 1);
   MTL::Size group_dims = MTL::Size(
       std::min(nthreads, kernel->maxTotalThreadsPerThreadgroup()), 1, 1);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
 
   switch (out.dtype()) {
@@ -427,7 +427,7 @@ void Arange::eval_gpu(const std::vector<array>& inputs, array& out) {
       throw std::runtime_error("[Arange::eval_gpu] Does not support complex64");
   }
 
-  set_array_buffer(compute_encoder, out, 2);
+  compute_encoder.set_output_array(out, 2);
   compute_encoder->dispatchThreads(grid_dims, group_dims);
 }
 
@@ -487,7 +487,7 @@ void ArgReduce::eval_gpu(const std::vector<array>& inputs, array& out) {
   // ArgReduce
   int simd_size = 32;
   int n_reads = 4;
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   {
     auto kernel = d.get_kernel(op_name + type_to_name(in));
     NS::UInteger thread_group_size = std::min(
@@ -502,8 +502,8 @@ void ArgReduce::eval_gpu(const std::vector<array>& inputs, array& out) {
     MTL::Size grid_dims = MTL::Size(n_threads, 1, 1);
     MTL::Size group_dims = MTL::Size(thread_group_size, 1, 1);
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     if (ndim == 0) {
       // Pass place holders so metal doesn't complain
       int shape_ = 0;
@@ -552,6 +552,9 @@ void Concatenate::eval_gpu(const std::vector<array>& inputs, array& out) {
   flags.row_contiguous = false;
   flags.col_contiguous = false;
   flags.contiguous = false;
+  auto& d = metal::device(stream().device);
+  auto& compute_encoder = d.get_command_encoder(stream().index);
+  auto concurrent_ctx = compute_encoder.start_concurrent();
   for (int i = 0; i < inputs.size(); i++) {
     array out_slice(inputs[i].shape(), out.dtype(), nullptr, {});
     size_t data_offset = strides[axis_] * sizes[i];
@@ -615,6 +618,10 @@ void Exp::eval_gpu(const std::vector<array>& inputs, array& out) {
   unary_op(inputs, out, "exp");
 }
 
+void Expm1::eval_gpu(const std::vector<array>& inputs, array& out) {
+  unary_op(inputs, out, "expm1");
+}
+
 void Full::eval_gpu(const std::vector<array>& inputs, array& out) {
   auto in = inputs[0];
   CopyType ctype;
@@ -787,10 +794,10 @@ void RandomBits::eval_gpu(const std::vector<array>& inputs, array& out) {
   MTL::Size grid_dims = MTL::Size(num_keys, half_size + odd, 1);
   NS::UInteger thread_group_size = kernel->maxTotalThreadsPerThreadgroup();
   MTL::Size group_dims = MTL::Size(thread_group_size, 1, 1);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   compute_encoder->setComputePipelineState(kernel);
-  set_array_buffer(compute_encoder, keys, 0);
-  set_array_buffer(compute_encoder, out, 1);
+  compute_encoder.set_input_array(keys, 0);
+  compute_encoder.set_output_array(out, 1);
   compute_encoder->setBytes(&odd, sizeof(bool), 2);
   compute_encoder->setBytes(&bytes_per_key, sizeof(size_t), 3);
 
@@ -822,7 +829,7 @@ void Reshape::eval_gpu(const std::vector<array>& inputs, array& out) {
 void Round::eval_gpu(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
   const auto& in = inputs[0];
-  if (not is_integral(in.dtype())) {
+  if (issubdtype(in.dtype(), inexact)) {
     unary_op(inputs, out, "round");
   } else {
     // No-op integer types

mlx/backend/metal/quantized.cpp

@@ -48,7 +48,7 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
             << bits_ << "_fast";
 
       // Encode and dispatch kernel
-      auto compute_encoder = d.get_command_encoder(s.index);
+      auto& compute_encoder = d.get_command_encoder(s.index);
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
@@ -57,11 +57,11 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
       MTL::Size group_dims = MTL::Size(bd, 2, 1);
       MTL::Size grid_dims = MTL::Size(1, O / bo, B);
 
-      set_array_buffer(compute_encoder, w, 0);
-      set_array_buffer(compute_encoder, scales, 1);
-      set_array_buffer(compute_encoder, biases, 2);
-      set_array_buffer(compute_encoder, x, 3);
-      set_array_buffer(compute_encoder, out, 4);
+      compute_encoder.set_input_array(w, 0);
+      compute_encoder.set_input_array(scales, 1);
+      compute_encoder.set_input_array(biases, 2);
+      compute_encoder.set_input_array(x, 3);
+      compute_encoder.set_output_array(out, 4);
       compute_encoder->setBytes(&D, sizeof(int), 5);
       compute_encoder->setBytes(&O, sizeof(int), 6);
 
@@ -75,7 +75,7 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
             << bits_;
 
       // Encode and dispatch kernel
-      auto compute_encoder = d.get_command_encoder(s.index);
+      auto& compute_encoder = d.get_command_encoder(s.index);
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
@@ -84,11 +84,11 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
       MTL::Size group_dims = MTL::Size(bd, 2, 1);
       MTL::Size grid_dims = MTL::Size(1, (O + bo - 1) / bo, B);
 
-      set_array_buffer(compute_encoder, w, 0);
-      set_array_buffer(compute_encoder, scales, 1);
-      set_array_buffer(compute_encoder, biases, 2);
-      set_array_buffer(compute_encoder, x, 3);
-      set_array_buffer(compute_encoder, out, 4);
+      compute_encoder.set_input_array(w, 0);
+      compute_encoder.set_input_array(scales, 1);
+      compute_encoder.set_input_array(biases, 2);
+      compute_encoder.set_input_array(x, 3);
+      compute_encoder.set_output_array(out, 4);
       compute_encoder->setBytes(&D, sizeof(int), 5);
       compute_encoder->setBytes(&O, sizeof(int), 6);
 
@@ -102,7 +102,7 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
             << bits_ << "_alN_" << std::boolalpha << ((O % 32) == 0);
 
       // Encode and dispatch kernel
-      auto compute_encoder = d.get_command_encoder(s.index);
+      auto& compute_encoder = d.get_command_encoder(s.index);
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
@@ -114,11 +114,11 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
       MTL::Size group_dims = MTL::Size(32, wn, wm);
       MTL::Size grid_dims = MTL::Size((O + bn - 1) / bn, (B + bm - 1) / bm, 1);
 
-      set_array_buffer(compute_encoder, x, 0);
-      set_array_buffer(compute_encoder, w, 1);
-      set_array_buffer(compute_encoder, scales, 2);
-      set_array_buffer(compute_encoder, biases, 3);
-      set_array_buffer(compute_encoder, out, 4);
+      compute_encoder.set_input_array(x, 0);
+      compute_encoder.set_input_array(w, 1);
+      compute_encoder.set_input_array(scales, 2);
+      compute_encoder.set_input_array(biases, 3);
+      compute_encoder.set_output_array(out, 4);
       compute_encoder->setBytes(&B, sizeof(int), 5);
       compute_encoder->setBytes(&O, sizeof(int), 6);
       compute_encoder->setBytes(&D, sizeof(int), 7);
@@ -133,20 +133,20 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
             << bits_;
 
       // Encode and dispatch kernel
-      auto compute_encoder = d.get_command_encoder(s.index);
+      auto& compute_encoder = d.get_command_encoder(s.index);
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
-      int bo = std::min(32, O);
+      int bo = 8;
       int bd = 32;
       MTL::Size group_dims = MTL::Size(bd, bo, 1);
       MTL::Size grid_dims = MTL::Size(1, (O + bo - 1) / bo, B);
 
-      set_array_buffer(compute_encoder, x, 0);
-      set_array_buffer(compute_encoder, w, 1);
-      set_array_buffer(compute_encoder, scales, 2);
-      set_array_buffer(compute_encoder, biases, 3);
-      set_array_buffer(compute_encoder, out, 4);
+      compute_encoder.set_input_array(x, 0);
+      compute_encoder.set_input_array(w, 1);
+      compute_encoder.set_input_array(scales, 2);
+      compute_encoder.set_input_array(biases, 3);
+      compute_encoder.set_output_array(out, 4);
       compute_encoder->setBytes(&D, sizeof(int), 5);
       compute_encoder->setBytes(&O, sizeof(int), 6);
 
@@ -160,7 +160,7 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
             << bits_;
 
       // Encode and dispatch kernel
-      auto compute_encoder = d.get_command_encoder(s.index);
+      auto& compute_encoder = d.get_command_encoder(s.index);
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
@@ -179,11 +179,11 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
         throw std::runtime_error(msg.str());
       }
 
-      set_array_buffer(compute_encoder, x, 0);
-      set_array_buffer(compute_encoder, w, 1);
-      set_array_buffer(compute_encoder, scales, 2);
-      set_array_buffer(compute_encoder, biases, 3);
-      set_array_buffer(compute_encoder, out, 4);
+      compute_encoder.set_input_array(x, 0);
+      compute_encoder.set_input_array(w, 1);
+      compute_encoder.set_input_array(scales, 2);
+      compute_encoder.set_input_array(biases, 3);
+      compute_encoder.set_output_array(out, 4);
       compute_encoder->setBytes(&B, sizeof(int), 5);
       compute_encoder->setBytes(&O, sizeof(int), 6);
       compute_encoder->setBytes(&D, sizeof(int), 7);

mlx/backend/metal/reduce.cpp

@@ -4,10 +4,10 @@
 #include <cassert>
 #include <sstream>
 
-#include "mlx/backend/common/reduce.h"
 #include "mlx/backend/metal/copy.h"
 #include "mlx/backend/metal/device.h"
 #include "mlx/backend/metal/kernels/defines.h"
+#include "mlx/backend/metal/reduce.h"
 #include "mlx/backend/metal/utils.h"
 #include "mlx/primitives.h"
 #include "mlx/utils.h"
@@ -18,8 +18,6 @@ namespace mlx::core {
 // Case wise reduce dispatch
 //////////////////////////////////////////////////////////////////////
 
-namespace {
-
 inline auto safe_div(size_t n, size_t m) {
   return m == 0 ? 0 : (n + m - 1) / m;
 }
@@ -37,7 +35,7 @@ void all_reduce_dispatch(
     const array& in,
     array& out,
     const std::string& op_name,
-    MTL::ComputeCommandEncoder* compute_encoder,
+    CommandEncoder& compute_encoder,
     metal::Device& d,
     const Stream& s) {
   Dtype out_dtype = out.dtype();
@@ -73,8 +71,8 @@ void all_reduce_dispatch(
 
   // Encode buffers and dispatch
   if (is_out_64b_int == false || n_thread_groups == 1) {
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&in_size, sizeof(size_t), 2);
     compute_encoder->dispatchThreads(grid_dims, group_dims);
 
@@ -87,14 +85,14 @@ void all_reduce_dispatch(
     std::vector<array> intermediates = {intermediate};
 
     // First dispatch
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, intermediate, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(intermediate, 1);
     compute_encoder->setBytes(&in_size, sizeof(size_t), 2);
     compute_encoder->dispatchThreads(grid_dims, group_dims);
 
     // Second pass to reduce intermediate reduction results written to DRAM
-    set_array_buffer(compute_encoder, intermediate, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(intermediate, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&intermediate_size, sizeof(size_t), 2);
 
     mod_in_size = (intermediate_size + n_reads - 1) / n_reads;
@@ -125,7 +123,7 @@ void row_reduce_general_dispatch(
     const std::string& op_name,
     const ReductionPlan& plan,
     const std::vector<int>& axes,
-    MTL::ComputeCommandEncoder* compute_encoder,
+    CommandEncoder& compute_encoder,
     metal::Device& d,
     const Stream& s) {
   Dtype out_dtype = out.dtype();
@@ -210,8 +208,8 @@ void row_reduce_general_dispatch(
   // Dispatch kernel
   if (!is_out_64b_int || non_row_reductions == 1) {
     // Set the arguments for the kernel
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 3);
     compute_encoder->setBytes(&non_row_reductions, sizeof(size_t), 4);
@@ -232,8 +230,8 @@ void row_reduce_general_dispatch(
     std::vector<array> intermediates = {intermediate};
 
     // Set the arguments for the kernel
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, intermediate, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(intermediate, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 3);
     compute_encoder->setBytes(&non_row_reductions, sizeof(size_t), 4);
@@ -260,8 +258,8 @@ void row_reduce_general_dispatch(
     ndim = new_shape.size();
 
     // Set the arguments for the kernel
-    set_array_buffer(compute_encoder, intermediate, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(intermediate, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 3);
     compute_encoder->setBytes(&non_row_reductions, sizeof(size_t), 4);
@@ -303,7 +301,7 @@ void strided_reduce_general_dispatch(
     const std::string& op_name,
     const ReductionPlan& plan,
     const std::vector<int>& axes,
-    MTL::ComputeCommandEncoder* compute_encoder,
+    CommandEncoder& compute_encoder,
     metal::Device& d,
     const Stream& s) {
   Dtype out_dtype = out.dtype();
@@ -351,8 +349,8 @@ void strided_reduce_general_dispatch(
     }
 
     // Encode arrays
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&reduction_stride, sizeof(size_t), 3);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 4);
@@ -417,8 +415,8 @@ void strided_reduce_general_dispatch(
 
   if (is_out_64b_int == false) {
     // Set the arguments for the kernel
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&reduction_stride, sizeof(size_t), 3);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 4);
@@ -452,8 +450,8 @@ void strided_reduce_general_dispatch(
     std::vector<array> intermediates = {intermediate};
 
     // Set the arguments for the kernel
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, intermediate, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(intermediate, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&reduction_stride, sizeof(size_t), 3);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 4);
@@ -496,8 +494,8 @@ void strided_reduce_general_dispatch(
         "row_reduce_general_no_atomics_" + op_name +
         type_to_name(intermediate));
     compute_encoder->setComputePipelineState(row_reduce_kernel);
-    set_array_buffer(compute_encoder, intermediate, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(intermediate, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 2);
     compute_encoder->setBytes(&out_size, sizeof(size_t), 3);
     compute_encoder->setBytes(&reduction_size, sizeof(size_t), 4);
@@ -534,8 +532,6 @@ void strided_reduce_general_dispatch(
   }
 }
 
-} // namespace
-
 //////////////////////////////////////////////////////////////////////
 // Main reduce dispatch
 //////////////////////////////////////////////////////////////////////
@@ -577,7 +573,7 @@ void Reduce::eval_gpu(const std::vector<array>& inputs, array& out) {
   // Initialize output
   auto& s = stream();
   auto& d = metal::device(s.device);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   {
     auto kernel = d.get_kernel("i" + op_name + type_to_name(out));
     size_t nthreads = out.size();
@@ -588,7 +584,7 @@ void Reduce::eval_gpu(const std::vector<array>& inputs, array& out) {
     }
     MTL::Size group_dims = MTL::Size(thread_group_size, 1, 1);
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(compute_encoder, out, 0);
+    compute_encoder.set_output_array(out, 0);
     compute_encoder->dispatchThreads(grid_dims, group_dims);
   }
 

mlx/backend/metal/reduce.h

@@ -0,0 +1,41 @@
+// Copyright @ 2023 - 2024 Apple Inc.
+
+#pragma once
+
+#include "mlx/backend/common/reduce.h"
+#include "mlx/backend/metal/device.h"
+#include "mlx/stream.h"
+
+namespace mlx::core {
+
+using metal::CommandEncoder;
+
+void all_reduce_dispatch(
+    const array& in,
+    array& out,
+    const std::string& op_name,
+    CommandEncoder& compute_encoder,
+    metal::Device& d,
+    const Stream& s);
+
+void row_reduce_general_dispatch(
+    const array& in,
+    array& out,
+    const std::string& op_name,
+    const ReductionPlan& plan,
+    const std::vector<int>& axes,
+    CommandEncoder& compute_encoder,
+    metal::Device& d,
+    const Stream& s);
+
+void strided_reduce_general_dispatch(
+    const array& in,
+    array& out,
+    const std::string& op_name,
+    const ReductionPlan& plan,
+    const std::vector<int>& axes,
+    CommandEncoder& compute_encoder,
+    metal::Device& d,
+    const Stream& s);
+
+} // namespace mlx::core

mlx/backend/metal/rope.cpp

@@ -63,14 +63,15 @@ void RoPE::eval_gpu(
   out_strides[2] = out.strides()[ndim - 1];
 
   std::ostringstream kname;
-  kname << "rope_" << (traditional_ ? "traditional_" : "") << type_to_name(in);
+  kname << "rope_" << (forward_ ? "" : "vjp_")
+        << (traditional_ ? "traditional_" : "") << type_to_name(in);
   auto kernel = d.get_kernel(kname.str());
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
 
   float base = std::log2(base_);
   compute_encoder->setComputePipelineState(kernel);
-  set_array_buffer(compute_encoder, donated ? out : in, 0);
-  set_array_buffer(compute_encoder, out, 1);
+  compute_encoder.set_input_array(donated ? out : in, 0);
+  compute_encoder.set_output_array(out, 1);
   compute_encoder->setBytes(&strides, 3 * sizeof(size_t), 2);
   compute_encoder->setBytes(&out_strides, 3 * sizeof(size_t), 3);
   compute_encoder->setBytes(&offset_, sizeof(int), 4);

mlx/backend/metal/scaled_dot_product_attention.cpp

@@ -71,7 +71,7 @@ void sdpa_metal(
 
   std::string kname_suffix = kname_suffix_tile_size + kname_suffix_nsimdgroups;
   kname_partials << kname_suffix;
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname_partials.str());
   compute_encoder->setComputePipelineState(kernel);
 
@@ -87,15 +87,15 @@ void sdpa_metal(
   MLXScaledDotProductAttentionParams params{
       query_sequence_length, n_q_heads, n_kv_heads, n_tiles, alpha};
 
-  set_array_buffer(compute_encoder, q, 0);
-  set_array_buffer(compute_encoder, k, 1);
-  set_array_buffer(compute_encoder, v, 2);
+  compute_encoder.set_input_array(q, 0);
+  compute_encoder.set_input_array(k, 1);
+  compute_encoder.set_input_array(v, 2);
   compute_encoder->setBytes(&KV_sequence_length, sizeof(KV_sequence_length), 3);
   compute_encoder->setBytes(
       &params, sizeof(MLXScaledDotProductAttentionParams), 4);
-  set_array_buffer(compute_encoder, o_partial, 5);
-  set_array_buffer(compute_encoder, p_lse, 6);
-  set_array_buffer(compute_encoder, p_rowmaxes, 7);
+  compute_encoder.set_input_array(o_partial, 5);
+  compute_encoder.set_input_array(p_lse, 6);
+  compute_encoder.set_input_array(p_rowmaxes, 7);
 
   constexpr const uint tgroupMemorySize = 32768;
   compute_encoder->setThreadgroupMemoryLength(tgroupMemorySize, 0);
@@ -104,12 +104,12 @@ void sdpa_metal(
   {
     auto kernel_accum = d.get_kernel(kname_reduce.str());
     compute_encoder->setComputePipelineState(kernel_accum);
-    set_array_buffer(compute_encoder, o_partial, 0);
-    set_array_buffer(compute_encoder, p_lse, 1);
-    set_array_buffer(compute_encoder, p_rowmaxes, 2);
+    compute_encoder.set_input_array(o_partial, 0);
+    compute_encoder.set_input_array(p_lse, 1);
+    compute_encoder.set_input_array(p_rowmaxes, 2);
     compute_encoder->setBytes(
         &params, sizeof(MLXScaledDotProductAttentionParams), 3);
-    set_array_buffer(compute_encoder, out, 4);
+    compute_encoder.set_output_array(out, 4);
 
     MTL::Size grid_dims_reduce = MTL::Size(heads, 1, batch);
     MTL::Size group_dims_reduce = MTL::Size(128, 1, 1);
@@ -127,7 +127,7 @@ void ScaledDotProductAttention::eval_gpu(
     const std::vector<array>& inputs,
     array& out) {
   assert(inputs.size() >= 3);
-  if (!is_floating_point(out.dtype())) {
+  if (!issubdtype(out.dtype(), floating)) {
     throw std::runtime_error(
         "[ScaledDotProductAttention] Does not yet support non-floating point types.");
   }

mlx/backend/metal/scan.cpp

@@ -52,10 +52,10 @@ void Scan::eval_gpu(const std::vector<array>& inputs, array& out) {
     kname << type_to_name(in) << "_" << type_to_name(out);
 
     auto kernel = d.get_kernel(kname.str());
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     size_t size = in.shape(axis_);
     compute_encoder->setBytes(&size, sizeof(size_t), 2);
 
@@ -101,10 +101,10 @@ void Scan::eval_gpu(const std::vector<array>& inputs, array& out) {
     kname << type_to_name(in) << "_" << type_to_name(out);
 
     auto kernel = d.get_kernel(kname.str());
-    auto compute_encoder = d.get_command_encoder(s.index);
+    auto& compute_encoder = d.get_command_encoder(s.index);
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(out, 1);
     size_t size = in.shape(axis_);
     size_t stride = in.strides()[axis_];
     compute_encoder->setBytes(&size, sizeof(size_t), 2);

mlx/backend/metal/softmax.cpp

@@ -12,7 +12,7 @@ namespace mlx::core {
 
 void Softmax::eval_gpu(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() == 1);
-  if (!is_floating_point(out.dtype())) {
+  if (!issubdtype(out.dtype(), floating)) {
     throw std::runtime_error(
         "[softmax] Does not support non-floating point types.");
   }
@@ -21,7 +21,7 @@ void Softmax::eval_gpu(const std::vector<array>& inputs, array& out) {
 
   // Make sure that the last dimension is contiguous
   std::vector<array> copies;
-  auto check_input = [&copies, &s](const array& x) {
+  auto check_input = [&copies, &s](const array& x) -> const array& {
     bool no_copy = x.strides()[x.ndim() - 1] == 1;
     if (x.ndim() > 1) {
       auto s = x.strides()[x.ndim() - 2];
@@ -30,10 +30,9 @@ void Softmax::eval_gpu(const std::vector<array>& inputs, array& out) {
     if (no_copy) {
       return x;
     } else {
-      array x_copy(x.shape(), x.dtype(), nullptr, {});
-      copy_gpu(x, x_copy, CopyType::General, s);
-      copies.push_back(x_copy);
-      return x_copy;
+      copies.push_back(array(x.shape(), x.dtype(), nullptr, {}));
+      copy_gpu(x, copies.back(), CopyType::General, s);
+      return copies.back();
     }
   };
   const array& in = check_input(inputs[0]);
@@ -57,8 +56,11 @@ void Softmax::eval_gpu(const std::vector<array>& inputs, array& out) {
   if (axis_size > looped_limit) {
     op_name += "looped_";
   }
+  if (in.dtype() != float32 && precise_) {
+    op_name += "precise_";
+  }
   op_name += type_to_name(out);
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   {
     auto kernel = d.get_kernel(op_name);
 
@@ -79,13 +81,10 @@ void Softmax::eval_gpu(const std::vector<array>& inputs, array& out) {
     }
 
     compute_encoder->setComputePipelineState(kernel);
-    set_array_buffer(
-        compute_encoder, in.data_shared_ptr() == nullptr ? out : in, 0);
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, out, 1);
+    compute_encoder.set_input_array(
+        in.data_shared_ptr() == nullptr ? out : in, 0);
+    compute_encoder.set_output_array(out, 1);
     compute_encoder->setBytes(&axis_size, sizeof(int), 2);
-    compute_encoder->setThreadgroupMemoryLength(simd_size * in.itemsize(), 0);
-    compute_encoder->setThreadgroupMemoryLength(simd_size * in.itemsize(), 1);
     compute_encoder->dispatchThreads(grid_dims, group_dims);
   }
   d.get_command_buffer(s.index)->addCompletedHandler(

mlx/backend/metal/sort.cpp

@@ -1,4 +1,4 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #include <algorithm>
 
@@ -57,13 +57,13 @@ void single_block_sort(
   }
 
   // Prepare command encoder
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
   auto kernel = d.get_kernel(kname.str());
   compute_encoder->setComputePipelineState(kernel);
 
   // Set inputs
-  set_array_buffer(compute_encoder, in, 0);
-  set_array_buffer(compute_encoder, out, 1);
+  compute_encoder.set_input_array(in, 0);
+  compute_encoder.set_output_array(out, 1);
   compute_encoder->setBytes(&size_sorted_axis, sizeof(int), 2);
   compute_encoder->setBytes(&stride_sorted_axis, sizeof(int), 3);
 
@@ -102,6 +102,11 @@ void multi_block_sort(
 
   int nc_dim = nc_shape.size();
 
+  if (nc_dim == 0) {
+    nc_shape = {0};
+    nc_str = {1};
+  }
+
   int size_sorted_axis = in.shape(axis);
   int stride_sorted_axis = in.strides()[axis];
 
@@ -126,7 +131,7 @@ void multi_block_sort(
       dev_vals_0, dev_vals_1, dev_idxs_0, dev_idxs_1, block_partitions};
 
   // Prepare command encoder
-  auto compute_encoder = d.get_command_encoder(s.index);
+  auto& compute_encoder = d.get_command_encoder(s.index);
 
   // Do blockwise sort
   {
@@ -137,14 +142,15 @@ void multi_block_sort(
     auto kernel = d.get_kernel(kname.str());
     compute_encoder->setComputePipelineState(kernel);
 
-    set_array_buffer(compute_encoder, in, 0);
-    set_array_buffer(compute_encoder, dev_vals_0, 1);
-    set_array_buffer(compute_encoder, dev_idxs_0, 2);
+    compute_encoder.set_input_array(in, 0);
+    compute_encoder.set_output_array(dev_vals_0, 1);
+    compute_encoder.set_output_array(dev_idxs_0, 2);
     compute_encoder->setBytes(&size_sorted_axis, sizeof(int), 3);
     compute_encoder->setBytes(&stride_sorted_axis, sizeof(int), 4);
     compute_encoder->setBytes(&nc_dim, sizeof(int), 5);
-    compute_encoder->setBytes(nc_shape.data(), nc_dim * sizeof(int), 6);
-    compute_encoder->setBytes(nc_str.data(), nc_dim * sizeof(size_t), 7);
+    compute_encoder->setBytes(
+        nc_shape.data(), nc_shape.size() * sizeof(int), 6);
+    compute_encoder->setBytes(nc_str.data(), nc_str.size() * sizeof(size_t), 7);
 
     MTL::Size group_dims = MTL::Size(bn, 1, 1);
     MTL::Size grid_dims = MTL::Size(n_blocks, n_rows, 1);
@@ -158,7 +164,8 @@ void multi_block_sort(
   array dev_idxs_in = dev_idxs_0;
   array dev_vals_out = dev_vals_1;
   array dev_idxs_out = dev_idxs_1;
-  for (int merge_tiles = 2; merge_tiles <= n_blocks; merge_tiles *= 2) {
+
+  for (int merge_tiles = 2; (merge_tiles / 2) < n_blocks; merge_tiles *= 2) {
     dev_vals_in = ping ? dev_vals_1 : dev_vals_0;
     dev_idxs_in = ping ? dev_idxs_1 : dev_idxs_0;
     dev_vals_out = ping ? dev_vals_0 : dev_vals_1;
@@ -174,9 +181,9 @@ void multi_block_sort(
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
-      set_array_buffer(compute_encoder, block_partitions, 0);
-      set_array_buffer(compute_encoder, dev_vals_in, 1);
-      set_array_buffer(compute_encoder, dev_idxs_in, 2);
+      compute_encoder.set_output_array(block_partitions, 0);
+      compute_encoder.set_input_array(dev_vals_in, 1);
+      compute_encoder.set_input_array(dev_idxs_in, 2);
       compute_encoder->setBytes(&size_sorted_axis, sizeof(int), 3);
       compute_encoder->setBytes(&merge_tiles, sizeof(int), 4);
 
@@ -195,11 +202,11 @@ void multi_block_sort(
       auto kernel = d.get_kernel(kname.str());
       compute_encoder->setComputePipelineState(kernel);
 
-      set_array_buffer(compute_encoder, block_partitions, 0);
-      set_array_buffer(compute_encoder, dev_vals_in, 1);
-      set_array_buffer(compute_encoder, dev_idxs_in, 2);
-      set_array_buffer(compute_encoder, dev_vals_out, 3);
-      set_array_buffer(compute_encoder, dev_idxs_out, 4);
+      compute_encoder.set_input_array(block_partitions, 0);
+      compute_encoder.set_input_array(dev_vals_in, 1);
+      compute_encoder.set_input_array(dev_idxs_in, 2);
+      compute_encoder.set_output_array(dev_vals_out, 3);
+      compute_encoder.set_output_array(dev_idxs_out, 4);
       compute_encoder->setBytes(&size_sorted_axis, sizeof(int), 5);
       compute_encoder->setBytes(&merge_tiles, sizeof(int), 6);
       compute_encoder->setBytes(&n_blocks, sizeof(int), 7);

mlx/backend/metal/utils.h

@@ -1,37 +1,20 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #pragma once
 
 #include "mlx/array.h"
 #include "mlx/backend/metal/device.h"
+#include "mlx/primitives.h"
 
 namespace mlx::core {
 
 namespace {
 
-inline void
-set_array_buffer(MTL::ComputeCommandEncoder* enc, const array& a, int idx) {
-  auto a_buf = static_cast<const MTL::Buffer*>(a.buffer().ptr());
-  auto offset = a.data<char>() -
-      static_cast<char*>(const_cast<MTL::Buffer*>(a_buf)->contents());
-  enc->setBuffer(a_buf, offset, idx);
-}
-
-inline void set_array_buffer(
-    MTL::ComputeCommandEncoder* enc,
-    const array& a,
-    int64_t offset,
-    int idx) {
-  auto a_buf = static_cast<const MTL::Buffer*>(a.buffer().ptr());
-  auto base_offset = a.data<char>() -
-      static_cast<char*>(const_cast<MTL::Buffer*>(a_buf)->contents());
-  base_offset += offset;
-  enc->setBuffer(a_buf, base_offset, idx);
-}
+using metal::CommandEncoder;
 
 template <typename T>
 inline void set_vector_bytes(
-    MTL::ComputeCommandEncoder* enc,
+    CommandEncoder& enc,
     const std::vector<T>& vec,
     size_t nelems,
     int idx) {
@@ -39,10 +22,8 @@ inline void set_vector_bytes(
 }
 
 template <typename T>
-inline void set_vector_bytes(
-    MTL::ComputeCommandEncoder* enc,
-    const std::vector<T>& vec,
-    int idx) {
+inline void
+set_vector_bytes(CommandEncoder& enc, const std::vector<T>& vec, int idx) {
   return set_vector_bytes(enc, vec, vec.size(), idx);
 }
 
@@ -123,6 +104,32 @@ MTL::Size get_block_dims(int dim0, int dim1, int dim2) {
   return MTL::Size{1ul << pows[0], 1ul << pows[1], 1ul << pows[2]};
 }
 
+inline NS::String* make_string(std::ostringstream& os) {
+  std::string string = os.str();
+  return NS::String::string(string.c_str(), NS::UTF8StringEncoding);
+}
+
+inline void debug_set_stream_queue_label(MTL::CommandQueue* queue, int index) {
+#ifdef MLX_METAL_DEBUG
+  std::ostringstream label;
+  label << "Stream " << index;
+  queue->setLabel(make_string(label));
+#endif
+}
+
+inline void debug_set_primitive_buffer_label(
+    MTL::CommandBuffer* command_buffer,
+    Primitive& primitive) {
+#ifdef MLX_METAL_DEBUG
+  std::ostringstream label;
+  if (auto cbuf_label = command_buffer->label(); cbuf_label) {
+    label << cbuf_label->utf8String();
+  }
+  primitive.print(label);
+  command_buffer->setLabel(make_string(label));
+#endif
+}
+
 } // namespace
 
 } // namespace mlx::core

mlx/backend/no_metal/metal.cpp

@@ -3,6 +3,7 @@
 #include <stdexcept>
 
 #include "mlx/backend/metal/metal.h"
+#include "mlx/backend/metal/metal_impl.h"
 
 namespace mlx::core::metal {
 
@@ -39,5 +40,9 @@ size_t set_memory_limit(size_t, bool) {
 size_t set_cache_limit(size_t) {
   return 0;
 }
+bool start_capture(std::string path) {
+  return false;
+}
+void stop_capture() {}
 
 } // namespace mlx::core::metal

mlx/backend/no_metal/primitives.cpp

@@ -49,6 +49,7 @@ NO_GPU(Equal)
 NO_GPU(Erf)
 NO_GPU(ErfInv)
 NO_GPU(Exp)
+NO_GPU(Expm1)
 NO_GPU(FFT)
 NO_GPU(Floor)
 NO_GPU(Full)
@@ -103,7 +104,9 @@ NO_GPU(Inverse)
 
 namespace fast {
 NO_GPU_MULTI(LayerNorm)
+NO_GPU_MULTI(LayerNormVJP)
 NO_GPU_MULTI(RMSNorm)
+NO_GPU_MULTI(RMSNormVJP)
 NO_GPU_MULTI(RoPE)
 NO_GPU(ScaledDotProductAttention)
 } // namespace fast

mlx/compile.cpp

@@ -32,7 +32,7 @@ bool is_unary(const Primitive& p) {
       typeid(p) == typeid(Sign) || typeid(p) == typeid(Sin) ||
       typeid(p) == typeid(Sinh) || typeid(p) == typeid(Square) ||
       typeid(p) == typeid(Sqrt) || typeid(p) == typeid(Tan) ||
-      typeid(p) == typeid(Tanh));
+      typeid(p) == typeid(Tanh) || typeid(p) == typeid(Expm1));
 }
 
 bool is_binary(const Primitive& p) {
@@ -162,44 +162,51 @@ CompileMode& compile_mode() {
   return compile_mode_;
 }
 
-using CompileFn = std::function<std::vector<array>(const std::vector<array>&)>;
 using ParentsMap =
     std::unordered_map<std::uintptr_t, std::vector<std::pair<array, int>>>;
 
+// Helper like below but only merges the two provided arrays. If the src has
+// siblings then these won't be merged to the dst.
+void merge_one(array& dst, array& src, ParentsMap& parents_map) {
+  auto src_parents = parents_map.find(src.id());
+  if (src_parents == parents_map.end()) {
+    return;
+  }
+  auto& pairs = parents_map[dst.id()];
+  for (auto& parent : src_parents->second) {
+    parent.first.inputs()[parent.second] = dst;
+    pairs.push_back(parent);
+  }
+  // Remove the source from the map to avoid fusing with it again
+  parents_map.erase(src_parents);
+};
+
 // Helper that merges two arrays in the graph by setting the parents of the
-// source to point to the destination
+// source to point to the destination. The arrays are assumed to be coming from
+// equivalent primitives so their siblings are merged as well.
 void merge(array& dst, array& src, ParentsMap& parents_map) {
   // Canonicalize the order of the primitives outputs
   auto sources = src.outputs();
   auto dests = dst.outputs();
   // For each src parent, point it to the corresponding dst
   for (int i = 0; i < sources.size(); ++i) {
-    auto src_parents = parents_map.find(sources[i].id());
-    if (src_parents == parents_map.end()) {
-      continue;
-    }
-    auto& pairs = parents_map[dests[i].id()];
-    for (auto& parent : src_parents->second) {
-      parent.first.inputs()[parent.second] = dests[i];
-      pairs.push_back(parent);
-    }
-    // Remove the source from the map to avoid fusing with it again
-    parents_map.erase(src_parents);
+    merge_one(dests[i], sources[i], parents_map);
   }
 };
 
 template <typename T, typename... U>
-size_t getAddress(std::function<T(U...)> f) {
-  typedef T(fnType)(U...);
-  fnType** fnPointer = f.template target<fnType*>();
-  if (fnPointer == nullptr) {
+std::uintptr_t get_function_address(const std::function<T(U...)>& fun) {
+  using FunType = T (*)(U...);
+  const FunType* fun_ptr = fun.template target<FunType>();
+  if (fun_ptr == nullptr) {
     throw std::invalid_argument(
         "[compile] Cannot compile a non-addressable function.");
   }
-  return (size_t)*fnPointer;
+  return reinterpret_cast<std::uintptr_t>(*fun_ptr);
 }
 
-struct CompilerCache {
+class CompilerCache {
+ public:
   struct CacheEntry {
     std::vector<array> inputs;
     std::vector<array> outputs;
@@ -211,20 +218,20 @@ struct CompilerCache {
   // Returns a reference to a CacheEntry which can be updated
   // by the caller to avoid copying large tapes / inputs / outputs
   CacheEntry& find(
-      size_t fun_id,
+      std::uintptr_t fun_id,
       const std::vector<array>& inputs,
       bool shapeless,
       const std::vector<uint64_t>& constants) {
-    // Try to find the entry
-    auto [entry_it, inserted] = cache_.insert({fun_id, {}});
-    auto& entries = entry_it->second;
-    auto is_match = [shapeless](
-                        const std::vector<array>& in1,
-                        const std::vector<array>& in2) {
+    // Find the cache entries for |fun_id|.
+    std::vector<CacheEntry>& entries = cache_[fun_id];
+    // Compare if 2 arrays have same shape and dtype.
+    auto has_same_shape_and_dtype = [shapeless](
+                                        const std::vector<array>& in1,
+                                        const std::vector<array>& in2) {
       if (in1.size() != in2.size()) {
         return false;
       }
-      for (int i = 0; i < in1.size(); ++i) {
+      for (size_t i = 0; i < in1.size(); ++i) {
         if (in1[i].ndim() != in2[i].ndim()) {
           return false;
         }
@@ -237,14 +244,14 @@ struct CompilerCache {
       }
       return true;
     };
-
     // Loop over entries and check inputs match i.e. shapes and types must be
     // equal. Note this could get really slow if one compiles the same
     // function with many different shapes. May want to store entries in a
     // more easily searchable structure.
-    for (auto& entry : entries) {
+    for (CacheEntry& entry : entries) {
       // Check the inputs match and return if so
-      if (is_match(inputs, entry.inputs) && constants == entry.constants) {
+      if (has_same_shape_and_dtype(inputs, entry.inputs) &&
+          constants == entry.constants) {
         return entry;
       }
     }
@@ -253,7 +260,7 @@ struct CompilerCache {
     return entries.back();
   };
 
-  void erase(size_t fun_id) {
+  void erase(std::uintptr_t fun_id) {
     cache_.erase(fun_id);
   }
 
@@ -263,8 +270,9 @@ struct CompilerCache {
     // initialized before the compiler cache
     allocator::allocator();
   }
+
   friend CompilerCache& compiler_cache();
-  std::unordered_map<size_t, std::vector<CacheEntry>> cache_;
+  std::unordered_map<std::uintptr_t, std::vector<CacheEntry>> cache_;
 };
 
 CompilerCache& compiler_cache() {
@@ -523,9 +531,14 @@ void compile_fuse(
     //  - Collect inputs to the new compiled primitive
     //  - Add fusable primitives to a tape in the correct order
 
-    std::function<void(const array&, int, const Stream&)> recurse;
+    std::function<void(
+        const array&, int, const Stream&, const std::vector<int>&)>
+        recurse;
     std::unordered_set<uintptr_t> cache;
-    recurse = [&](const array& a, int depth, const Stream& s) {
+    recurse = [&](const array& a,
+                  int depth,
+                  const Stream& s,
+                  const std::vector<int>& shape) {
       if (cache.find(a.id()) != cache.end()) {
         return;
       }
@@ -535,8 +548,10 @@ void compile_fuse(
       // - Constant input
       // - Stream mismatch
       // - Non fusable primitive
+      // - Is global output but has a different shape
       if (depth >= max_compile_depth || !a.has_primitive() ||
-          a.primitive().stream() != s || !is_fusable(a.primitive())) {
+          a.primitive().stream() != s || !is_fusable(a.primitive()) ||
+          (output_map.find(a.id()) != output_map.end() && a.shape() != shape)) {
         return;
       }
 
@@ -563,13 +578,13 @@ void compile_fuse(
       cache.insert({a.id()});
 
       for (auto& in : a.inputs()) {
-        recurse(in, depth + 1, s);
+        recurse(in, depth + 1, s, shape);
       }
     };
 
     if (arr.has_primitive()) {
       Stream s = arr.primitive().stream();
-      recurse(arr, 0, s);
+      recurse(arr, 0, s, arr.shape());
     }
 
     // Not worth fusing a single primitive
@@ -633,6 +648,10 @@ void compile_fuse(
     std::vector<std::vector<int>> shapes;
     std::vector<Dtype> types;
     for (auto& o : old_outputs) {
+      if (o.shape() != old_outputs.back().shape()) {
+        throw std::runtime_error(
+            "[compile] Compilation failed. Tried to fuse operations with different output shapes");
+      }
       shapes.push_back(o.shape());
       types.push_back(o.dtype());
     }
@@ -675,7 +694,7 @@ void compile_fuse(
     // - Update outputs parents to point to compiled outputs
     // - Update any overall graph outputs to be compiled outputs
     for (int o = 0; o < old_outputs.size(); ++o) {
-      merge(compiled_outputs[o], old_outputs[o], parents_map);
+      merge_one(compiled_outputs[o], old_outputs[o], parents_map);
       if (auto it = output_map.find(old_outputs[o].id());
           it != output_map.end()) {
         it->second = compiled_outputs[o];
@@ -774,7 +793,7 @@ void compile_validate_shapeless(const std::vector<array>& tape) {
 
 std::function<std::vector<array>(const std::vector<array>&)> compile(
     const std::function<std::vector<array>(const std::vector<array>&)>& fun,
-    size_t fun_id,
+    std::uintptr_t fun_id,
     bool shapeless /* = false */,
     std::vector<uint64_t> constants /* = {} */) {
   if (compile_mode() == CompileMode::disabled ||
@@ -833,7 +852,7 @@ std::function<std::vector<array>(const std::vector<array>&)> compile(
   };
 }
 
-void compile_erase(size_t fun_id) {
+void compile_erase(std::uintptr_t fun_id) {
   detail::compiler_cache().erase(fun_id);
 }
 
@@ -845,7 +864,7 @@ std::function<std::vector<array>(const std::vector<array>&)> compile(
   if (detail::compile_mode() == CompileMode::disabled) {
     return fun;
   }
-  auto fun_id = detail::getAddress(fun);
+  auto fun_id = detail::get_function_address(fun);
   return detail::compile(fun, fun_id, shapeless);
 }
 

mlx/dtype.cpp

@@ -1,4 +1,4 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #include <cstdint>
 #include <sstream>
@@ -12,6 +12,7 @@ namespace mlx::core {
 namespace {
 
 constexpr int num_types = 13;
+constexpr int num_cats = 8;
 
 constexpr Dtype::Kind type_kinds[num_types] = {
     Dtype::Kind::b, // bool_,
@@ -49,18 +50,37 @@ constexpr Dtype type_rules[num_types][num_types] = {
   {complex64, complex64, complex64, complex64, complex64, complex64, complex64, complex64, complex64, complex64, complex64, complex64, complex64}, // complex64
 };
 
+
+constexpr bool subcategory_to_category[num_cats][num_cats] = {
+// complexfloating floating inexact signedinteger unsignedinteger integer number generic
+  {true,           false,   true,   false,        false,          false,  true,  true}, // complexfloating
+  {false,          true,    true,   false,        false,          false,  true,  true}, // floating
+  {false,          false,   true,   false,        false,          false,  true,  true}, // inexact
+  {false,          false,   false,  true,         false,          true,   true,  true}, // signedinteger
+  {false,          false,   false,  false,        true,           true,   true,  true}, // unsignedinteger
+  {false,          false,   false,  false,        false,          true,   true,  true}, // integer
+  {false,          false,   false,  false,        false,          false,  true,  true}, // number
+  {false,          false,   false,  false,        false,          false,  false, true}, // generic
+};
+
+constexpr Dtype::Category type_to_category[num_types] = {
+    Dtype::Category::generic, // bool_,
+    Dtype::Category::unsignedinteger, // uint8,
+    Dtype::Category::unsignedinteger, // uint16,
+    Dtype::Category::unsignedinteger, // uint32,
+    Dtype::Category::unsignedinteger, // uint64,
+    Dtype::Category::signedinteger, // int8,
+    Dtype::Category::signedinteger, // int16,
+    Dtype::Category::signedinteger, // int32,
+    Dtype::Category::signedinteger, // int64,
+    Dtype::Category::floating, // float16,
+    Dtype::Category::floating, // float32,
+    Dtype::Category::floating, // bfloat16,
+    Dtype::Category::complexfloating, // complex64,
+};
+
 // clang-format on
 
-inline bool is_big_endian() {
-  union ByteOrder {
-    int32_t i;
-    uint8_t c[4];
-  };
-  ByteOrder b = {0x01234567};
-
-  return b.c[0] == 0x01;
-}
-
 } // namespace
 
 Dtype promote_types(const Dtype& t1, const Dtype& t2) {
@@ -141,6 +161,23 @@ TypeToDtype<complex64_t>::operator Dtype() {
   return complex64;
 }
 
+bool issubdtype(const Dtype& a, const Dtype& b) {
+  return a == b;
+}
+
+bool issubdtype(const Dtype::Category& cat, const Dtype& type) {
+  return false;
+}
+
+bool issubdtype(const Dtype& type, const Dtype::Category& cat) {
+  return issubdtype(type_to_category[static_cast<uint32_t>(type.val)], cat);
+}
+
+bool issubdtype(const Dtype::Category& a, const Dtype::Category& b) {
+  return subcategory_to_category[static_cast<uint32_t>(a)]
+                                [static_cast<uint32_t>(b)];
+}
+
 // Array protocol typestring for Dtype
 std::string dtype_to_array_protocol(const Dtype& t) {
   std::ostringstream r;
@@ -153,9 +190,9 @@ std::string dtype_to_array_protocol(const Dtype& t) {
 }
 
 // Dtype from array protocol type string
-Dtype dtype_from_array_protocol(const std::string& t) {
+Dtype dtype_from_array_protocol(std::string_view t) {
   if (t.length() == 2 || t.length() == 3) {
-    std::string r = t.length() == 3 ? t.substr(1, 2) : t;
+    std::string_view r = t.length() == 3 ? t.substr(1, 2) : t;
 
     if (r == "V2") {
       return bfloat16;
@@ -201,7 +238,7 @@ Dtype dtype_from_array_protocol(const std::string& t) {
   }
 
   throw std::invalid_argument(
-      "[from_str] Invalid array protocol type-string: " + t);
+      "[from_str] Invalid array protocol type-string: " + std::string(t));
 }
 
 } // namespace mlx::core

mlx/dtype.h

@@ -1,4 +1,4 @@
-// Copyright © 2023 Apple Inc.
+// Copyright © 2023-2024 Apple Inc.
 
 #pragma once
 
@@ -38,6 +38,17 @@ struct Dtype {
     V, /* void - used for brain float */
   };
 
+  enum class Category {
+    complexfloating,
+    floating,
+    inexact,
+    signedinteger,
+    unsignedinteger,
+    integer,
+    number,
+    generic
+  };
+
   Val val;
   const uint8_t size;
   constexpr explicit Dtype(Val val, uint8_t size) : val(val), size(size){};
@@ -63,6 +74,22 @@ inline constexpr Dtype float32{Dtype::Val::float32, sizeof(float)};
 inline constexpr Dtype bfloat16{Dtype::Val::bfloat16, sizeof(uint16_t)};
 inline constexpr Dtype complex64{Dtype::Val::complex64, sizeof(complex64_t)};
 
+inline constexpr Dtype::Category complexfloating =
+    Dtype::Category::complexfloating;
+inline constexpr Dtype::Category floating = Dtype::Category::floating;
+inline constexpr Dtype::Category inexact = Dtype::Category::inexact;
+inline constexpr Dtype::Category signedinteger = Dtype::Category::signedinteger;
+inline constexpr Dtype::Category unsignedinteger =
+    Dtype::Category::unsignedinteger;
+inline constexpr Dtype::Category integer = Dtype::Category::integer;
+inline constexpr Dtype::Category number = Dtype::Category::number;
+inline constexpr Dtype::Category generic = Dtype::Category::generic;
+
+bool issubdtype(const Dtype& a, const Dtype& b);
+bool issubdtype(const Dtype::Category& a, const Dtype& b);
+bool issubdtype(const Dtype& a, const Dtype::Category& b);
+bool issubdtype(const Dtype::Category& a, const Dtype::Category& b);
+
 Dtype promote_types(const Dtype& t1, const Dtype& t2);
 
 inline uint8_t size_of(const Dtype& t) {
@@ -71,23 +98,6 @@ inline uint8_t size_of(const Dtype& t) {
 
 Dtype::Kind kindof(const Dtype& t);
 
-inline bool is_unsigned(const Dtype& t) {
-  return kindof(t) == Dtype::Kind::u || kindof(t) == Dtype::Kind::b;
-}
-
-inline bool is_floating_point(const Dtype& t) {
-  return kindof(t) == Dtype::Kind::f || kindof(t) == Dtype::Kind::V ||
-      kindof(t) == Dtype::Kind::c;
-}
-
-inline bool is_complex(const Dtype& t) {
-  return kindof(t) == Dtype::Kind::c;
-}
-
-inline bool is_integral(const Dtype& t) {
-  return !(is_floating_point(t));
-}
-
 template <typename T>
 struct TypeToDtype {
   operator Dtype();
@@ -96,6 +106,6 @@ struct TypeToDtype {
 // Array protocol typestring for Dtype
 std::string dtype_to_array_protocol(const Dtype& t);
 // Dtype from array protocol type string
-Dtype dtype_from_array_protocol(const std::string& t);
+Dtype dtype_from_array_protocol(std::string_view t);
 
 } // namespace mlx::core

mlx/fast.cpp

@@ -1,5 +1,8 @@
 // Copyright © 2023-2024 Apple Inc.
 
+#include <cassert>
+#include <numeric>
+
 #include "mlx/fast.h"
 #include "mlx/fast_primitives.h"
 #include "mlx/ops.h"
@@ -64,7 +67,7 @@ array rms_norm(
     throw std::invalid_argument(msg.str());
   }
   auto out_type = result_type(x, weight);
-  if (!is_floating_point(out_type) || is_complex(out_type)) {
+  if (!issubdtype(out_type, floating)) {
     std::ostringstream msg;
     msg << "[rms_norm] Received unsupported type " << out_type << ".";
     throw std::invalid_argument(msg.str());
@@ -94,11 +97,69 @@ array rms_norm(
   return fallback({x, weight})[0];
 }
 
+std::vector<array> RMSNorm::vjp(
+    const std::vector<array>& primals,
+    const std::vector<array>& cotangents,
+    const std::vector<int>& argnums,
+    const std::vector<array>& outputs) {
+  assert(primals.size() == 2);
+  assert(outputs.size() == 1);
+  assert(cotangents.size() == 1);
+
+  auto s = stream();
+  auto fallback = [eps = eps_, s](const std::vector<array>& inputs) {
+    auto& x = inputs[0];
+    auto& w = inputs[1];
+    auto& g = inputs[2];
+
+    std::vector<array> vjps;
+
+    auto n = rsqrt(
+        add(mean(square(x, s), /* axis= */ -1, /* keepdims= */ true, s),
+            array(eps, x.dtype()),
+            s),
+        s);
+    auto n3 = power(n, array(3, x.dtype()), s);
+
+    // df/dx
+    auto gw = multiply(g, w, s);
+    auto t = mean(multiply(gw, x, s), /* axis= */ -1, /* keepdims= */ true, s);
+    t = multiply(multiply(x, t, s), n3, s);
+    vjps.push_back(subtract(multiply(gw, n, s), t, s));
+
+    // df/dw
+    std::vector<int> axes(g.ndim() - 1);
+    std::iota(axes.begin(), axes.end(), 0);
+    vjps.push_back(
+        sum(multiply(g, multiply(x, n, s), s), axes, /* keepdims= */ false, s));
+
+    return vjps;
+  };
+
+  auto vjps = array::make_arrays(
+      {primals[0].shape(), primals[1].shape()},
+      {primals[0].dtype(), primals[1].dtype()},
+      std::make_shared<RMSNormVJP>(s, fallback, eps_),
+      {primals[0], primals[1], cotangents[0]});
+
+  std::vector<array> returned_vjps;
+  for (auto& arg : argnums) {
+    returned_vjps.push_back(std::move(vjps[arg]));
+  }
+
+  return returned_vjps;
+}
+
 bool RMSNorm::is_equivalent(const Primitive& other) const {
   const RMSNorm& a_other = static_cast<const RMSNorm&>(other);
   return eps_ == a_other.eps_;
 }
 
+bool RMSNormVJP::is_equivalent(const Primitive& other) const {
+  const RMSNormVJP& a_other = static_cast<const RMSNormVJP&>(other);
+  return eps_ == a_other.eps_;
+}
+
 array layer_norm(
     const array& x,
     const std::optional<array>& weight,
@@ -128,7 +189,7 @@ array layer_norm(
       ? ((bias.has_value()) ? result_type(x, *weight, *bias)
                             : result_type(x, *weight))
       : x.dtype();
-  if (!is_floating_point(out_type) || is_complex(out_type)) {
+  if (!issubdtype(out_type, floating)) {
     std::ostringstream msg;
     msg << "[layer_norm] Received unsupported type " << out_type << ".";
     throw std::invalid_argument(msg.str());
@@ -176,11 +237,90 @@ array layer_norm(
   return fallback({x, passed_weight, passed_bias})[0];
 }
 
+std::vector<array> LayerNorm::vjp(
+    const std::vector<array>& primals,
+    const std::vector<array>& cotangents,
+    const std::vector<int>& argnums,
+    const std::vector<array>& outputs) {
+  assert(primals.size() == 3);
+  assert(outputs.size() == 1);
+  assert(cotangents.size() == 1);
+
+  auto s = stream();
+  auto fallback = [eps = eps_, s](const std::vector<array>& inputs) {
+    auto& x = inputs[0];
+    auto& w = inputs[1];
+    auto& b = inputs[2];
+    auto& g = inputs[3];
+
+    std::vector<array> vjps;
+
+    auto norm = number_of_elements(x, {-1}, true, x.dtype(), s);
+    auto sumx = sum(x, /* axis= */ -1, /* keepdims= */ true, s);
+    auto sumx2 = sum(square(x, s), /* axis= */ -1, /* keepdims= */ true, s);
+    auto mu = multiply(sumx, norm, s);
+    auto mu2 = multiply(sumx2, norm, s);
+    auto var = subtract(mu2, square(mu, s), s);
+    auto n = rsqrt(add(var, array(eps, x.dtype()), s));
+    auto n3 = power(n, array(3, x.dtype()), s);
+    auto x_c = subtract(x, mu, s);
+
+    // df/dx
+    auto wg = multiply(w, g, s);
+    auto sumwg =
+        multiply(sum(wg, /* axis= */ -1, /* keepdims= */ true, s), norm, s);
+    auto sumwgxc = multiply(
+        sum(multiply(wg, x_c, s), /* axis= */ -1, /* keepdims= */ true, s),
+        norm,
+        s);
+    auto t1 = multiply(multiply(x_c, sumwgxc, s), n3, s);
+    auto t2 = multiply(subtract(wg, sumwg, s), n, s);
+    vjps.push_back(subtract(t2, t1, s));
+
+    // df/dw
+    std::vector<int> axes(g.ndim() - 1);
+    std::iota(axes.begin(), axes.end(), 0);
+    if (w.ndim() == 0) {
+      vjps.push_back(zeros_like(w, s));
+    } else {
+      vjps.push_back(sum(
+          multiply(g, multiply(x_c, n, s), s), axes, /* keepdims= */ false, s));
+    }
+
+    // df/db
+    if (b.ndim() == 0) {
+      vjps.push_back(zeros_like(w, s));
+    } else {
+      vjps.push_back(sum(g, axes, /* keepdims= */ false, s));
+    }
+
+    return vjps;
+  };
+
+  auto vjps = array::make_arrays(
+      {primals[0].shape(), primals[1].shape(), primals[2].shape()},
+      {primals[0].dtype(), primals[1].dtype(), primals[2].dtype()},
+      std::make_shared<LayerNormVJP>(s, fallback, eps_),
+      {primals[0], primals[1], primals[2], cotangents[0]});
+
+  std::vector<array> returned_vjps;
+  for (auto& arg : argnums) {
+    returned_vjps.push_back(std::move(vjps[arg]));
+  }
+
+  return returned_vjps;
+}
+
 bool LayerNorm::is_equivalent(const Primitive& other) const {
   const LayerNorm& a_other = static_cast<const LayerNorm&>(other);
   return eps_ == a_other.eps_;
 }
 
+bool LayerNormVJP::is_equivalent(const Primitive& other) const {
+  const LayerNormVJP& a_other = static_cast<const LayerNormVJP&>(other);
+  return eps_ == a_other.eps_;
+}
+
 array rope(
     const array& x,
     int dims,
@@ -188,19 +328,16 @@ array rope(
     float base,
     float scale,
     int offset,
-    StreamOrDevice s /* = {} */) {
+    bool forward,
+    StreamOrDevice s) {
   if (x.ndim() < 3) {
     std::ostringstream msg;
     msg << "[rope] Input must have at least 3 dimensions but got input with "
         << x.ndim() << " dimensions.";
     throw std::invalid_argument(msg.str());
   }
-  if (traditional && x.shape(-1) != dims) {
-    throw std::invalid_argument(
-        "[rope] Does not support partial traditional application.");
-  }
 
-  auto fallback = [dims, traditional, base, scale, offset, s](
+  auto fallback = [dims, traditional, base, scale, offset, forward, s](
                       const std::vector<array>& inputs) {
     auto& shape = inputs[0].shape();
     int ndim = shape.size();
@@ -217,16 +354,39 @@ array rope(
     auto coss = cos(theta, s);
     auto sins = sin(theta, s);
 
-    if (traditional) {
-      auto x1 = slice(x, {0, 0, 0}, x.shape(), {1, 1, 2}, s);
-      auto x2 = slice(x, {0, 0, 1}, x.shape(), {1, 1, 2}, s);
+    auto apply_rope = [forward, s](
+                          const array& x1,
+                          const array& x2,
+                          const array& coss,
+                          const array& sins) {
       std::vector<array> outs;
-      outs.push_back(subtract(multiply(x1, coss, s), multiply(x2, sins, s), s));
-      outs.push_back(add(multiply(x1, sins, s), multiply(x2, coss, s), s));
+      if (forward) {
+        outs.push_back(
+            subtract(multiply(x1, coss, s), multiply(x2, sins, s), s));
+        outs.push_back(add(multiply(x1, sins, s), multiply(x2, coss, s), s));
+      } else {
+        outs.push_back(add(multiply(x2, sins, s), multiply(x1, coss, s), s));
+        outs.push_back(
+            subtract(multiply(x2, coss, s), multiply(x1, sins, s), s));
+      }
+      return outs;
+    };
+
+    if (traditional) {
+      auto x1 =
+          slice(x, {0, 0, 0}, {x.shape(0), x.shape(1), dims}, {1, 1, 2}, s);
+      auto x2 =
+          slice(x, {0, 0, 1}, {x.shape(0), x.shape(1), dims}, {1, 1, 2}, s);
+      auto outs = apply_rope(x1, x2, coss, sins);
       for (auto& o : outs) {
         o = expand_dims(o, 3, s);
       }
-      return std::vector<array>{reshape(concatenate(outs, 3, s), shape, s)};
+      auto out = concatenate(outs, 3, s);
+      if (dims < x.shape(-1)) {
+        out = reshape(out, {x.shape(0), x.shape(1), dims});
+        out = concatenate({out, slice(x, {0, 0, dims}, x.shape(), s)}, 2, s);
+      }
+      return std::vector<array>{reshape(out, shape, s)};
     } else {
       auto out_s = x.shape();
       out_s.back() = half_dims;
@@ -234,9 +394,7 @@ array rope(
       out_s.back() = dims;
       auto x2 = slice(x, {0, 0, half_dims}, out_s, s);
 
-      std::vector<array> outs;
-      outs.push_back(subtract(multiply(x1, coss, s), multiply(x2, sins, s), s));
-      outs.push_back(add(multiply(x1, sins, s), multiply(x2, coss, s), s));
+      auto outs = apply_rope(x1, x2, coss, sins);
       if (dims < x.shape(-1)) {
         outs.push_back(slice(x, {0, 0, dims}, x.shape(), s));
       }
@@ -249,18 +407,54 @@ array rope(
         x.shape(),
         x.dtype(),
         std::make_shared<RoPE>(
-            stream, fallback, dims, traditional, base, scale, offset),
+            stream, fallback, dims, traditional, base, scale, offset, forward),
         {x});
   }
   return fallback({x})[0];
 }
 
+array rope(
+    const array& x,
+    int dims,
+    bool traditional,
+    float base,
+    float scale,
+    int offset,
+    StreamOrDevice s /* = {} */) {
+  return rope(x, dims, traditional, base, scale, offset, true, s);
+}
+
+std::vector<array> RoPE::vjp(
+    const std::vector<array>& primals,
+    const std::vector<array>& cotangents,
+    const std::vector<int>& argnums,
+    const std::vector<array>& outputs) {
+  auto s = stream();
+  auto fallback = [dims = dims_,
+                   traditional = traditional_,
+                   base = base_,
+                   scale = scale_,
+                   offset = offset_,
+                   forward = forward_,
+                   s](std::vector<array> inputs) {
+    return std::vector<array>{
+        rope(inputs[0], dims, traditional, base, scale, offset, !forward, s)};
+  };
+
+  return {array(
+      cotangents[0].shape(),
+      cotangents[0].dtype(),
+      std::make_shared<RoPE>(
+          s, fallback, dims_, traditional_, base_, scale_, offset_, !forward_),
+      cotangents)};
+}
+
 bool RoPE::is_equivalent(const Primitive& other) const {
   const RoPE& a_other = static_cast<const RoPE&>(other);
   return (
       dims_ == a_other.dims_ && base_ == a_other.base_ &&
       scale_ == a_other.scale_ && traditional_ == a_other.traditional_ &&
-      offset_ == a_other.offset_);
+      offset_ == a_other.offset_ && forward_ == a_other.forward_);
 }
 
 /** Computes: O = softmax(Q @ K.T) @ V **/
@@ -319,7 +513,7 @@ array scaled_dot_product_attention(
   }
 
   auto final_type = result_type(queries, keys, values);
-  if (!is_floating_point(final_type) || is_complex(final_type)) {
+  if (!issubdtype(final_type, floating)) {
     std::ostringstream msg;
     msg << "[scaled_dot_product_attention] Received unsupported type "
         << final_type << ".";
@@ -356,10 +550,7 @@ array scaled_dot_product_attention(
     if (needs_mask) {
       scores = add(scores, inputs[3], s);
     }
-    scores = astype(
-        softmax(astype(scores, float32, s), std::vector<int>{-1}, s),
-        final_type,
-        s);
+    scores = softmax(scores, std::vector<int>{-1}, true, s);
     auto out = matmul(scores, v, s);
     if (n_repeats > 1) {
       out = reshape(out, {B, n_q_heads, L, -1}, s);

mlx/fast_primitives.h

@@ -48,6 +48,12 @@ class RMSNorm : public Custom {
   void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override;
 
+  std::vector<array> vjp(
+      const std::vector<array>& primals,
+      const std::vector<array>& cotangents,
+      const std::vector<int>& argnums,
+      const std::vector<array>& outputs) override;
+
   DEFINE_PRINT(RMSNorm)
   bool is_equivalent(const Primitive& other) const override;
 
@@ -56,6 +62,29 @@ class RMSNorm : public Custom {
   float eps_;
 };
 
+class RMSNormVJP : public Custom {
+ public:
+  RMSNormVJP(
+      Stream stream,
+      std::function<std::vector<array>(std::vector<array>)> fallback,
+      float eps)
+      : Custom(stream, fallback), eps_(eps){};
+
+  void eval_cpu(const std::vector<array>& inputs, std::vector<array>& outputs)
+      override {
+    throw std::runtime_error("NYI");
+  };
+  void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)
+      override;
+
+  DEFINE_PRINT(RMSNormVJP)
+  bool is_equivalent(const Primitive& other) const override;
+
+ private:
+  std::function<std::vector<array>(std::vector<array>)> fallback_;
+  float eps_;
+};
+
 class LayerNorm : public Custom {
  public:
   LayerNorm(
@@ -71,6 +100,12 @@ class LayerNorm : public Custom {
   void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override;
 
+  std::vector<array> vjp(
+      const std::vector<array>& primals,
+      const std::vector<array>& cotangents,
+      const std::vector<int>& argnums,
+      const std::vector<array>& outputs) override;
+
   DEFINE_PRINT(LayerNorm)
   bool is_equivalent(const Primitive& other) const override;
 
@@ -79,6 +114,29 @@ class LayerNorm : public Custom {
   float eps_;
 };
 
+class LayerNormVJP : public Custom {
+ public:
+  LayerNormVJP(
+      Stream stream,
+      std::function<std::vector<array>(std::vector<array>)> fallback,
+      float eps)
+      : Custom(stream, fallback), eps_(eps){};
+
+  void eval_cpu(const std::vector<array>& inputs, std::vector<array>& outputs)
+      override {
+    throw std::runtime_error("NYI");
+  };
+  void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)
+      override;
+
+  DEFINE_PRINT(LayerNormVJP)
+  bool is_equivalent(const Primitive& other) const override;
+
+ private:
+  std::function<std::vector<array>(std::vector<array>)> fallback_;
+  float eps_;
+};
+
 class RoPE : public Custom {
  public:
   RoPE(
@@ -88,13 +146,15 @@ class RoPE : public Custom {
       bool traditional,
       float base,
       float scale,
-      int offset)
+      int offset,
+      bool forward)
       : Custom(stream, fallback),
         dims_(dims),
         traditional_(traditional),
         base_(base),
         scale_(scale),
-        offset_(offset){};
+        offset_(offset),
+        forward_(forward){};
 
   void eval_cpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override {
@@ -103,6 +163,12 @@ class RoPE : public Custom {
   void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override;
 
+  std::vector<array> vjp(
+      const std::vector<array>& primals,
+      const std::vector<array>& cotangents,
+      const std::vector<int>& argnums,
+      const std::vector<array>& outputs) override;
+
   DEFINE_PRINT(RoPE)
   bool is_equivalent(const Primitive& other) const override;
 
@@ -113,6 +179,7 @@ class RoPE : public Custom {
   float base_;
   float scale_;
   int offset_;
+  bool forward_;
 };
 
 class ScaledDotProductAttention : public Custom {
@@ -126,7 +193,7 @@ class ScaledDotProductAttention : public Custom {
 
   void eval_cpu(const std::vector<array>& inputs, std::vector<array>& outputs)
       override {
-    outputs[0] = fallback_(inputs)[0];
+    throw std::runtime_error("NYI");
   };
 
   void eval_gpu(const std::vector<array>& inputs, std::vector<array>& outputs)

mlx/fft.cpp

@@ -95,7 +95,7 @@ array fft_impl(
   return array(
       out_shape,
       out_type,
-      std::make_unique<FFT>(to_stream(s), valid_axes, inverse, real),
+      std::make_shared<FFT>(to_stream(s), valid_axes, inverse, real),
       {astype(in, in_type, s)});
 }
 

mlx/fft.h

@@ -6,12 +6,10 @@
 
 #include "array.h"
 #include "device.h"
-#include "stream.h"
+#include "utils.h"
 
 namespace mlx::core::fft {
 
-using StreamOrDevice = std::variant<std::monostate, Stream, Device>;
-
 /** Compute the n-dimensional Fourier Transform. */
 array fftn(
     const array& a,

mlx/graph_utils.cpp

@@ -23,8 +23,7 @@ const std::string& NodeNamer::get_name(const array& x) {
       letters.push_back('A' + (var_num - 1) % 26);
       var_num = (var_num - 1) / 26;
     }
-    std::string name(letters.rbegin(), letters.rend());
-    names.insert({x.id(), name});
+    names.emplace(x.id(), std::string(letters.rbegin(), letters.rend()));
 
     return get_name(x);
   }

mlx/graph_utils.h

@@ -14,15 +14,15 @@ struct NodeNamer {
 
 void print_graph(std::ostream& os, const std::vector<array>& outputs);
 
-template <typename... Arrays>
-void print_graph(std::ostream& os, Arrays... outputs) {
+template <typename... Arrays, typename = enable_for_arrays_t<Arrays...>>
+void print_graph(std::ostream& os, Arrays&&... outputs) {
   print_graph(os, std::vector<array>{std::forward<Arrays>(outputs)...});
 }
 
 void export_to_dot(std::ostream& os, const std::vector<array>& outputs);
 
-template <typename... Arrays>
-void export_to_dot(std::ostream& os, Arrays... outputs) {
+template <typename... Arrays, typename = enable_for_arrays_t<Arrays...>>
+void export_to_dot(std::ostream& os, Arrays&&... outputs) {
   export_to_dot(os, std::vector<array>{std::forward<Arrays>(outputs)...});
 }
 

mlx/io.h

@@ -23,13 +23,13 @@ using SafetensorsLoad = std::pair<
 void save(std::shared_ptr<io::Writer> out_stream, array a);
 
 /** Save array to file in .npy format */
-void save(const std::string& file, array a);
+void save(std::string file, array a);
 
 /** Load array from reader in .npy format */
 array load(std::shared_ptr<io::Reader> in_stream, StreamOrDevice s = {});
 
 /** Load array from file in .npy format */
-array load(const std::string& file, StreamOrDevice s = {});
+array load(std::string file, StreamOrDevice s = {});
 
 /** Load array map from .safetensors file format */
 SafetensorsLoad load_safetensors(
@@ -44,7 +44,7 @@ void save_safetensors(
     std::unordered_map<std::string, array>,
     std::unordered_map<std::string, std::string> metadata = {});
 void save_safetensors(
-    const std::string& file,
+    std::string file,
     std::unordered_map<std::string, array>,
     std::unordered_map<std::string, std::string> metadata = {});
 

mlx/io/gguf.cpp

@@ -206,7 +206,7 @@ std::unordered_map<std::string, array> load_arrays(gguf_ctx* ctx) {
   std::unordered_map<std::string, array> array_map;
   gguf_tensor tensor;
 
-  auto check_insert = [](auto inserted) {
+  auto check_insert = [](const auto& inserted) {
     if (!inserted.second) {
       std::ostringstream msg;
       msg << "[load_gguf] Duplicate parameter name " << inserted.first->second
@@ -216,6 +216,7 @@ std::unordered_map<std::string, array> load_arrays(gguf_ctx* ctx) {
   };
 
   while (gguf_get_tensor(ctx, &tensor)) {
+    std::string name(tensor.name, tensor.namelen);
     if (tensor.type == GGUF_TYPE_Q4_0 || tensor.type == GGUF_TYPE_Q4_1 ||
         tensor.type == GGUF_TYPE_Q8_0) {
       gguf_load_quantized(array_map, tensor);
@@ -224,14 +225,14 @@ std::unordered_map<std::string, array> load_arrays(gguf_ctx* ctx) {
 
       const auto& [data, dtype] = extract_tensor_data(&tensor);
       array loaded_array = array(data, get_shape(tensor), dtype);
-      array_map.insert({name, loaded_array});
+      check_insert(array_map.insert({name, loaded_array}));
     }
   }
   return array_map;
 }
 
 GGUFLoad load_gguf(const std::string& file, StreamOrDevice s) {
-  gguf_ctx* ctx = gguf_open(file.c_str());
+  gguf_ctx* ctx = gguf_open(file.data());
   if (!ctx) {
     throw std::runtime_error("[load_gguf] gguf_init failed");
   }

mlx/io/gguf_quants.cpp

@@ -105,7 +105,8 @@ void gguf_load_quantized(
     weights_per_byte = 1;
   }
 
-  std::string name = std::string(tensor.name, tensor.namelen);
+  std::string name(tensor.name, tensor.namelen);
+
   std::vector<int> shape = get_shape(tensor);
   const uint64_t weights_per_block = 32;
   if (shape[shape.size() - 1] % weights_per_block != 0) {
@@ -136,9 +137,9 @@ void gguf_load_quantized(
     extract_q8_0_data(tensor, weights, scales, biases);
   }
 
-  a.insert({name, weights});
+  a.emplace(name, std::move(weights));
 
-  auto check_insert = [](auto inserted) {
+  auto check_insert = [](const auto& inserted) {
     if (!inserted.second) {
       std::ostringstream msg;
       msg << "[load_gguf] Duplicate parameter name " << inserted.first->second
@@ -147,11 +148,11 @@ void gguf_load_quantized(
     }
   };
 
-  const std::string weight_suffix = ".weight";
+  constexpr std::string_view weight_suffix = ".weight";
   const std::string name_prefix =
       name.substr(0, name.length() - weight_suffix.length());
-  check_insert(a.insert({name_prefix + ".scales", scales}));
-  check_insert(a.insert({name_prefix + ".biases", biases}));
+  check_insert(a.emplace(name_prefix + ".scales", std::move(scales)));
+  check_insert(a.emplace(name_prefix + ".biases", std::move(biases)));
 }
 
 } // namespace mlx::core

mlx/io/load.cpp

@@ -26,16 +26,6 @@ constexpr uint8_t MAGIC[] = {
     0x59,
 };
 
-inline bool is_big_endian_() {
-  union ByteOrder {
-    int32_t i;
-    uint8_t c[4];
-  };
-  ByteOrder b = {0x01234567};
-
-  return b.c[0] == 0x01;
-}
-
 } // namespace
 
 /** Save array to out stream in .npy format */
@@ -73,8 +63,7 @@ void save(std::shared_ptr<io::Writer> out_stream, array a) {
   std::string fortran_order = a.flags().col_contiguous ? "True" : "False";
   std::ostringstream header;
   header << "{'descr': '" << dtype_to_array_protocol(a.dtype()) << "',"
-         << " 'fortran_order': " << fortran_order << ","
-         << " 'shape': (";
+         << " 'fortran_order': " << fortran_order << "," << " 'shape': (";
   for (auto i : a.shape()) {
     header << i << ", ";
   }
@@ -94,7 +83,7 @@ void save(std::shared_ptr<io::Writer> out_stream, array a) {
     uint16_t v1_header_len = header.tellp();
     const char* len_bytes = reinterpret_cast<const char*>(&v1_header_len);
 
-    if (!is_big_endian_()) {
+    if (!is_big_endian()) {
       magic_ver_len.write(len_bytes, 2);
     } else {
       magic_ver_len.write(len_bytes + 1, 1);
@@ -106,7 +95,7 @@ void save(std::shared_ptr<io::Writer> out_stream, array a) {
     uint32_t v2_header_len = header.tellp();
     const char* len_bytes = reinterpret_cast<const char*>(&v2_header_len);
 
-    if (!is_big_endian_()) {
+    if (!is_big_endian()) {
       magic_ver_len.write(len_bytes, 4);
     } else {
       magic_ver_len.write(len_bytes + 3, 1);
@@ -124,16 +113,13 @@ void save(std::shared_ptr<io::Writer> out_stream, array a) {
 }
 
 /** Save array to file in .npy format */
-void save(const std::string& file_, array a) {
-  // Open and check file
-  std::string file = file_;
-
+void save(std::string file, array a) {
   // Add .npy to file name if it is not there
   if (file.length() < 4 || file.substr(file.length() - 4, 4) != ".npy")
     file += ".npy";
 
   // Serialize array
-  save(std::make_shared<io::FileWriter>(file), a);
+  save(std::make_shared<io::FileWriter>(std::move(file)), a);
 }
 
 /** Load array from reader in .npy format */
@@ -219,7 +205,7 @@ array load(std::shared_ptr<io::Reader> in_stream, StreamOrDevice s) {
   // Build primitive
 
   size_t offset = 8 + header_len_size + header.length();
-  bool swap_endianness = read_is_big_endian != is_big_endian_();
+  bool swap_endianness = read_is_big_endian != is_big_endian();
 
   if (col_contiguous) {
     std::reverse(shape.begin(), shape.end());
@@ -227,7 +213,7 @@ array load(std::shared_ptr<io::Reader> in_stream, StreamOrDevice s) {
   auto loaded_array = array(
       shape,
       dtype,
-      std::make_unique<Load>(to_stream(s), in_stream, offset, swap_endianness),
+      std::make_shared<Load>(to_stream(s), in_stream, offset, swap_endianness),
       std::vector<array>{});
   if (col_contiguous) {
     loaded_array = transpose(loaded_array, s);
@@ -237,8 +223,8 @@ array load(std::shared_ptr<io::Reader> in_stream, StreamOrDevice s) {
 }
 
 /** Load array from file in .npy format */
-array load(const std::string& file, StreamOrDevice s) {
-  return load(std::make_shared<io::FileReader>(file), s);
+array load(std::string file, StreamOrDevice s) {
+  return load(std::make_shared<io::FileReader>(std::move(file)), s);
 }
 
 } // namespace mlx::core

mlx/io/safetensor.cpp

@@ -60,7 +60,7 @@ std::string dtype_to_safetensor_str(Dtype t) {
   }
 }
 
-Dtype dtype_from_safetensor_str(std::string str) {
+Dtype dtype_from_safetensor_str(std::string_view str) {
   if (str == ST_F32) {
     return float32;
   } else if (str == ST_F16) {
@@ -88,7 +88,8 @@ Dtype dtype_from_safetensor_str(std::string str) {
   } else if (str == ST_C64) {
     return complex64;
   } else {
-    throw std::runtime_error("[safetensor] unsupported dtype " + str);
+    throw std::runtime_error(
+        "[safetensor] unsupported dtype " + std::string(str));
   }
 }
 
@@ -129,14 +130,14 @@ SafetensorsLoad load_safetensors(
       }
       continue;
     }
-    std::string dtype = item.value().at("dtype");
-    std::vector<int> shape = item.value().at("shape");
-    std::vector<size_t> data_offsets = item.value().at("data_offsets");
+    const std::string& dtype = item.value().at("dtype");
+    const std::vector<int>& shape = item.value().at("shape");
+    const std::vector<size_t>& data_offsets = item.value().at("data_offsets");
     Dtype type = dtype_from_safetensor_str(dtype);
     auto loaded_array = array(
         shape,
         type,
-        std::make_unique<Load>(
+        std::make_shared<Load>(
             to_stream(s), in_stream, offset + data_offsets.at(0), false),
         std::vector<array>{});
     res.insert({item.key(), loaded_array});
@@ -207,19 +208,17 @@ void save_safetensors(
 }
 
 void save_safetensors(
-    const std::string& file_,
+    std::string file,
     std::unordered_map<std::string, array> a,
     std::unordered_map<std::string, std::string> metadata /* = {} */) {
-  // Open and check file
-  std::string file = file_;
-
   // Add .safetensors to file name if it is not there
   if (file.length() < 12 ||
       file.substr(file.length() - 12, 12) != ".safetensors")
     file += ".safetensors";
 
   // Serialize array
-  save_safetensors(std::make_shared<io::FileWriter>(file), a, metadata);
+  save_safetensors(
+      std::make_shared<io::FileWriter>(std::move(file)), a, metadata);
 }
 
 } // namespace mlx::core

mlx/linalg.cpp

@@ -11,7 +11,7 @@
 namespace mlx::core::linalg {
 
 Dtype at_least_float(const Dtype& d) {
-  return is_floating_point(d) ? d : promote_types(d, float32);
+  return issubdtype(d, inexact) ? d : promote_types(d, float32);
 }
 
 inline array l2_norm(
@@ -19,7 +19,7 @@ inline array l2_norm(
     const std::vector<int>& axis,
     bool keepdims,
     StreamOrDevice s) {
-  if (is_complex(a.dtype())) {
+  if (issubdtype(a.dtype(), complexfloating)) {
     return sqrt(sum(abs(a, s) * abs(a, s), axis, keepdims, s), s);
   } else {
     return sqrt(sum(square(a, s), axis, keepdims, s), s);

mlx/ops.cpp

@@ -47,7 +47,7 @@ std::pair<std::vector<int>, std::vector<int>> compute_reduce_shape(
 }
 
 Dtype at_least_float(const Dtype& d) {
-  return is_floating_point(d) ? d : promote_types(d, float32);
+  return issubdtype(d, inexact) ? d : promote_types(d, float32);
 }
 
 } // namespace
@@ -158,50 +158,64 @@ array linspace(
       to_stream(s));
 }
 
-array astype(const array& a, Dtype dtype, StreamOrDevice s /* = {} */) {
+array astype(array a, Dtype dtype, StreamOrDevice s /* = {} */) {
   if (dtype == a.dtype()) {
-    return a;
+    return std::move(a);
   }
+  auto copied_shape = a.shape(); // |a| will be moved
   return array(
-      a.shape(), dtype, std::make_shared<AsType>(to_stream(s), dtype), {a});
+      std::move(copied_shape),
+      dtype,
+      std::make_shared<AsType>(to_stream(s), dtype),
+      {std::move(a)});
 }
 
 array as_strided(
-    const array& a,
+    array a,
     std::vector<int> shape,
     std::vector<size_t> strides,
     size_t offset,
     StreamOrDevice s /* = {} */) {
-  // Force the input array to be contiguous
-  auto x = reshape(a, {-1}, s);
+  auto copied_shape = shape; // |shape| will be moved
+  auto dtype = a.dtype(); // |a| will be moved
   return array(
-      shape,
-      a.dtype(),
-      std::make_shared<AsStrided>(to_stream(s), shape, strides, offset),
-      {x});
+      std::move(copied_shape),
+      dtype,
+      std::make_shared<AsStrided>(
+          to_stream(s), std::move(shape), std::move(strides), offset),
+      // Force the input array to be contiguous.
+      {reshape(std::move(a), {-1}, s)});
 }
 
-array copy(const array& a, StreamOrDevice s /* = {} */) {
-  return array(a.shape(), a.dtype(), std::make_shared<Copy>(to_stream(s)), {a});
+array copy(array a, StreamOrDevice s /* = {} */) {
+  auto copied_shape = a.shape(); // |a| will be moved
+  auto dtype = a.dtype();
+  return array(
+      std::move(copied_shape),
+      dtype,
+      std::make_shared<Copy>(to_stream(s)),
+      {std::move(a)});
 }
 
 array full(
-    const std::vector<int>& shape,
-    const array& vals,
+    std::vector<int> shape,
+    array vals,
     Dtype dtype,
     StreamOrDevice s /* = {} */) {
-  if (std::any_of(shape.begin(), shape.end(), [](auto i) { return i < 0; })) {
+  if (std::any_of(shape.begin(), shape.end(), [](int i) { return i < 0; })) {
     throw std::invalid_argument("[full] Negative dimensions not allowed.");
   }
-  auto in = broadcast_to(astype(vals, dtype, s), shape, s);
-  return array(shape, dtype, std::make_shared<Full>(to_stream(s)), {in});
+  auto copied_shape = shape; // |shape| will be moved
+  return array(
+      std::move(copied_shape),
+      dtype,
+      std::make_shared<Full>(to_stream(s)),
+      {broadcast_to(astype(std::move(vals), dtype, s), std::move(shape), s)});
 }
 
-array full(
-    const std::vector<int>& shape,
-    const array& vals,
-    StreamOrDevice s /* = {} */) {
-  return full(shape, vals, vals.dtype(), to_stream(s));
+array full(std::vector<int> shape, array vals, StreamOrDevice s /* = {} */) {
+  auto dtype = vals.dtype(); // |vals| will be moved
+  return full(std::move(shape), std::move(vals), dtype, to_stream(s));
 }
 
 array zeros(
@@ -682,6 +696,41 @@ split(const array& a, int num_splits, StreamOrDevice s /* = {} */) {
   return split(a, num_splits, 0, to_stream(s));
 }
 
+std::vector<array> meshgrid(
+    const std::vector<array>& arrays,
+    bool sparse /* = false */,
+    std::string indexing /* = "xy" */,
+    StreamOrDevice s /* = {} */) {
+  if (indexing != "xy" && indexing != "ij") {
+    throw std::invalid_argument(
+        "[meshgrid] Invalid indexing value. Valid values are 'xy' and 'ij'.");
+  }
+
+  auto ndim = arrays.size();
+  std::vector<array> outputs;
+  for (int i = 0; i < ndim; ++i) {
+    std::vector<int> shape(ndim, 1);
+    shape[i] = -1;
+    outputs.push_back(reshape(arrays[i], std::move(shape), s));
+  }
+
+  if (indexing == "xy" and ndim > 1) {
+    std::vector<int> shape(ndim, 1);
+
+    shape[1] = arrays[0].size();
+    outputs[0] = reshape(arrays[0], shape, s);
+    shape[1] = 1;
+    shape[0] = arrays[1].size();
+    outputs[1] = reshape(arrays[1], std::move(shape), s);
+  }
+
+  if (!sparse) {
+    outputs = broadcast_arrays(outputs, s);
+  }
+
+  return outputs;
+}
+
 array clip(
     const array& a,
     const std::optional<array>& a_min,
@@ -883,15 +932,15 @@ array pad(
     if (low_pad_size[i] < 0) {
       std::ostringstream msg;
       msg << "Invalid low padding size (" << low_pad_size[i]
-          << ") passed to pad"
-          << " for axis " << i << ". Padding sizes must be non-negative";
+          << ") passed to pad" << " for axis " << i
+          << ". Padding sizes must be non-negative";
       throw std::invalid_argument(msg.str());
     }
     if (high_pad_size[i] < 0) {
       std::ostringstream msg;
       msg << "Invalid high padding size (" << high_pad_size[i]
-          << ") passed to pad"
-          << " for axis " << i << ". Padding sizes must be non-negative";
+          << ") passed to pad" << " for axis " << i
+          << ". Padding sizes must be non-negative";
       throw std::invalid_argument(msg.str());
     }
 
@@ -1001,19 +1050,30 @@ array transpose(
   for (auto& ax : axes) {
     ax = ax < 0 ? ax + a.ndim() : ax;
   }
-  std::set dims(axes.begin(), axes.end());
-  if (dims.size() != axes.size()) {
-    throw std::invalid_argument("Repeat axes not allowed in transpose.");
+  if (axes.size() != a.ndim()) {
+    std::ostringstream msg;
+    msg << "[transpose] Recived " << axes.size() << " axes for array with "
+        << a.ndim() << " dimensions.";
+    throw std::invalid_argument(msg.str());
   }
-  if (dims.size() != a.ndim() ||
-      a.ndim() > 0 &&
-          (*dims.begin() != 0 || *dims.rbegin() != (a.ndim() - 1))) {
-    throw std::invalid_argument("Transpose axes don't match array dimensions.");
+
+  // Check in bounds and for duplicates
+  std::vector<int> shape(axes.size(), 0);
+  for (auto& ax : axes) {
+    if (ax < 0 || ax >= a.ndim()) {
+      std::ostringstream msg;
+      msg << "[transpose] Invalid axis (" << ax << ") for array with "
+          << a.ndim() << " dimensions.";
+      throw std::invalid_argument(msg.str());
+    }
+    if (shape[ax] != 0) {
+      throw std::invalid_argument("[transpose] Repeat axes not allowed.");
+    }
+    shape[ax] = 1;
   }
-  std::vector<int> shape;
-  shape.reserve(axes.size());
-  for (auto ax : axes) {
-    shape.push_back(a.shape()[ax]);
+
+  for (int i = 0; i < axes.size(); ++i) {
+    shape[i] = a.shape()[axes[i]];
   }
   return array(
       std::move(shape),
@@ -1140,7 +1200,7 @@ array array_equal(
     return array(false);
   } else {
     auto dtype = promote_types(a.dtype(), b.dtype());
-    equal_nan &= is_floating_point(dtype);
+    equal_nan &= issubdtype(dtype, inexact);
     return all(
         array(
             a.shape(),
@@ -1153,7 +1213,7 @@ array array_equal(
 }
 
 array isnan(const array& a, StreamOrDevice s /* = {} */) {
-  if (is_integral(a.dtype())) {
+  if (issubdtype(a.dtype(), integer) || a.dtype() == bool_) {
     return full(a.shape(), false, bool_, s);
   }
   return not_equal(a, a, s);
@@ -1164,14 +1224,14 @@ array isinf(const array& a, StreamOrDevice s /* = {} */) {
 }
 
 array isposinf(const array& a, StreamOrDevice s /* = {} */) {
-  if (is_integral(a.dtype())) {
+  if (issubdtype(a.dtype(), integer) || a.dtype() == bool_) {
     return full(a.shape(), false, bool_, s);
   }
   return equal(a, array(std::numeric_limits<float>::infinity(), a.dtype()), s);
 }
 
 array isneginf(const array& a, StreamOrDevice s /* = {} */) {
-  if (is_integral(a.dtype())) {
+  if (issubdtype(a.dtype(), integer) || a.dtype() == bool_) {
     return full(a.shape(), false, bool_, s);
   }
   return equal(a, array(-std::numeric_limits<float>::infinity(), a.dtype()), s);
@@ -1416,6 +1476,34 @@ array var(
   return var(a, std::vector<int>{axis}, keepdims, ddof, to_stream(s));
 }
 
+array std(
+    const array& a,
+    bool keepdims,
+    int ddof /* = 0*/,
+    StreamOrDevice s /* = {}*/) {
+  std::vector<int> axes(a.ndim());
+  std::iota(axes.begin(), axes.end(), 0);
+  return std(a, axes, keepdims, ddof, to_stream(s));
+}
+
+array std(
+    const array& a,
+    const std::vector<int>& axes,
+    bool keepdims /* = false */,
+    int ddof /* = 0*/,
+    StreamOrDevice s /* = {}*/) {
+  return sqrt(var(a, axes, keepdims, ddof, s), s);
+}
+
+array std(
+    const array& a,
+    int axis,
+    bool keepdims /* = false */,
+    int ddof /* = 0*/,
+    StreamOrDevice s /* = {} */) {
+  return std(a, std::vector<int>{axis}, keepdims, ddof, to_stream(s));
+}
+
 array prod(const array& a, bool keepdims, StreamOrDevice s /* = {}*/) {
   std::vector<int> axes(a.ndim());
   std::iota(axes.begin(), axes.end(), 0);
@@ -1929,7 +2017,7 @@ array floor_divide(
     const array& b,
     StreamOrDevice s /* = {} */) {
   auto dtype = promote_types(a.dtype(), b.dtype());
-  if (is_floating_point(dtype)) {
+  if (issubdtype(dtype, inexact)) {
     return floor(divide(a, b, s), s);
   }
 
@@ -1957,7 +2045,7 @@ array operator%(const array& a, const array& b) {
 std::vector<array>
 divmod(const array& a, const array& b, StreamOrDevice s /* = {} */) {
   auto dtype = promote_types(a.dtype(), b.dtype());
-  if (is_complex(dtype)) {
+  if (issubdtype(dtype, complexfloating)) {
     throw std::invalid_argument("[divmod] Complex type not supported.");
   }
   auto inputs =
@@ -2019,6 +2107,13 @@ array exp(const array& a, StreamOrDevice s /* = {} */) {
   return array(a.shape(), dtype, std::make_shared<Exp>(to_stream(s)), {input});
 }
 
+array expm1(const array& a, StreamOrDevice s /* = {} */) {
+  auto dtype = at_least_float(a.dtype());
+  auto input = astype(a, dtype, s);
+  return array(
+      a.shape(), dtype, std::make_shared<Expm1>(to_stream(s)), {input});
+}
+
 array sin(const array& a, StreamOrDevice s /* = {} */) {
   auto dtype = at_least_float(a.dtype());
   auto input = astype(a, dtype, s);
@@ -2220,7 +2315,7 @@ array matmul(
   }
   // Type promotion
   auto out_type = promote_types(a.dtype(), b.dtype());
-  if (!is_floating_point(out_type) || is_complex(out_type)) {
+  if (!issubdtype(out_type, floating)) {
     std::ostringstream msg;
     msg << "[matmul] Only real floating point types are supported but "
         << a.dtype() << " and " << b.dtype() << " were provided which results"
@@ -2330,7 +2425,7 @@ array gather(
 
   // Promote indices to the same type
   auto dtype = result_type(indices);
-  if (!is_integral(dtype)) {
+  if (issubdtype(dtype, inexact)) {
     throw std::invalid_argument(
         "[gather] Got indices with invalid dtype. Indices must be integral.");
   }
@@ -2413,8 +2508,8 @@ array take_along_axis(
     StreamOrDevice s /* = {} */) {
   if (axis + a.ndim() < 0 || axis >= static_cast<int>(a.ndim())) {
     std::ostringstream msg;
-    msg << "[take_along_axis] Received invalid axis "
-        << " for array with " << a.ndim() << " dimensions.";
+    msg << "[take_along_axis] Received invalid axis " << " for array with "
+        << a.ndim() << " dimensions.";
     throw std::invalid_argument(msg.str());
   }
 
@@ -2521,7 +2616,7 @@ array scatter(
 
   // Promote indices to the same type
   auto dtype = result_type(indices);
-  if (!is_integral(dtype)) {
+  if (issubdtype(dtype, inexact)) {
     throw std::invalid_argument(
         "[scatter] Got indices with invalid dtype. Indices must be integral.");
   }
@@ -2605,25 +2700,34 @@ array rsqrt(const array& a, StreamOrDevice s /* = {} */) {
 array softmax(
     const array& a,
     const std::vector<int>& axes,
+    bool precise /* = false */,
     StreamOrDevice s /* = {}*/) {
   if (axes.size() == 1 && (a.ndim() == axes[0] + 1 || axes[0] == -1)) {
     auto dtype = at_least_float(a.dtype());
     return array(
         a.shape(),
         dtype,
-        std::make_shared<Softmax>(to_stream(s)),
+        std::make_shared<Softmax>(to_stream(s), precise),
         {astype(a, dtype, s)});
   } else {
-    auto a_max = stop_gradient(max(a, axes, /*keepdims = */ true, s), s);
-    auto ex = exp(subtract(a, a_max, s), s);
-    return divide(ex, sum(ex, axes, /*keepdims = */ true, s), s);
+    auto in = a;
+    if (precise) {
+      in = astype(a, float32, s);
+    }
+    auto a_max = stop_gradient(max(in, axes, /*keepdims = */ true, s), s);
+    auto ex = exp(subtract(in, a_max, s), s);
+    return astype(
+        divide(ex, sum(ex, axes, /*keepdims = */ true, s), s), a.dtype(), s);
   }
 }
 
-array softmax(const array& a, StreamOrDevice s /* = {}*/) {
+array softmax(
+    const array& a,
+    bool precise /* = false */,
+    StreamOrDevice s /* = {}*/) {
   std::vector<int> axes(a.ndim());
   std::iota(axes.begin(), axes.end(), 0);
-  return softmax(a, axes, s);
+  return softmax(a, axes, precise, s);
 }
 
 array power(const array& a, const array& b, StreamOrDevice s /* = {} */) {
@@ -2800,15 +2904,15 @@ inline std::vector<int> conv_out_shape(
 
     if (pads_lo[i - 1] < 0 || pads_hi[i - 1] < 0) {
       std::ostringstream msg;
-      msg << "[conv] Padding sizes must be non-negative."
-          << " Got padding " << pads_lo << " | " << pads_hi << ".";
+      msg << "[conv] Padding sizes must be non-negative." << " Got padding "
+          << pads_lo << " | " << pads_hi << ".";
       throw std::invalid_argument(msg.str());
     }
 
     if (strides[i - 1] <= 0) {
       std::ostringstream msg;
-      msg << "[conv] Stride sizes must be positive."
-          << " Got strides " << strides << ".";
+      msg << "[conv] Stride sizes must be positive." << " Got strides "
+          << strides << ".";
       throw std::invalid_argument(msg.str());
     }
 
@@ -2834,7 +2938,7 @@ inline std::vector<int> conv_out_shape(
 }
 
 inline void run_conv_checks(const array& in, const array& wt, int n_dim) {
-  if (!is_floating_point(in.dtype()) && kindof(in.dtype()) != Dtype::Kind::c) {
+  if (!issubdtype(in.dtype(), floating)) {
     std::ostringstream msg;
     msg << "[conv] Invalid input array with type " << in.dtype() << "."
         << " Convolution currently only supports floating point types";
@@ -2844,8 +2948,7 @@ inline void run_conv_checks(const array& in, const array& wt, int n_dim) {
   if (in.ndim() != n_dim + 2) {
     std::ostringstream msg;
     msg << "[conv] Invalid input array with " << in.ndim() << " dimensions for "
-        << n_dim << "D convolution."
-        << " Expected an array with " << n_dim + 2
+        << n_dim << "D convolution." << " Expected an array with " << n_dim + 2
         << " dimensions following the format [N, ..., C_in].";
     throw std::invalid_argument(msg.str());
   }
@@ -2971,6 +3074,35 @@ array conv_general(
     input_dilation = std::vector<int>(spatial_dims, input_dilation_int);
   }
 
+  // Check for negative padding
+  bool has_neg_padding = false;
+  for (auto& pd : padding_lo) {
+    has_neg_padding = (pd < 0);
+  }
+  for (auto& pd : padding_hi) {
+    has_neg_padding = (pd < 0);
+  }
+
+  // Handle negative padding
+  if (has_neg_padding) {
+    std::vector<int> starts(in.ndim(), 0);
+    std::vector<int> stops = in.shape();
+
+    for (int i = 0; i < spatial_dims; i++) {
+      if (padding_lo[i] < 0) {
+        starts[i + 1] -= padding_lo[i];
+        padding_lo[i] = 0;
+      }
+
+      if (padding_hi[i] < 0) {
+        stops[i + 1] += padding_hi[i];
+        padding_hi[i] = 0;
+      }
+    }
+
+    in = slice(in, std::move(starts), std::move(stops), s);
+  }
+
   // Get output shapes
   std::vector<int> out_shape = conv_out_shape(
       in.shape(),
@@ -3005,7 +3137,6 @@ array quantized_matmul(
     int bits /* = 4 */,
     StreamOrDevice s /* = {} */) {
   array x = in_x;
-
   if (w.dtype() != uint32) {
     std::ostringstream msg;
     msg << "[quantized_matmul] The weight matrix should be uint32 "
@@ -3022,12 +3153,6 @@ array quantized_matmul(
   // Keep x's batch dimensions to reshape it back after the matmul
   auto original_shape = x.shape();
   int x_inner_dims = original_shape.back();
-  original_shape.pop_back();
-
-  // Reshape x into a matrix if it isn't already one
-  if (x.ndim() != 2) {
-    x = reshape(x, {-1, x_inner_dims}, s);
-  }
 
   if (scales.ndim() != 2 || scales.shape() != biases.shape()) {
     std::ostringstream msg;
@@ -3062,7 +3187,7 @@ array quantized_matmul(
   }
 
   auto dtype = result_type(x, scales, biases);
-  if (!is_floating_point(dtype) || is_complex(dtype)) {
+  if (!issubdtype(dtype, floating)) {
     std::ostringstream msg;
     msg << "[quantized_matmul] Only real floating types are supported but "
         << "the passed types where x.dtype() == " << x.dtype()
@@ -3070,9 +3195,10 @@ array quantized_matmul(
         << " and biases.dtype() == " << biases.dtype();
     throw std::invalid_argument(msg.str());
   }
-
-  auto out = array(
-      {x.shape(0), w_outer_dims},
+  std::vector<array> inputs;
+  original_shape.back() = w_outer_dims;
+  return array(
+      std::move(original_shape),
       dtype,
       std::make_shared<QuantizedMatmul>(
           to_stream(s), group_size, bits, transpose),
@@ -3080,14 +3206,6 @@ array quantized_matmul(
        w,
        astype(scales, dtype, s),
        astype(biases, dtype, s)});
-
-  // If needed reshape x to the original batch shape
-  if (original_shape.size() != 1) {
-    original_shape.push_back(w_outer_dims);
-    out = reshape(out, std::move(original_shape), s);
-  }
-
-  return out;
 }
 
 std::tuple<array, array, array> quantize(
@@ -3117,8 +3235,7 @@ std::tuple<array, array, array> quantize(
     std::ostringstream msg;
     msg << "[quantize] The last dimension of the matrix needs to be divisible by "
         << "the quantization group size " << group_size
-        << ". However the provided "
-        << " matrix has shape " << w.shape();
+        << ". However the provided " << " matrix has shape " << w.shape();
     throw std::invalid_argument(msg.str());
   }
 
@@ -3364,7 +3481,7 @@ array addmm(
 
   // Type promotion
   auto out_type = result_type(a, b, c);
-  if (!is_floating_point(out_type) || is_complex(out_type)) {
+  if (!issubdtype(out_type, floating)) {
     std::ostringstream msg;
     msg << "[addmm] Only real floating point types are supported but "
         << c.dtype() << ", " << a.dtype() << " and " << b.dtype()

mlx/ops.h

@@ -41,40 +41,33 @@ array linspace(
     StreamOrDevice s = {});
 
 /** Convert an array to the given data type. */
-array astype(const array& a, Dtype dtype, StreamOrDevice s = {});
+array astype(array a, Dtype dtype, StreamOrDevice s = {});
 
 /** Create a view of an array with the given shape and strides. */
 array as_strided(
-    const array& a,
+    array a,
     std::vector<int> shape,
     std::vector<size_t> strides,
     size_t offset,
     StreamOrDevice s = {});
 
 /** Copy another array. */
-array copy(const array& a, StreamOrDevice s = {});
+array copy(array a, StreamOrDevice s = {});
 
 /** Fill an array of the given shape with the given value(s). */
 array full(
-    const std::vector<int>& shape,
-    const array& vals,
+    std::vector<int> shape,
+    array vals,
     Dtype dtype,
     StreamOrDevice s = {});
-array full(
-    const std::vector<int>& shape,
-    const array& vals,
-    StreamOrDevice s = {});
+array full(std::vector<int> shape, array vals, StreamOrDevice s = {});
 template <typename T>
-array full(
-    const std::vector<int>& shape,
-    T val,
-    Dtype dtype,
-    StreamOrDevice s = {}) {
-  return full(shape, array(val, dtype), to_stream(s));
+array full(std::vector<int> shape, T val, Dtype dtype, StreamOrDevice s = {}) {
+  return full(std::move(shape), array(val, dtype), to_stream(s));
 }
 template <typename T>
-array full(const std::vector<int>& shape, T val, StreamOrDevice s = {}) {
-  return full(shape, array(val), to_stream(s));
+array full(std::vector<int> shape, T val, StreamOrDevice s = {}) {
+  return full(std::move(shape), array(val), to_stream(s));
 }
 
 /** Fill an array of the given shape with zeros. */
@@ -204,6 +197,13 @@ std::vector<array> split(
 std::vector<array>
 split(const array& a, const std::vector<int>& indices, StreamOrDevice s = {});
 
+/** A vector of coordinate arrays from coordinate vectors. */
+std::vector<array> meshgrid(
+    const std::vector<array>& arrays,
+    bool sparse = false,
+    std::string indexing = "xy",
+    StreamOrDevice s = {});
+
 /**
  * Clip (limit) the values in an array.
  */
@@ -514,13 +514,14 @@ array mean(
     bool keepdims = false,
     StreamOrDevice s = {});
 
-/** Computes the mean of the elements of an array. */
+/** Computes the variance of the elements of an array. */
 array var(const array& a, bool keepdims, int ddof = 0, StreamOrDevice s = {});
 inline array var(const array& a, StreamOrDevice s = {}) {
   return var(a, false, 0, to_stream(s));
 }
 
-/** Computes the var of the elements of an array along the given axes */
+/** Computes the variance of the elements of an array along the given
+ * axes */
 array var(
     const array& a,
     const std::vector<int>& axes,
@@ -528,7 +529,8 @@ array var(
     int ddof = 0,
     StreamOrDevice s = {});
 
-/** Computes the var of the elements of an array along the given axis */
+/** Computes the variance of the elements of an array along the given
+ * axis */
 array var(
     const array& a,
     int axis,
@@ -536,6 +538,30 @@ array var(
     int ddof = 0,
     StreamOrDevice s = {});
 
+/** Computes the standard deviation of the elements of an array. */
+array std(const array& a, bool keepdims, int ddof = 0, StreamOrDevice s = {});
+inline array std(const array& a, StreamOrDevice s = {}) {
+  return std(a, false, 0, to_stream(s));
+}
+
+/** Computes the standard deviatoin of the elements of an array along the given
+ * axes */
+array std(
+    const array& a,
+    const std::vector<int>& axes,
+    bool keepdims = false,
+    int ddof = 0,
+    StreamOrDevice s = {});
+
+/** Computes the standard deviation of the elements of an array along the given
+ * axis */
+array std(
+    const array& a,
+    int axis,
+    bool keepdims = false,
+    int ddof = 0,
+    StreamOrDevice s = {});
+
 /** The product of all elements of the array. */
 array prod(const array& a, bool keepdims, StreamOrDevice s = {});
 inline array prod(const array& a, StreamOrDevice s = {}) {
@@ -849,6 +875,9 @@ array erf(const array& a, StreamOrDevice s = {});
 /** Computes the inverse error function of the elements of an array. */
 array erfinv(const array& a, StreamOrDevice s = {});
 
+/** Computes the expm1 function of the elements of an array. */
+array expm1(const array& a, StreamOrDevice s = {});
+
 /** Stop the flow of gradients. */
 array stop_gradient(const array& a, StreamOrDevice s = {});
 
@@ -983,14 +1012,16 @@ array rsqrt(const array& a, StreamOrDevice s = {});
 array softmax(
     const array& a,
     const std::vector<int>& axes,
+    bool precise = false,
     StreamOrDevice s = {});
 
 /** Softmax of an array. */
-array softmax(const array& a, StreamOrDevice s = {});
+array softmax(const array& a, bool precise = false, StreamOrDevice s = {});
 
 /** Softmax of an array. */
-inline array softmax(const array& a, int axis, StreamOrDevice s = {}) {
-  return softmax(a, std::vector<int>{axis}, s);
+inline array
+softmax(const array& a, int axis, bool precise = false, StreamOrDevice s = {}) {
+  return softmax(a, std::vector<int>{axis}, precise, s);
 }
 
 /** Raise elements of a to the power of b element-wise */

mlx/primitives.cpp

@@ -1239,6 +1239,34 @@ std::pair<std::vector<array>, std::vector<int>> Exp::vmap(
   return {{exp(inputs[0], stream())}, axes};
 }
 
+std::vector<array> Expm1::vjp(
+    const std::vector<array>& primals,
+    const std::vector<array>& cotangents,
+    const std::vector<int>& argnums,
+    const std::vector<array>& outputs) {
+  return {multiply(
+      cotangents[0],
+      add(outputs[0], array(1.0f, outputs[0].dtype()), stream()),
+      stream())};
+}
+
+std::vector<array> Expm1::jvp(
+    const std::vector<array>& primals,
+    const std::vector<array>& tangents,
+    const std::vector<int>& argnums) {
+  assert(primals.size() == 1);
+  assert(argnums.size() == 1);
+  return {multiply(tangents[0], exp(primals[0], stream()), stream())};
+}
+
+std::pair<std::vector<array>, std::vector<int>> Expm1::vmap(
+    const std::vector<array>& inputs,
+    const std::vector<int>& axes) {
+  assert(inputs.size() == 1);
+  assert(axes.size() == 1);
+  return {{expm1(inputs[0], stream())}, axes};
+}
+
 bool FFT::is_equivalent(const Primitive& other) const {
   const FFT& r_other = static_cast<const FFT&>(other);
   return axes_ == r_other.axes_ && inverse_ == r_other.inverse_ &&
@@ -1267,7 +1295,7 @@ std::pair<std::vector<array>, std::vector<int>> FFT::vmap(
       {array(
           out_shape,
           real_ && inverse_ ? float32 : complex64,
-          std::make_unique<FFT>(stream(), fft_axes, inverse_, real_),
+          std::make_shared<FFT>(stream(), fft_axes, inverse_, real_),
           {in})},
       {ax}};
 }
@@ -1377,7 +1405,7 @@ std::pair<std::vector<array>, std::vector<int>> Full::vmap(
   assert(axes.size() == 1);
   auto& in = inputs[0];
   auto out =
-      array(in.shape(), in.dtype(), std::make_unique<Full>(stream()), {in});
+      array(in.shape(), in.dtype(), std::make_shared<Full>(stream()), {in});
   return {{out}, axes};
 }
 
@@ -1604,7 +1632,7 @@ std::pair<std::vector<array>, std::vector<int>> Log::vmap(
       {array(
           in.shape(),
           in.dtype(),
-          std::make_unique<Log>(stream(), base_),
+          std::make_shared<Log>(stream(), base_),
           {in})},
       axes};
 }
@@ -2259,7 +2287,7 @@ std::pair<std::vector<array>, std::vector<int>> RandomBits::vmap(
   auto out = array(
       shape,
       get_dtype(),
-      std::make_unique<RandomBits>(stream(), shape, width_),
+      std::make_shared<RandomBits>(stream(), shape, width_),
       {key});
   return {{out}, {kax}};
 }
@@ -2493,7 +2521,7 @@ std::pair<std::vector<array>, std::vector<int>> Scan::vmap(
       {array(
           in.shape(),
           out_dtype,
-          std::make_unique<Scan>(
+          std::make_shared<Scan>(
               stream(), reduce_type_, axis_ + axis_left, reverse_, inclusive_),
           {in})},
       axes};
@@ -2975,7 +3003,7 @@ std::pair<std::vector<array>, std::vector<int>> Softmax::vmap(
   } else {
     softmax_axes.push_back(-2);
   }
-  return {{softmax(inputs[0], softmax_axes, stream())}, axes};
+  return {{softmax(inputs[0], softmax_axes, precise_, stream())}, axes};
 }
 
 std::vector<array> Softmax::vjp(
@@ -2998,13 +3026,18 @@ std::vector<array> Softmax::jvp(
     const std::vector<int>& argnums) {
   assert(primals.size() == 1);
   assert(tangents.size() == 1);
-  auto s = softmax(primals[0], std::vector<int>{-1}, stream());
+  auto s = softmax(primals[0], std::vector<int>{-1}, precise_, stream());
   auto sv = multiply(s, tangents[0], stream());
   return {subtract(
       sv,
       multiply(s, sum(sv, std::vector<int>{-1}, true, stream()), stream()))};
 }
 
+bool Softmax::is_equivalent(const Primitive& other) const {
+  const Softmax& s_other = static_cast<const Softmax&>(other);
+  return precise_ == s_other.precise_;
+}
+
 std::pair<std::vector<array>, std::vector<int>> Sort::vmap(
     const std::vector<array>& inputs,
     const std::vector<int>& axes) {
@@ -3303,7 +3336,7 @@ std::pair<std::vector<array>, std::vector<int>> NumberOfElements::vmap(
   array out = array(
       std::vector<int>{},
       dtype_,
-      std::make_unique<NumberOfElements>(stream(), new_axes, inverted_, dtype_),
+      std::make_shared<NumberOfElements>(stream(), new_axes, inverted_, dtype_),
       inputs);
 
   return {{out}, {-1}};