From dcb8319f3d6eed77387dc7dda8af64dc32f469c5 Mon Sep 17 00:00:00 2001
From: Awni Hannun <awni@apple.com>
Date: Fri, 25 Jul 2025 12:13:19 -0700
Subject: [PATCH] update install docs and requirements (#2419)

---
 README.md                      | 21 +++++++++++++--------
 docs/src/install.rst           | 22 +++++++++++++++++++---
 python/scripts/repair_linux.sh |  1 +
 3 files changed, 33 insertions(+), 11 deletions(-)

diff --git a/README.md b/README.md
index 3666509f2..87dcfd18e 100644
--- a/README.md
+++ b/README.md
@@ -11,10 +11,10 @@ brought to you by Apple machine learning research.
 
 Some key features of MLX include:
 
- - **Familiar APIs**: MLX has a Python API that closely follows NumPy.  MLX
+ - **Familiar APIs**: MLX has a Python API that closely follows NumPy. MLX
    also has fully featured C++, [C](https://github.com/ml-explore/mlx-c), and
    [Swift](https://github.com/ml-explore/mlx-swift/) APIs, which closely mirror
-   the Python API.  MLX has higher-level packages like `mlx.nn` and
+   the Python API. MLX has higher-level packages like `mlx.nn` and
    `mlx.optimizers` with APIs that closely follow PyTorch to simplify building
    more complex models.
 
@@ -68,18 +68,23 @@ in the documentation.
 
 ## Installation
 
-MLX is available on [PyPI](https://pypi.org/project/mlx/). To install the Python API, run:
+MLX is available on [PyPI](https://pypi.org/project/mlx/). To install MLX on
+macOS, run:
 
-**With `pip`**:
-
-```
+```bash
 pip install mlx
 ```
 
-**With `conda`**:
+To install the CUDA backend on Linux, run:
 
+```bash
+pip install "mlx[cuda]"
 ```
-conda install -c conda-forge mlx
+
+To install a CPU-only Linux package, run:
+
+```bash
+pip install "mlx[cpu]"
 ```
 
 Checkout the
diff --git a/docs/src/install.rst b/docs/src/install.rst
index 70491ac64..268141567 100644
--- a/docs/src/install.rst
+++ b/docs/src/install.rst
@@ -13,7 +13,7 @@ silicon computer is
 
     pip install mlx
 
-To install from PyPI you must meet the following requirements:
+To install from PyPI your system must meet the following requirements:
 
 - Using an M series chip (Apple silicon)
 - Using a native Python >= 3.9
@@ -26,13 +26,22 @@ To install from PyPI you must meet the following requirements:
 CUDA
 ^^^^
 
-MLX has a CUDA backend which you can use on any Linux platform with CUDA 12
-and SM 7.0 (Volta) and up. To install MLX with CUDA support, run:
+MLX has a CUDA backend which you can install with:
 
 .. code-block:: shell
 
     pip install "mlx[cuda]"
 
+To install the CUDA package from PyPi your system must meet the following
+requirements:
+
+- Nvidia architecture >= SM 7.0 (Volta)
+- Nvidia driver >= 550.54.14
+- CUDA toolkit >= 12.0
+- Linux distribution with glibc >= 2.35
+- Python >= 3.9
+
+
 CPU-only (Linux)
 ^^^^^^^^^^^^^^^^
 
@@ -42,6 +51,13 @@ For a CPU-only version of MLX that runs on Linux use:
 
     pip install "mlx[cpu]"
 
+To install the CPU-only package from PyPi your system must meet the following
+requirements:
+
+- Linux distribution with glibc >= 2.35
+- Python >= 3.9
+
+
 Troubleshooting
 ^^^^^^^^^^^^^^^
 
diff --git a/python/scripts/repair_linux.sh b/python/scripts/repair_linux.sh
index 82cf49060..880c43f4b 100644
--- a/python/scripts/repair_linux.sh
+++ b/python/scripts/repair_linux.sh
@@ -2,6 +2,7 @@
 
 auditwheel repair dist/* \
   --plat manylinux_2_35_x86_64 \
+  --only-plat \
   --exclude libmlx* \
   -w wheel_tmp