-
+
+
@@ -370,6 +370,7 @@
right_shift()operator>>()view()roll()roll()roll()roll()roll()roll()Roll elements along an axis and introduce them on the other side.
+right_shift()operator>>()view()roll()roll()roll()roll()roll()roll()
-
+
+
@@ -369,6 +369,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -367,6 +367,7 @@
-
+
+
@@ -369,6 +369,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
A device to run operations on.
Methods
diff --git a/docs/build/html/python/_autosummary/mlx.core.Dtype.html b/docs/build/html/python/_autosummary/mlx.core.Dtype.html index 7cdee3468..c4f1514b1 100644 --- a/docs/build/html/python/_autosummary/mlx.core.Dtype.html +++ b/docs/build/html/python/_autosummary/mlx.core.Dtype.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Element-wise absolute value.
-
+
+
@@ -370,6 +370,7 @@
Element-wise addition.
Add two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.addmm.html b/docs/build/html/python/_autosummary/mlx.core.addmm.html index d436862ac..f8b301efe 100644 --- a/docs/build/html/python/_autosummary/mlx.core.addmm.html +++ b/docs/build/html/python/_autosummary/mlx.core.addmm.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Matrix multiplication with addition and optional scaling.
Perform the (possibly batched) matrix multiplication of two arrays and add to the result with optional scaling factors.
@@ -881,9 +883,9 @@ with optional scaling factors.c (array) – Input array or scalar.
a (array) – Input array or scalar.
b (array) – Input array or scalar.
alpha (float, optional) – Scaling factor for the +
alpha (float, optional) – Scaling factor for the
matrix product of a and b (default: 1)
beta (float, optional) – Scaling factor for c (default: 1)
beta (float, optional) – Scaling factor for c (default: 1)
-
+
+
@@ -370,6 +370,7 @@
An and reduction over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Approximate comparison of two arrays.
Infinite values are considered equal if they have the same sign, NaN values are not equal unless equal_nan is True.
The arrays are considered equal if:
@@ -885,9 +887,9 @@ broadcasting.a (array) – Input array.
b (array) – Input array.
rtol (float) – Relative tolerance.
atol (float) – Absolute tolerance.
equal_nan (bool) – If True, NaNs are considered equal.
+
rtol (float) – Relative tolerance.
atol (float) – Absolute tolerance.
equal_nan (bool) – If True, NaNs are considered equal.
Defaults to False.
-
+
+
@@ -370,6 +370,7 @@
An or reduction over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Generates ranges of numbers.
+arange(stop: int | float, step: None | int | float = None, dtype: Dtype | None = None, *, stream: None | Stream | Device = None) array +Overloaded function.
+arange(start : Union[int, float], stop : Union[int, float], step : Union[None, int, float], dtype: Optional[Dtype] = None, *, stream: Union[None, Stream, Device] = None) -> array
++Generates ranges of numbers.
Generate numbers in the half-open interval
-[start, stop)in increments ofstep.-
+- Parameters:
-- +
-
+- -
start (float or int, optional) – Starting value which defaults to
0.- -
- -
step (float or int, optional) – Increment which defaults to
1.- -
dtype (Dtype, optional) – Specifies the data type of the output. -If unspecified will default to
float32if any ofstart, -stop, orsteparefloat. Otherwise will default to -int32.+
-- Args:
- -
start (float or int, optional): Starting value which defaults to
0. +stop (float or int): Stopping value. +step (float or int, optional): Increment which defaults to1. +dtype (Dtype, optional): Specifies the data type of the output. If unspecified will default tofloat32if any ofstart,stop, orsteparefloat. Otherwise will default toint32.- Returns:
-The range of values.
+- Returns:
- -
array: The range of values.
- Return type:
-- -
--+Note
-Following the Numpy convention the actual increment used to +
- Note:
Following the Numpy convention the actual increment used to generate numbers is
-dtype(start + step) - dtype(start). This can lead to unexpected results for example if start + step is a fractional value and the dtype is integral.
arange(stop : Union[int, float], step : Union[None, int, float] = None, dtype: Optional[Dtype] = None, *, stream: Union[None, Stream, Device] = None) -> array
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse cosine.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse hyperbolic cosine.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse sine.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse hyperbolic sine.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse tangent.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse tangent of the ratio of two arrays.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse hyperbolic tangent.
-
+
+
@@ -370,6 +370,7 @@
Indices of the maximum values along the axis.
a (array) – Input array.
axis (int, optional) – Optional axis to reduce over. If unspecified +
axis (int, optional) – Optional axis to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Indices of the minimum values along the axis.
a (array) – Input array.
axis (int, optional) – Optional axis to reduce over. If unspecified +
axis (int, optional) – Optional axis to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Returns the indices that partition the array.
The ordering of the elements within a partition in given by the indices is undefined.
@@ -879,12 +881,12 @@ is undefined.a (array) – Input array.
kth (int) – Element index at the kth position in the output will
+
kth (int) – Element index at the kth position in the output will
give the sorted position. All indices before the kth position
will be of elements less or equal to the element at the kth
index and all indices after will be of elements greater or equal
to the element at the kth index.
axis (int or None, optional) – Optional axis to partition over. +
axis (int or None, optional) – Optional axis to partition over.
If None, this partitions over the flattened array.
If unspecified, it defaults to -1.
-
+
+
@@ -370,6 +370,7 @@
Returns the indices that sort the array.
a (array) – Input array.
axis (int or None, optional) – Optional axis to sort over. +
axis (int or None, optional) – Optional axis to sort over.
If None, this sorts over the flattened array.
If unspecified, it defaults to -1 (sorting over the last axis).
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
See abs().
-
+
+
@@ -370,6 +370,7 @@
See all().
-
+
+
@@ -370,6 +370,7 @@
See any().
-
+
+
@@ -370,6 +370,7 @@
See argmax().
-
+
+
@@ -370,6 +370,7 @@
See argmin().
-
+
+
@@ -370,6 +370,7 @@
Cast the array to a specified type.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
See conj().
-
+
+
@@ -370,6 +370,7 @@
See cos().
-
+
+
@@ -370,6 +370,7 @@
See cummax().
-
+
+
@@ -370,6 +370,7 @@
See cummin().
-
+
+
@@ -370,6 +370,7 @@
See cumprod().
-
+
+
@@ -370,6 +370,7 @@
See cumsum().
-
+
+
@@ -370,6 +370,7 @@
Extract a diagonal or construct a diagonal matrix.
-
+
+
@@ -370,6 +370,7 @@
See diagonal().
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
See exp().
-
+
+
@@ -370,6 +370,7 @@
See flatten().
-
+
+
@@ -370,6 +370,7 @@
An N-dimensional array object.
Methods
@@ -972,7 +974,7 @@ document.write(`See reciprocal().
reshape(self, *shape[, stream])
Equivalent to reshape() but the shape can be passed either as a tuple or as separate arguments.
Equivalent to reshape() but the shape can be passed either as a tuple or as separate arguments.
round(self[, decimals, stream])
See round().
See swapaxes().
tolist(self)
Convert the array to a Python list.
Convert the array to a Python list.
transpose(self, *axes[, stream])
Equivalent to transpose() but the axes can be passed either as a tuple or as separate arguments.
-
+
+
@@ -370,6 +370,7 @@
Access the value of a scalar array.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
See log().
-
+
+
@@ -370,6 +370,7 @@
See log10().
-
+
+
@@ -370,6 +370,7 @@
See log1p().
-
+
+
@@ -370,6 +370,7 @@
See log2().
-
+
+
@@ -370,6 +370,7 @@
See logsumexp().
-
+
+
@@ -370,6 +370,7 @@
See max().
-
+
+
@@ -370,6 +370,7 @@
See mean().
-
+
+
@@ -370,6 +370,7 @@
See min().
-
+
+
@@ -370,6 +370,7 @@
See moveaxis().
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
See prod().
-
+
+
@@ -370,6 +370,7 @@
See reciprocal().
-
+
+
@@ -370,6 +370,7 @@
Equivalent to reshape() but the shape can be passed either as a
-tuple or as separate arguments.
tuple or as separate arguments.
See reshape() for full documentation.
-
+
+
@@ -370,6 +370,7 @@
See round().
-
+
+
@@ -370,6 +370,7 @@
See rsqrt().
-
+
+
@@ -370,6 +370,7 @@
A tuple containing the sizes of each dimension.
-
+
+
@@ -370,6 +370,7 @@
See sin().
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
See split().
-
+
+
@@ -370,6 +370,7 @@
See sqrt().
-
+
+
@@ -370,6 +370,7 @@
See square().
-
+
+
@@ -370,6 +370,7 @@
See squeeze().
-
+
+
@@ -370,6 +370,7 @@
See std().
-
+
+
@@ -370,6 +370,7 @@
See sum().
-
+
+
@@ -370,6 +370,7 @@
See swapaxes().
-
+
+
@@ -370,6 +370,7 @@
Convert the array to a Python list.
Convert the array to a Python list.
-
+
+
@@ -370,6 +370,7 @@
Equivalent to transpose() but the axes can be passed either as
a tuple or as separate arguments.
See transpose() for full documentation.
-
+
+
@@ -370,6 +370,7 @@
See var().
-
+
+
@@ -370,6 +370,7 @@
See view().
-
+
+
@@ -370,6 +370,7 @@
Array equality check.
Compare two arrays for equality. Returns True if and only if the arrays
have the same shape and their values are equal. The arrays need not have
@@ -881,7 +883,7 @@ the same type to be considered equal.
-
+
+
@@ -370,6 +370,7 @@
Create a view into the array with the given shape and strides.
The resulting array will always be as if the provided array was row contiguous regardless of the provided arrays storage order and current @@ -887,12 +889,12 @@ result into crashes.
a (array) – Input array
shape (list(int), optional) – The shape of the resulting array. If +
shape (list(int), optional) – The shape of the resulting array. If
None it defaults to a.shape().
strides (list(int), optional) – The strides of the resulting array. If +
strides (list(int), optional) – The strides of the resulting array. If
None it defaults to the reverse exclusive cumulative product of
a.shape().
offset (int) – Skip that many elements from the beginning of the input +
offset (int) – Skip that many elements from the beginning of the input array.
-
+
+
@@ -370,6 +370,7 @@
Convert all arrays to have at least one dimension.
-
+
+
@@ -370,6 +370,7 @@
Convert all arrays to have at least two dimensions.
-
+
+
@@ -370,6 +370,7 @@
Convert all arrays to have at least three dimensions.
-
+
+
@@ -370,6 +370,7 @@
Element-wise bitwise and.
Take the bitwise and of two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.bitwise_or.html b/docs/build/html/python/_autosummary/mlx.core.bitwise_or.html index f23a06561..38ef639a7 100644 --- a/docs/build/html/python/_autosummary/mlx.core.bitwise_or.html +++ b/docs/build/html/python/_autosummary/mlx.core.bitwise_or.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise bitwise or.
Take the bitwise or of two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.bitwise_xor.html b/docs/build/html/python/_autosummary/mlx.core.bitwise_xor.html index e299b8a1e..6a6547f73 100644 --- a/docs/build/html/python/_autosummary/mlx.core.bitwise_xor.html +++ b/docs/build/html/python/_autosummary/mlx.core.bitwise_xor.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise bitwise xor.
Take the bitwise exclusive or of two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be diff --git a/docs/build/html/python/_autosummary/mlx.core.block_masked_mm.html b/docs/build/html/python/_autosummary/mlx.core.block_masked_mm.html index 4dc4b70fc..909acc9a1 100644 --- a/docs/build/html/python/_autosummary/mlx.core.block_masked_mm.html +++ b/docs/build/html/python/_autosummary/mlx.core.block_masked_mm.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Matrix multiplication with block masking.
Perform the (possibly batched) matrix multiplication of two arrays and with blocks
of size block_size x block_size optionally masked out.
block_size<
a (array) – Input array or scalar.
b (array) – Input array or scalar.
block_size (int) – Size of blocks to be masked. Must be 32 or 64. Default: 64.
block_size (int) – Size of blocks to be masked. Must be 32 or 64. Default: 64.
mask_out (array, optional) – Mask for output. Default: None.
mask_lhs (array, optional) – Mask for a. Default: None.
mask_rhs (array, optional) – Mask for b. Default: None.
mlx.core.broadcast_to — MLX 0.18.0 documentation
+ mlx.core.broadcast_to — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
Broadcast an array to the given shape.
The broadcasting semantics are the same as Numpy.
-
+
+
@@ -370,6 +370,7 @@
Element-wise ceil.
-
+
+
@@ -370,6 +370,7 @@
Clip the values of the array between the given minimum and maximum.
If either a_min or a_max are None, then corresponding edge
is ignored. At least one of a_min and a_max cannot be None.
diff --git a/docs/build/html/python/_autosummary/mlx.core.compile.html b/docs/build/html/python/_autosummary/mlx.core.compile.html
index 345fe18b8..b28ccd89b 100644
--- a/docs/build/html/python/_autosummary/mlx.core.compile.html
+++ b/docs/build/html/python/_autosummary/mlx.core.compile.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Returns a compiled function which produces the same output as fun.
fun (Callable) – A function which takes a variable number of
array or trees of array and returns
a variable number of array or trees of array.
inputs (list or dict, optional) – These inputs will be captured during +
inputs (list or dict, optional) – These inputs will be captured during
the function compilation along with the inputs to fun. The inputs
-can be a list or a dict containing arbitrarily nested
+can be a list or a dict containing arbitrarily nested
lists, dictionaries, or arrays. Leaf nodes that are not
array are ignored. Default: None
outputs (list or dict, optional) – These outputs will be captured and +
outputs (list or dict, optional) – These outputs will be captured and
updated in a compiled function. The outputs can be a
-list or a dict containing arbitrarily nested lists,
+list or a dict containing arbitrarily nested lists,
dictionaries, or arrays. Leaf nodes that are not array are ignored.
Default: None
shapeless (bool, optional) – A function compiled with the shapeless
+
shapeless (bool, optional) – A function compiled with the shapeless
option enabled will not be recompiled when the input shape changes. Not all
functions can be compiled with shapeless enabled. Attempting to compile
such functions with shapeless enabled will throw. Note, changing the number
diff --git a/docs/build/html/python/_autosummary/mlx.core.concatenate.html b/docs/build/html/python/_autosummary/mlx.core.concatenate.html
index 213bbb104..b4836e3a5 100644
--- a/docs/build/html/python/_autosummary/mlx.core.concatenate.html
+++ b/docs/build/html/python/_autosummary/mlx.core.concatenate.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Concatenate the arrays along the given axis.
-
+
+
@@ -370,6 +370,7 @@
Return the elementwise complex conjugate of the input. Alias for mx.conjugate.
-
+
+
@@ -370,6 +370,7 @@
Return the elementwise complex conjugate of the input. Alias for mx.conj.
-
+
+
@@ -370,6 +370,7 @@
1D convolution over an input with several channels
input (array) – Input array of shape (N, H, C_in).
weight (array) – Weight array of shape (C_out, H, C_in).
stride (int, optional) – Kernel stride. Default: 1.
padding (int, optional) – Input padding. Default: 0.
dilation (int, optional) – Kernel dilation. Default: 1.
groups (int, optional) – Input feature groups. Default: 1.
stride (int, optional) – Kernel stride. Default: 1.
padding (int, optional) – Input padding. Default: 0.
dilation (int, optional) – Kernel dilation. Default: 1.
groups (int, optional) – Input feature groups. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
2D convolution over an input with several channels
input (array) – Input array of shape (N, H, W, C_in).
weight (array) – Weight array of shape (C_out, H, W, C_in).
stride (int or tuple(int), optional) – tuple of size 2 with
+
stride (int or tuple(int), optional) – tuple of size 2 with
kernel strides. All spatial dimensions get the same stride if
only one number is specified. Default: 1.
padding (int or tuple(int), optional) – tuple of size 2 with
+
padding (int or tuple(int), optional) – tuple of size 2 with
symmetric input padding. All spatial dimensions get the same
padding if only one number is specified. Default: 0.
dilation (int or tuple(int), optional) – tuple of size 2 with
+
dilation (int or tuple(int), optional) – tuple of size 2 with
kernel dilation. All spatial dimensions get the same dilation
if only one number is specified. Default: 1
groups (int, optional) – input feature groups. Default: 1.
groups (int, optional) – input feature groups. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
3D convolution over an input with several channels
Note: Only the default groups=1 is currently supported.
input (array) – Input array of shape (N, D, H, W, C_in).
weight (array) – Weight array of shape (C_out, D, H, W, C_in).
stride (int or tuple(int), optional) – tuple of size 3 with
+
stride (int or tuple(int), optional) – tuple of size 3 with
kernel strides. All spatial dimensions get the same stride if
only one number is specified. Default: 1.
padding (int or tuple(int), optional) – tuple of size 3 with
+
padding (int or tuple(int), optional) – tuple of size 3 with
symmetric input padding. All spatial dimensions get the same
padding if only one number is specified. Default: 0.
dilation (int or tuple(int), optional) – tuple of size 3 with
+
dilation (int or tuple(int), optional) – tuple of size 3 with
kernel dilation. All spatial dimensions get the same dilation
if only one number is specified. Default: 1
groups (int, optional) – input feature groups. Default: 1.
groups (int, optional) – input feature groups. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
General convolution over an input with several channels
input (array) – Input array of shape (N, ..., C_in).
weight (array) – Weight array of shape (C_out, ..., C_in).
stride (int or list(int), optional) – list with kernel strides.
+
stride (int or list(int), optional) – list with kernel strides.
All spatial dimensions get the same stride if
only one number is specified. Default: 1.
padding (int, list(int), or tuple(list(int), list(int)), optional) – list with input padding. All spatial dimensions get the same
+
padding (int, list(int), or tuple(list(int), list(int)), optional) – list with input padding. All spatial dimensions get the same
padding if only one number is specified. Default: 0.
kernel_dilation (int or list(int), optional) – list with
kernel dilation. All spatial dimensions get the same dilation
if only one number is specified. Default: 1
input_dilation (int or list(int), optional) – list with
input dilation. All spatial dimensions get the same dilation
if only one number is specified. Default: 1
groups (int, optional) – Input feature groups. Default: 1.
flip (bool, optional) – Flip the order in which the spatial dimensions of +
groups (int, optional) – Input feature groups. Default: 1.
flip (bool, optional) – Flip the order in which the spatial dimensions of
the weights are processed. Performs the cross-correlation operator when
flip is False and the convolution operator otherwise.
Default: False.
-
+
+
@@ -370,6 +370,7 @@
1D transposed convolution over an input with several channels
input (array) – Input array of shape (N, H, C_in).
weight (array) – Weight array of shape (C_out, H, C_in).
stride (int, optional) – Kernel stride. Default: 1.
padding (int, optional) – Input padding. Default: 0.
dilation (int, optional) – Kernel dilation. Default: 1.
groups (int, optional) – Input feature groups. Default: 1.
stride (int, optional) – Kernel stride. Default: 1.
padding (int, optional) – Input padding. Default: 0.
dilation (int, optional) – Kernel dilation. Default: 1.
groups (int, optional) – Input feature groups. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
2D transposed convolution over an input with several channels
Note: Only the default groups=1 is currently supported.
input (array) – Input array of shape (N, H, W, C_in).
weight (array) – Weight array of shape (C_out, H, W, C_in).
stride (int or tuple(int), optional) – tuple of size 2 with
+
stride (int or tuple(int), optional) – tuple of size 2 with
kernel strides. All spatial dimensions get the same stride if
only one number is specified. Default: 1.
padding (int or tuple(int), optional) – tuple of size 2 with
+
padding (int or tuple(int), optional) – tuple of size 2 with
symmetric input padding. All spatial dimensions get the same
padding if only one number is specified. Default: 0.
dilation (int or tuple(int), optional) – tuple of size 2 with
+
dilation (int or tuple(int), optional) – tuple of size 2 with
kernel dilation. All spatial dimensions get the same dilation
if only one number is specified. Default: 1
groups (int, optional) – input feature groups. Default: 1.
groups (int, optional) – input feature groups. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
3D transposed convolution over an input with several channels
Note: Only the default groups=1 is currently supported.
input (array) – Input array of shape (N, D, H, W, C_in).
weight (array) – Weight array of shape (C_out, D, H, W, C_in).
stride (int or tuple(int), optional) – tuple of size 3 with
+
stride (int or tuple(int), optional) – tuple of size 3 with
kernel strides. All spatial dimensions get the same stride if
only one number is specified. Default: 1.
padding (int or tuple(int), optional) – tuple of size 3 with
+
padding (int or tuple(int), optional) – tuple of size 3 with
symmetric input padding. All spatial dimensions get the same
padding if only one number is specified. Default: 0.
dilation (int or tuple(int), optional) – tuple of size 3 with
+
dilation (int or tuple(int), optional) – tuple of size 3 with
kernel dilation. All spatial dimensions get the same dilation
if only one number is specified. Default: 1
groups (int, optional) – input feature groups. Default: 1.
groups (int, optional) – input feature groups. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
The discrete convolution of 1D arrays.
If v is longer than a, then they are swapped.
The conv filter is flipped following signal processing convention.
-
+
+
@@ -370,6 +370,7 @@
Element-wise cosine.
-
+
+
@@ -370,6 +370,7 @@
Element-wise hyperbolic cosine.
-
+
+
@@ -370,6 +370,7 @@
Return the cumulative maximum of the elements along the given axis.
a (array) – Input array
axis (int, optional) – Optional axis to compute the cumulative maximum +
axis (int, optional) – Optional axis to compute the cumulative maximum over. If unspecified the cumulative maximum of the flattened array is returned.
reverse (bool) – Perform the cumulative maximum in reverse.
inclusive (bool) – The i-th element of the output includes the i-th +
reverse (bool) – Perform the cumulative maximum in reverse.
inclusive (bool) – The i-th element of the output includes the i-th element of the input.
-
+
+
@@ -370,6 +370,7 @@
Return the cumulative minimum of the elements along the given axis.
a (array) – Input array
axis (int, optional) – Optional axis to compute the cumulative minimum +
axis (int, optional) – Optional axis to compute the cumulative minimum over. If unspecified the cumulative minimum of the flattened array is returned.
reverse (bool) – Perform the cumulative minimum in reverse.
inclusive (bool) – The i-th element of the output includes the i-th +
reverse (bool) – Perform the cumulative minimum in reverse.
inclusive (bool) – The i-th element of the output includes the i-th element of the input.
-
+
+
@@ -370,6 +370,7 @@
Return the cumulative product of the elements along the given axis.
a (array) – Input array
axis (int, optional) – Optional axis to compute the cumulative product +
axis (int, optional) – Optional axis to compute the cumulative product over. If unspecified the cumulative product of the flattened array is returned.
reverse (bool) – Perform the cumulative product in reverse.
inclusive (bool) – The i-th element of the output includes the i-th +
reverse (bool) – Perform the cumulative product in reverse.
inclusive (bool) – The i-th element of the output includes the i-th element of the input.
-
+
+
@@ -370,6 +370,7 @@
Return the cumulative sum of the elements along the given axis.
a (array) – Input array
axis (int, optional) – Optional axis to compute the cumulative sum +
axis (int, optional) – Optional axis to compute the cumulative sum over. If unspecified the cumulative sum of the flattened array is returned.
reverse (bool) – Perform the cumulative sum in reverse.
inclusive (bool) – The i-th element of the output includes the i-th +
reverse (bool) – Perform the cumulative sum in reverse.
inclusive (bool) – The i-th element of the output includes the i-th element of the input.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Convert angles from radians to degrees.
-
+
+
@@ -371,6 +371,7 @@
Dequantize the matrix w using the provided scales and
biases and the group_size and bits configuration.
Formally, given the notation in quantize(), we compute
@@ -886,9 +888,9 @@ document.write(`
w (array) – Matrix to be quantized
scales (array) – The scales to use per group_size elements of w
biases (array) – The biases to use per group_size elements of w
group_size (int, optional) – The size of the group in w that shares a
+
group_size (int, optional) – The size of the group in w that shares a
scale and bias. Default: 64.
bits (int, optional) – The number of bits occupied by each element in +
bits (int, optional) – The number of bits occupied by each element in
w. Default: 4.
-
+
+
@@ -371,6 +371,7 @@
Extract a diagonal or construct a diagonal matrix.
If a is 1-D then a diagonal matrix is constructed with a on the
\(k\)-th diagonal. If a is 2-D then the \(k\)-th diagonal is
@@ -881,7 +883,7 @@ returned.
-
+
+
@@ -370,6 +370,7 @@
Return specified diagonals.
If a is 2-D, then a 1-D array containing the diagonal at the given
offset is returned.
a (array) – Input array
offset (int, optional) – Offset of the diagonal from the main diagonal. +
offset (int, optional) – Offset of the diagonal from the main diagonal.
Can be positive or negative. Default: 0.
axis1 (int, optional) – The first axis of the 2-D sub-arrays from which +
axis1 (int, optional) – The first axis of the 2-D sub-arrays from which
the diagonals should be taken. Default: 0.
axis2 (int, optional) – The second axis of the 2-D sub-arrays from which +
axis2 (int, optional) – The second axis of the 2-D sub-arrays from which
the diagonals should be taken. Default: 1.
-
+
+
@@ -370,6 +370,7 @@
Globally disable compilation. Setting the environment variable
MLX_DISABLE_COMPILE can also be used to disable compilation.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Gather arrays from all processes.
Gather the x arrays from all processes in the group and concatenate
them along the first axis. The arrays should all have the same shape.
-
+
+
@@ -370,6 +370,7 @@
All reduce sum.
Sum the x arrays from all processes in the group.
-
+
+
@@ -370,6 +370,7 @@
Initialize the communication backend and create the global communication group.
strict (bool, optional) – If set to False it returns a singleton group +
strict (bool, optional) – If set to False it returns a singleton group
in case mx.distributed.is_available() returns False otherwise
it throws a runtime error. Default: False
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Recv an array with shape shape and dtype dtype from process
with rank src.
shape (Tuple[int]) – The shape of the array we are receiving.
shape (Tuple[int]) – The shape of the array we are receiving.
dtype (Dtype) – The data type of the array we are receiving.
src (int) – Rank of the source process in the group.
src (int) – Rank of the source process in the group.
group (Group) – The group of processes that will participate in the
recv. If set to None the global group is used. Default:
None.
-
+
+
@@ -370,6 +370,7 @@
Recv an array with shape and type like x from process with rank
src.
It is equivalent to calling mx.distributed.recv(x.shape, x.dtype, src).
x (array) – An array defining the shape and dtype of the array we are receiving.
src (int) – Rank of the source process in the group.
src (int) – Rank of the source process in the group.
group (Group) – The group of processes that will participate in the
recv. If set to None the global group is used. Default:
None.
-
+
+
@@ -370,6 +370,7 @@
Send an array from the current process to the process that has rank
dst in the group.
x (array) – Input array.
dst (int) – Rank of the destination process in the group.
dst (int) – Rank of the destination process in the group.
group (Group) – The group of processes that will participate in the
sned. If set to None the global group is used. Default:
None.
-
+
+
@@ -370,6 +370,7 @@
Element-wise division.
Divide two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.divmod.html b/docs/build/html/python/_autosummary/mlx.core.divmod.html index 317bcd10c..fc85bdafd 100644 --- a/docs/build/html/python/_autosummary/mlx.core.divmod.html +++ b/docs/build/html/python/_autosummary/mlx.core.divmod.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise quotient and remainder.
The fuction divmod(a, b) is equivalent to but faster than
(a // b, a % b). The function uses numpy-style broadcasting
@@ -887,7 +889,7 @@ semantics. Either or both input arrays can also be scalars.
The quotient a // b and remainder a % b.
-
+
+
@@ -370,6 +370,7 @@
Perform the Einstein summation convention on the operands.
-
+
+
@@ -370,6 +370,7 @@
Compute the contraction order for the given Einstein summation.
tuple(list(tuple(int, int)), str)
+
-
+
+
@@ -370,6 +370,7 @@
Globally enable compilation. This will override the environment
variable MLX_DISABLE_COMPILE if set.
-
+
+
@@ -370,6 +370,7 @@
Element-wise equality.
Equality comparison on two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.erf.html b/docs/build/html/python/_autosummary/mlx.core.erf.html index 62b8dd14d..d99d3bc00 100644 --- a/docs/build/html/python/_autosummary/mlx.core.erf.html +++ b/docs/build/html/python/_autosummary/mlx.core.erf.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Element-wise error function.
-
+
+
@@ -370,6 +370,7 @@
Element-wise inverse of erf().
-
+
+
@@ -370,6 +370,7 @@
Evaluate an array or tree of array.
*args (arrays or trees of arrays) – Each argument can be a single array
or a tree of arrays. If a tree is given the nodes can be a Python
-list, tuple or dict. Leaves which are not
+list, tuple or dict. Leaves which are not
arrays are ignored.
-
+
+
@@ -370,6 +370,7 @@
Element-wise exponential.
-
+
+
@@ -370,6 +370,7 @@
Add a size one dimension at the given axis.
-
+
+
@@ -370,6 +370,7 @@
Element-wise exponential minus 1.
Computes exp(x) - 1 with greater precision for small x.
-
+
+
@@ -370,6 +370,7 @@
Create an identity matrix or a general diagonal matrix.
n (int) – The number of rows in the output.
m (int, optional) – The number of columns in the output. Defaults to n.
k (int, optional) – Index of the diagonal. Defaults to 0 (main diagonal).
n (int) – The number of rows in the output.
m (int, optional) – The number of columns in the output. Defaults to n.
k (int, optional) – Index of the diagonal. Defaults to 0 (main diagonal).
dtype (Dtype, optional) – Data type of the output array. Defaults to float32.
stream (Stream, optional) – Stream or device. Defaults to None.
-
+
+
@@ -371,6 +371,7 @@
Quantize the matrix w using the provided scales and
biases and the group_size and bits configuration.
Formally, given the notation in quantize(), we compute
@@ -886,9 +888,9 @@ document.write(`
w (array) – Matrix to be quantize
scales (array) – The scales to use per group_size elements of w
biases (array) – The biases to use per group_size elements of w
group_size (int, optional) – The size of the group in w that shares a
+
group_size (int, optional) – The size of the group in w that shares a
scale and bias. (default: 64)
bits (int, optional) – The number of bits occupied by each element in +
bits (int, optional) – The number of bits occupied by each element in
w. (default: 4)
-
+
+
@@ -370,6 +370,7 @@
Layer normalization.
The normalization is with respect to the last axis of the input x.
bias (array, optional) – An additive offset to be added to the result.
The bias should be one-dimensional with the same size
as the last axis of x. If set to None then no translation happens.
eps (float) – A small additive constant for numerical stability.
eps (float) – A small additive constant for numerical stability.
-
+
+
@@ -370,6 +370,7 @@
A jit-compiled custom Metal kernel defined from a source string.
name (str) – Name for the kernel.
input_names (List[str]) – The parameter names of the inputs in the +
name (str) – Name for the kernel.
input_names (List[str]) – The parameter names of the inputs in the function signature.
output_names (List[str]) – The parameter names of the outputs in the +
output_names (List[str]) – The parameter names of the outputs in the function signature.
source (str) – Source code. This is the body of a function in Metal, +
source (str) – Source code. This is the body of a function in Metal, the function signature will be automatically generated.
header (str) – Header source code to include before the main function. +
header (str) – Header source code to include before the main function. Useful for helper functions or includes that should live outside of the main function body.
ensure_row_contiguous (bool) – Whether to ensure the inputs are row contiguous +
ensure_row_contiguous (bool) – Whether to ensure the inputs are row contiguous
before the kernel runs. Default: True.
atomic_outputs (bool) – Whether to use atomic outputs in the function signature +
atomic_outputs (bool) – Whether to use atomic outputs in the function signature
e.g. device atomic<float>. Default: False.
-
+
+
@@ -370,6 +370,7 @@
Root Mean Square normalization (RMS norm).
The normalization is with respect to the last axis of the input x.
weight (array) – A multiplicative weight to scale the result by.
The weight should be one-dimensional with the same size
as the last axis of x.
eps (float) – A small additive constant for numerical stability.
eps (float) – A small additive constant for numerical stability.
-
+
+
@@ -370,6 +370,7 @@
Apply rotary positional encoding to the input.
a (array) – Input array.
dims (int) – The feature dimensions to be rotated. If the input feature +
dims (int) – The feature dimensions to be rotated. If the input feature is larger than dims then the rest is left unchanged.
traditional (bool) – If set to True choose the traditional
+
traditional (bool) – If set to True choose the traditional
implementation which rotates consecutive dimensions.
base (float, optional) – The base used to compute angular frequency for +
base (float, optional) – The base used to compute angular frequency for
each dimension in the positional encodings. Exactly one of base and
freqs must be None.
scale (float) – The scale used to scale the positions.
offset (int) – The position offset to start at.
scale (float) – The scale used to scale the positions.
offset (int) – The position offset to start at.
freqs (array, optional) – Optional frequencies to use with RoPE.
If set, the base parameter must be None. Default: None.
-
+
+
@@ -370,6 +370,7 @@
A fast implementation of multi-head attention: O = softmax(Q @ K.T, dim=-1) @ V.
Supports:
v
q (array) – Input query array.
k (array) – Input keys array.
v (array) – Input values array.
scale (float) – Scale for queries (typically 1.0 / sqrt(q.shape(-1))
scale (float) – Scale for queries (typically 1.0 / sqrt(q.shape(-1))
mask (array, optional) – An additive mask to apply to the query-key scores.
-
+
+
@@ -370,6 +370,7 @@
One dimensional discrete Fourier Transform.
a (array) – The input array.
n (int, optional) – Size of the transformed axis. The +
n (int, optional) – Size of the transformed axis. The
corresponding axis in the input is truncated or padded with
zeros to match n. The default value is a.shape[axis].
axis (int, optional) – Axis along which to perform the FFT. The +
axis (int, optional) – Axis along which to perform the FFT. The
default is -1.
-
+
+
@@ -370,6 +370,7 @@
Two dimensional discrete Fourier Transform.
a (array) – The input array.
s (list(int), optional) – Sizes of the transformed axes. The +
s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in s. The default value is the
sizes of a along axes.
axes (list(int), optional) – Axes along which to perform the FFT. +
axes (list(int), optional) – Axes along which to perform the FFT.
The default is [-2, -1].
-
+
+
@@ -370,6 +370,7 @@
n-dimensional discrete Fourier Transform.
a (array) – The input array.
s (list(int), optional) – Sizes of the transformed axes. The +
s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in s. The default value is the
sizes of a along axes.
axes (list(int), optional) – Axes along which to perform the FFT. +
axes (list(int), optional) – Axes along which to perform the FFT.
The default is None in which case the FFT is over the last
len(s) axes are or all axes if s is also None.
-
+
+
@@ -370,6 +370,7 @@
One dimensional inverse discrete Fourier Transform.
a (array) – The input array.
n (int, optional) – Size of the transformed axis. The +
n (int, optional) – Size of the transformed axis. The
corresponding axis in the input is truncated or padded with
zeros to match n. The default value is a.shape[axis].
axis (int, optional) – Axis along which to perform the FFT. The +
axis (int, optional) – Axis along which to perform the FFT. The
default is -1.
-
+
+
@@ -370,6 +370,7 @@
Two dimensional inverse discrete Fourier Transform.
a (array) – The input array.
s (list(int), optional) – Sizes of the transformed axes. The +
s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in s. The default value is the
sizes of a along axes.
axes (list(int), optional) – Axes along which to perform the FFT. +
axes (list(int), optional) – Axes along which to perform the FFT.
The default is [-2, -1].
-
+
+
@@ -370,6 +370,7 @@
n-dimensional inverse discrete Fourier Transform.
a (array) – The input array.
s (list(int), optional) – Sizes of the transformed axes. The +
s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in s. The default value is the
sizes of a along axes.
axes (list(int), optional) – Axes along which to perform the FFT. +
axes (list(int), optional) – Axes along which to perform the FFT.
The default is None in which case the FFT is over the last
len(s) axes or all axes if s is also None.
-
+
+
@@ -370,6 +370,7 @@
The inverse of rfft().
The output has the same shape as the input except along axis in
which case it has size n.
Parameters:
a (array) – The input array.
n (int, optional) – Size of the transformed axis. The
+
n (int, optional) – Size of the transformed axis. The
corresponding axis in the input is truncated or padded with
zeros to match n // 2 + 1. The default value is
a.shape[axis] // 2 + 1.
axis (int, optional) – Axis along which to perform the FFT. The
+
axis (int, optional) – Axis along which to perform the FFT. The
default is -1.
mlx.core.fft.irfft2 — MLX 0.18.0 documentation
+ mlx.core.fft.irfft2 — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
The inverse of rfft2().
Note the input is generally complex. The dimensions of the input
specified in a (array) – The input array. s (list(int), optional) – Sizes of the transformed axes. The
+ s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in axes (list(int), optional) – Axes along which to perform the FFT.
+ axes (list(int), optional) – Axes along which to perform the FFT.
The default is axes are padded or truncated to match the sizes
@@ -881,12 +883,12 @@ and will have size
s except for the last axis
which has size s[-1] // 2 + 1. The default value is the
sizes of a along axes.[-2, -1].
-
+
+
@@ -370,6 +370,7 @@
The inverse of rfftn().
Note the input is generally complex. The dimensions of the input
specified in a (array) – The input array. s (list(int), optional) – Sizes of the transformed axes. The
+ s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in axes (list(int), optional) – Axes along which to perform the FFT.
+ axes (list(int), optional) – Axes along which to perform the FFT.
The default is axes are padded or truncated to match the sizes
@@ -881,11 +883,11 @@ and will have size
diff --git a/docs/build/html/python/_autosummary/mlx.core.fft.rfft.html b/docs/build/html/python/_autosummary/mlx.core.fft.rfft.html
index 733cabe5a..7befc5f70 100644
--- a/docs/build/html/python/_autosummary/mlx.core.fft.rfft.html
+++ b/docs/build/html/python/_autosummary/mlx.core.fft.rfft.html
@@ -8,7 +8,7 @@
- s. The default value is the
sizes of a along axes.None in which case the FFT is over the last
len(s) axes or all axes if s is also None.
-
+
+
@@ -370,6 +370,7 @@
One dimensional discrete Fourier Transform on a real input.
The output has the same shape as the input except along axis in
which case it has size n // 2 + 1.
a (array) – The input array. If the array is complex it will be silently
cast to a real type.
n (int, optional) – Size of the transformed axis. The
+
n (int, optional) – Size of the transformed axis. The
corresponding axis in the input is truncated or padded with
zeros to match n. The default value is a.shape[axis].
axis (int, optional) – Axis along which to perform the FFT. The
+
axis (int, optional) – Axis along which to perform the FFT. The
default is -1.
Two dimensional real discrete Fourier Transform.
The output has the same shape as the input except along the dimensions in
a (array) – The input array. If the array is complex it will be silently
cast to a real type. s (list(int), optional) – Sizes of the transformed axes. The
+ s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in axes (list(int), optional) – Axes along which to perform the FFT.
+ axes (list(int), optional) – Axes along which to perform the FFT.
The default is axes in which case it has sizes from s. The last axis in axes is
@@ -881,11 +883,11 @@ treated as the real axis and will have size
s. The default value is the
sizes of a along axes.[-2, -1].
-
+
+
@@ -370,6 +370,7 @@
n-dimensional real discrete Fourier Transform.
The output has the same shape as the input except along the dimensions in
a (array) – The input array. If the array is complex it will be silently
cast to a real type. s (list(int), optional) – Sizes of the transformed axes. The
+ s (list(int), optional) – Sizes of the transformed axes. The
corresponding axes in the input are truncated or padded with
zeros to match the sizes in axes (list(int), optional) – Axes along which to perform the FFT.
+ axes (list(int), optional) – Axes along which to perform the FFT.
The default is axes in which case it has sizes from s. The last axis in axes is
@@ -881,11 +883,11 @@ treated as the real axis and will have size
diff --git a/docs/build/html/python/_autosummary/mlx.core.flatten.html b/docs/build/html/python/_autosummary/mlx.core.flatten.html
index fc54b7563..566d88890 100644
--- a/docs/build/html/python/_autosummary/mlx.core.flatten.html
+++ b/docs/build/html/python/_autosummary/mlx.core.flatten.html
@@ -8,7 +8,7 @@
- s. The default value is the
sizes of a along axes.None in which case the FFT is over the last
len(s) axes or all axes if s is also None.
-
+
+
@@ -370,6 +370,7 @@
Flatten an array.
The axes flattened will be between start_axis and end_axis,
inclusive. Negative axes are supported. After converting negative axis to
@@ -881,8 +883,8 @@ positive, axes outside the valid range will be clamped to a valid value,
a (array) – Input array.
start_axis (int, optional) – The first dimension to flatten. Defaults to 0.
end_axis (int, optional) – The last dimension to flatten. Defaults to -1.
start_axis (int, optional) – The first dimension to flatten. Defaults to 0.
end_axis (int, optional) – The last dimension to flatten. Defaults to -1.
stream (Stream, optional) – Stream or device. Defaults to None
in which case the default stream of the default device is used.
-
+
+
@@ -370,6 +370,7 @@
Element-wise floor.
-
+
+
@@ -370,6 +370,7 @@
Element-wise integer division.
If either array is a floating point type then it is equivalent to
calling floor() after divide().
-
+
+
@@ -370,6 +370,7 @@
Construct an array with the given value.
Constructs an array of size shape filled with vals. If vals
is an array it must be broadcastable to the given shape.
vals (float or int or array) – Values to fill the array with.
vals (float or int or array) – Values to fill the array with.
dtype (Dtype, optional) – Data type of the output array. If
unspecified the output type is inferred from vals.
-
+
+
@@ -370,6 +370,7 @@
Matrix multiplication with matrix-level gather.
Performs a gather of the operands with the given indices followed by a (possibly batched) matrix multiplication of two arrays. This operation diff --git a/docs/build/html/python/_autosummary/mlx.core.gather_qmm.html b/docs/build/html/python/_autosummary/mlx.core.gather_qmm.html index 871df93a5..39126cb9b 100644 --- a/docs/build/html/python/_autosummary/mlx.core.gather_qmm.html +++ b/docs/build/html/python/_autosummary/mlx.core.gather_qmm.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Perform quantized matrix multiplication with matrix-level gather.
This operation is the quantized equivalent to gather_mm().
Similar to gather_mm(), the indices lhs_indices and
@@ -888,12 +890,12 @@ as w
biases (array) – The biases to use per group_size elements of w
lhs_indices (array, optional) – Integer indices for x. Default: None.
rhs_indices (array, optional) – Integer indices for w. Default: None.
transpose (bool, optional) – Defines whether to multiply with the +
transpose (bool, optional) – Defines whether to multiply with the
transposed w or not, namely whether we are performing
x @ w.T or x @ w. Default: True.
group_size (int, optional) – The size of the group in w that
+
group_size (int, optional) – The size of the group in w that
shares a scale and bias. Default: 64.
bits (int, optional) – The number of bits occupied by each element in +
bits (int, optional) – The number of bits occupied by each element in
w. Default: 4.
-
+
+
@@ -370,6 +370,7 @@
Returns a function which computes the gradient of fun.
fun (Callable) – A function which takes a variable number of
array or trees of array and returns
a scalar output array.
argnums (int or list(int), optional) – Specify the index (or indices) +
argnums (int or list(int), optional) – Specify the index (or indices)
of the positional arguments of fun to compute the gradient
with respect to. If neither argnums nor argnames are
provided argnums defaults to 0 indicating fun’s first
argument.
argnames (str or list(str), optional) – Specify keyword arguments of +
argnames (str or list(str), optional) – Specify keyword arguments of
fun to compute gradients with respect to. It defaults to [] so
no gradients for keyword arguments by default.
-
+
+
@@ -370,6 +370,7 @@
Element-wise greater than.
Strict greater than on two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.greater_equal.html b/docs/build/html/python/_autosummary/mlx.core.greater_equal.html index 79ea135aa..0efba75c3 100644 --- a/docs/build/html/python/_autosummary/mlx.core.greater_equal.html +++ b/docs/build/html/python/_autosummary/mlx.core.greater_equal.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise greater or equal.
Greater than or equal on two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.hadamard_transform.html b/docs/build/html/python/_autosummary/mlx.core.hadamard_transform.html index c0c863949..7f7e096f0 100644 --- a/docs/build/html/python/_autosummary/mlx.core.hadamard_transform.html +++ b/docs/build/html/python/_autosummary/mlx.core.hadamard_transform.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Perform the Walsh-Hadamard transform along the final axis.
Equivalent to:
from scipy.linalg import hadamard
@@ -885,7 +887,7 @@ document.write(`
-
+
+
Create a square identity matrix.
n (int) – The number of rows and columns in the output.
n (int) – The number of rows and columns in the output.
dtype (Dtype, optional) – Data type of the output array. Defaults to float32.
stream (Stream, optional) – Stream or device. Defaults to None.
-
+
+
Ordinary inner product of vectors for 1-D arrays, in higher dimensions a sum product over the last axes.
-
+
+
@@ -370,6 +370,7 @@
Returns a boolean array where two arrays are element-wise equal within a tolerance.
Infinite values are considered equal if they have the same sign, NaN values are
not equal unless equal_nan is True.
a (array) – Input array.
b (array) – Input array.
rtol (float) – Relative tolerance.
atol (float) – Absolute tolerance.
equal_nan (bool) – If True, NaNs are considered equal.
+
rtol (float) – Relative tolerance.
atol (float) – Absolute tolerance.
equal_nan (bool) – If True, NaNs are considered equal.
Defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Return a boolean array indicating which elements are finite.
An element is finite if it is not infinite or NaN.
-
+
+
@@ -370,6 +370,7 @@
Return a boolean array indicating which elements are +/- inifnity.
-
+
+
@@ -370,6 +370,7 @@
Return a boolean array indicating which elements are NaN.
-
+
+
@@ -370,6 +370,7 @@
Return a boolean array indicating which elements are negative infinity.
-
+
+
@@ -370,6 +370,7 @@
Return a boolean array indicating which elements are positive infinity.
-
+
+
@@ -370,6 +370,7 @@
Check if a Dtype or DtypeCategory is a subtype
of another.
Example
diff --git a/docs/build/html/python/_autosummary/mlx.core.jvp.html b/docs/build/html/python/_autosummary/mlx.core.jvp.html index 7a2367e6d..fce1015b0 100644 --- a/docs/build/html/python/_autosummary/mlx.core.jvp.html +++ b/docs/build/html/python/_autosummary/mlx.core.jvp.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Compute the Jacobian-vector product.
This computes the product of the Jacobian of a function fun evaluated
at primals with the tangents.
primals
fun (Callable) – A function which takes a variable number of array
and returns a single array or list of array.
primals (list(array)) – A list of array at which to
evaluate the Jacobian.
tangents (list(array)) – A list of array which are the
“vector” in the Jacobian-vector product. The tangents should be the
same in number, shape, and type as the inputs of fun (i.e. the primals).
@@ -892,7 +894,7 @@ same in number, shape, and type as the inputs of fun.
-
+
+
@@ -370,6 +370,7 @@
Element-wise left shift.
Shift the bits of the first input to the left by the second using numpy-style broadcasting semantics. Either or both input arrays can diff --git a/docs/build/html/python/_autosummary/mlx.core.less.html b/docs/build/html/python/_autosummary/mlx.core.less.html index 79f405ced..4f2ab6ff4 100644 --- a/docs/build/html/python/_autosummary/mlx.core.less.html +++ b/docs/build/html/python/_autosummary/mlx.core.less.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise less than.
Strict less than on two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.less_equal.html b/docs/build/html/python/_autosummary/mlx.core.less_equal.html index f8231dcf7..356973488 100644 --- a/docs/build/html/python/_autosummary/mlx.core.less_equal.html +++ b/docs/build/html/python/_autosummary/mlx.core.less_equal.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise less than or equal.
Less than or equal on two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.linalg.cholesky.html b/docs/build/html/python/_autosummary/mlx.core.linalg.cholesky.html index 168d94844..263fd99d8 100644 --- a/docs/build/html/python/_autosummary/mlx.core.linalg.cholesky.html +++ b/docs/build/html/python/_autosummary/mlx.core.linalg.cholesky.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Compute the Cholesky decomposition of a real symmetric positive semi-definite matrix.
This function supports arrays with at least 2 dimensions. When the input
has more than two dimensions, the Cholesky decomposition is computed for each matrix
@@ -881,7 +883,7 @@ in the last two dimensions of a (array) – Input array. upper (bool, optional) – If upper (bool, optional) – If stream (Stream, optional) – Stream or device. Defaults to Parameters:
True, return the upper triangular Cholesky factor.
+True, return the upper triangular Cholesky factor.
If False, return the lower triangular Cholesky factor. Default: False.None
in which case the default stream of the default device is used.
-
+
+
@@ -371,6 +371,7 @@
Compute the inverse of a real symmetric positive semi-definite matrix using it’s Cholesky decomposition.
Let \(\mathbf{A}\) be a real symmetric positive semi-definite matrix and \(\mathbf{L}\) its Cholesky decomposition such that:
L (array) – Input array.
upper (bool, optional) – If True, return the upper triangular Cholesky factor.
+
upper (bool, optional) – If True, return the upper triangular Cholesky factor.
If False, return the lower triangular Cholesky factor. Default: False.
stream (Stream, optional) – Stream or device. Defaults to None
in which case the default stream of the default device is used.
-
+
+
Compute the cross product of two arrays along a specified axis.
The cross product is defined for arrays with size 2 or 3 in the specified axis. If the size is 2 then the third value is assumed @@ -881,7 +883,7 @@ to be zero.
a (array) – Input array.
b (array) – Input array.
axis (int, optional) – Axis along which to compute the cross +
axis (int, optional) – Axis along which to compute the cross
product. Default: -1.
stream (Stream, optional) – Stream or device. Defaults to None
in which case the default stream of the default device is used.
-
+
+
@@ -370,6 +370,7 @@
Compute the inverse of a square matrix.
This function supports arrays with at least 2 dimensions. When the input has more than two dimensions, the inverse is computed for each matrix diff --git a/docs/build/html/python/_autosummary/mlx.core.linalg.norm.html b/docs/build/html/python/_autosummary/mlx.core.linalg.norm.html index 9024683e5..834e4bb6b 100644 --- a/docs/build/html/python/_autosummary/mlx.core.linalg.norm.html +++ b/docs/build/html/python/_autosummary/mlx.core.linalg.norm.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Matrix or vector norm.
This function computes vector or matrix norms depending on the value of
the ord and axis parameters.
orda (array) – Input array. If axis is None, a must be 1-D or 2-D,
unless ord is None. If both axis and ord are None, the
2-norm of a.flatten will be returned.
ord (int, float or str, optional) – Order of the norm (see table under Notes).
+
ord (int, float or str, optional) – Order of the norm (see table under Notes).
If None, the 2-norm (or Frobenius norm for matrices) will be computed
along the given axis. Default: None.
axis (int or list(int), optional) – If axis is an integer, it specifies the
+
axis (int or list(int), optional) – If axis is an integer, it specifies the
axis of a along which to compute the vector norms. If axis is a
2-tuple, it specifies the axes that hold 2-D matrices, and the matrix
norms of these matrices are computed. If axis is None then
either a vector norm (when a is 1-D) or a matrix norm (when a is
2-D) is returned. Default: None.
keepdims (bool, optional) – If True, the axes which are normed over are
+
keepdims (bool, optional) – If True, the axes which are normed over are
left in the result as dimensions with size one. Default False.
-
+
+
@@ -370,6 +370,7 @@
The QR factorization of the input matrix.
This function supports arrays with at least 2 dimensions. The matrices which are factorized are assumed to be in the last two dimensions of @@ -888,7 +890,7 @@ in which case the default stream of the default device is used.
The Q and R matrices.
Example
diff --git a/docs/build/html/python/_autosummary/mlx.core.linalg.svd.html b/docs/build/html/python/_autosummary/mlx.core.linalg.svd.html index c3f45c964..f7ee98201 100644 --- a/docs/build/html/python/_autosummary/mlx.core.linalg.svd.html +++ b/docs/build/html/python/_autosummary/mlx.core.linalg.svd.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
The Singular Value Decomposition (SVD) of the input matrix.
This function supports arrays with at least 2 dimensions. When the input has more than two dimensions, the function iterates over all indices of the first @@ -889,7 +891,7 @@ in which case the default stream of the default device is used.
A = U @ diag(S) @ Vt
-
+
+
@@ -370,6 +370,7 @@
Compute the inverse of a triangular square matrix.
This function supports arrays with at least 2 dimensions. When the input has more than two dimensions, the inverse is computed for each matrix diff --git a/docs/build/html/python/_autosummary/mlx.core.linspace.html b/docs/build/html/python/_autosummary/mlx.core.linspace.html index 2965bd13d..cba457235 100644 --- a/docs/build/html/python/_autosummary/mlx.core.linspace.html +++ b/docs/build/html/python/_autosummary/mlx.core.linspace.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Generate num evenly spaced numbers over interval [start, stop].
start (scalar) – Starting value.
stop (scalar) – Stopping value.
num (int, optional) – Number of samples, defaults to 50.
num (int, optional) – Number of samples, defaults to 50.
dtype (Dtype, optional) – Specifies the data type of the output,
default to float32.
-
+
+
@@ -370,6 +370,7 @@
Load array(s) from a binary file.
The supported formats are .npy, .npz, .safetensors, and
.gguf.
file (file, str) – File in which the array is saved.
format (str, optional) – Format of the file. If None, the
+
file (file, str) – File in which the array is saved.
format (str, optional) – Format of the file. If None, the
format is inferred from the file extension. Supported formats:
npy, npz, and safetensors. Default: None.
return_metadata (bool, optional) – Load the metadata for formats +
return_metadata (bool, optional) – Load the metadata for formats
which support matadata. The metadata will be returned as an
additional dictionary. Default: False.
Return type:
--
+
-
-
+
+
Element-wise natural logarithm.
-
+
+
@@ -370,6 +370,7 @@
Element-wise base-10 logarithm.
-
+
+
@@ -370,6 +370,7 @@
Element-wise natural log of one plus the array.
-
+
+
@@ -370,6 +370,7 @@
Element-wise base-2 logarithm.
-
+
+
@@ -370,6 +370,7 @@
Element-wise log-add-exp.
This is a numerically stable log-add-exp of two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.logical_and.html b/docs/build/html/python/_autosummary/mlx.core.logical_and.html index c7bc433ec..2557f028b 100644 --- a/docs/build/html/python/_autosummary/mlx.core.logical_and.html +++ b/docs/build/html/python/_autosummary/mlx.core.logical_and.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Element-wise logical and.
-
+
+
@@ -370,6 +370,7 @@
Element-wise logical not.
-
+
+
@@ -370,6 +370,7 @@
Element-wise logical or.
-
+
+
@@ -370,6 +370,7 @@
A log-sum-exp reduction over the given axes.
The log-sum-exp reduction is a numerically stable version of:
log(sum(exp(a), axis))
@@ -881,10 +883,10 @@ document.write(`
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
Matrix multiplication.
Perform the (possibly batched) matrix multiplication of two arrays. This function supports broadcasting for arrays with more than two dimensions.
diff --git a/docs/build/html/python/_autosummary/mlx.core.max.html b/docs/build/html/python/_autosummary/mlx.core.max.html index 54d55038f..23a627f7e 100644 --- a/docs/build/html/python/_autosummary/mlx.core.max.html +++ b/docs/build/html/python/_autosummary/mlx.core.max.html @@ -8,7 +8,7 @@ -
-
+
+
A max reduction over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Element-wise maximum.
Take the element-wise max of two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.mean.html b/docs/build/html/python/_autosummary/mlx.core.mean.html index e37032d5b..8b425a562 100644 --- a/docs/build/html/python/_autosummary/mlx.core.mean.html +++ b/docs/build/html/python/_autosummary/mlx.core.mean.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Compute the mean(s) over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Generate multidimensional coordinate grids from 1-D coordinate arrays
arrays (array) – Input arrays.
sparse (bool, optional) – If True, a sparse grid is returned in which each output
+
sparse (bool, optional) – If True, a sparse grid is returned in which each output
array has a single non-zero element. If False, a dense grid is returned.
Defaults to False.
indexing (str, optional) – Cartesian (‘xy’) or matrix (‘ij’) indexing of the output arrays. +
indexing (str, optional) – Cartesian (‘xy’) or matrix (‘ij’) indexing of the output arrays.
Defaults to 'xy'.
'xy'The output arrays.
-
+
+
@@ -370,6 +370,7 @@
Clear the memory cache.
After calling this, get_cache_memory() should return 0.
-
+
+
@@ -370,6 +370,7 @@
Get information about the GPU device and system settings.
Currently returns:
-
+
+
@@ -370,6 +370,7 @@
Get the actively used memory in bytes.
Note, this will not always match memory use reported by the system because it does not include cached memory buffers.
diff --git a/docs/build/html/python/_autosummary/mlx.core.metal.get_cache_memory.html b/docs/build/html/python/_autosummary/mlx.core.metal.get_cache_memory.html index 26a3944cd..5822bc78f 100644 --- a/docs/build/html/python/_autosummary/mlx.core.metal.get_cache_memory.html +++ b/docs/build/html/python/_autosummary/mlx.core.metal.get_cache_memory.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Get the cache size in bytes.
The cache includes memory not currently used that has not been returned to the system allocator.
diff --git a/docs/build/html/python/_autosummary/mlx.core.metal.get_peak_memory.html b/docs/build/html/python/_autosummary/mlx.core.metal.get_peak_memory.html index 19e1cc52b..4400384db 100644 --- a/docs/build/html/python/_autosummary/mlx.core.metal.get_peak_memory.html +++ b/docs/build/html/python/_autosummary/mlx.core.metal.get_peak_memory.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Get the peak amount of used memory in bytes.
The maximum memory used recorded from the beginning of the program
execution or since the last call to reset_peak_memory().
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Set the free cache limit.
If using more than the given limit, free memory will be reclaimed
from the cache on the next allocation. To disable the cache, set
@@ -880,13 +882,13 @@ the limit to 0set_memory_limit() for more details.
-
+
+
@@ -370,6 +370,7 @@
Set the memory limit.
Memory allocations will wait on scheduled tasks to complete if the limit is exceeded. If there are no more scheduled tasks an error will be raised @@ -882,8 +884,8 @@ size reported by the device.
The previous memory limit in bytes.
Start a Metal capture.
path (str) – The path to save the capture which should have +
path (str) – The path to save the capture which should have
the extension .gputrace.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
A min reduction over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Element-wise minimum.
Take the element-wise min of two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.moveaxis.html b/docs/build/html/python/_autosummary/mlx.core.moveaxis.html index ee07629e7..61c0b753a 100644 --- a/docs/build/html/python/_autosummary/mlx.core.moveaxis.html +++ b/docs/build/html/python/_autosummary/mlx.core.moveaxis.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Move an axis to a new position.
-
+
+
@@ -370,6 +370,7 @@
Element-wise multiplication.
Multiply two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.nan_to_num.html b/docs/build/html/python/_autosummary/mlx.core.nan_to_num.html index b4703f1c6..0a71b29d4 100644 --- a/docs/build/html/python/_autosummary/mlx.core.nan_to_num.html +++ b/docs/build/html/python/_autosummary/mlx.core.nan_to_num.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Replace NaN and Inf values with finite numbers.
a (array) – Input array
nan (float, optional) – Value to replace NaN with. Default: 0.
posinf (float, optional) – Value to replace positive infinities +
nan (float, optional) – Value to replace NaN with. Default: 0.
posinf (float, optional) – Value to replace positive infinities
with. If None, defaults to largest finite value for the
given data type. Default: None.
neginf (float, optional) – Value to replace negative infinities +
neginf (float, optional) – Value to replace negative infinities
with. If None, defaults to the negative of the largest
finite value for the given data type. Default: None.
-
+
+
@@ -370,6 +370,7 @@
Element-wise negation.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Element-wise not equal.
Not equal comparison on two arrays with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.ones.html b/docs/build/html/python/_autosummary/mlx.core.ones.html index ba2a3fc44..6df455f86 100644 --- a/docs/build/html/python/_autosummary/mlx.core.ones.html +++ b/docs/build/html/python/_autosummary/mlx.core.ones.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Construct an array of ones.
dtype (Dtype, optional) – Data type of the output array. If
unspecified the output type defaults to float32.
-
+
+
@@ -370,6 +370,7 @@
An array of ones like the input.
-
+
+
@@ -370,6 +370,7 @@
Compute the outer product of two 1-D arrays, if the array’s passed are not 1-D a flatten op will be run beforehand.
-
+
+
@@ -370,6 +370,7 @@
Pad an array with a constant value
a (array) – Input array.
pad_width (int, tuple(int), tuple(int, int) or list(tuple(int, int))) – Number of padded +
pad_width (int, tuple(int), tuple(int, int) or list(tuple(int, int))) – Number of padded
values to add to the edges of each axis:((before_1, after_1),
(before_2, after_2), ..., (before_N, after_N)). If a single pair
of integers is passed then (before_i, after_i) are all the same.
diff --git a/docs/build/html/python/_autosummary/mlx.core.partition.html b/docs/build/html/python/_autosummary/mlx.core.partition.html
index 2a3cc9676..276abab7c 100644
--- a/docs/build/html/python/_autosummary/mlx.core.partition.html
+++ b/docs/build/html/python/_autosummary/mlx.core.partition.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Returns a partitioned copy of the array such that the smaller kth
elements are first.
The ordering of the elements in partitions is undefined.
@@ -879,11 +881,11 @@ elements are first.a (array) – Input array.
kth (int) – Element at the kth index will be in its sorted
+
kth (int) – Element at the kth index will be in its sorted
position in the output. All elements before the kth index will
be less or equal to the kth element and all elements after
will be greater or equal to the kth element in the output.
axis (int or None, optional) – Optional axis to partition over. +
axis (int or None, optional) – Optional axis to partition over.
If None, this partitions over the flattened array.
If unspecified, it defaults to -1.
-
+
+
@@ -370,6 +370,7 @@
Element-wise power operation.
Raise the elements of a to the powers in elements of b with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.prod.html b/docs/build/html/python/_autosummary/mlx.core.prod.html index 7ddaec972..67611ffb6 100644 --- a/docs/build/html/python/_autosummary/mlx.core.prod.html +++ b/docs/build/html/python/_autosummary/mlx.core.prod.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
An product reduction over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Put values along an axis at the specified indices.
values (array) – Values array. These should be broadcastable with the indices.
axis (int or None) – Axis in the destination to put the values to. If +
axis (int or None) – Axis in the destination to put the values to. If
axis == None the destination is flattened prior to the put
operation.
-
+
+
@@ -371,6 +371,7 @@
Quantize the matrix w using bits bits per element.
Note, every w (array) – Matrix to be quantized group_size (int, optional) – The size of the group in group_size (int, optional) – The size of the group in bits (int, optional) – The number of bits occupied by each element of
+ bits (int, optional) – The number of bits occupied by each element of
group_size elements in a row of w are quantized
together. Hence, number of columns of w should be divisible by
@@ -904,9 +906,9 @@ save \(s\) and Parameters:
w that shares a
+w that shares a
scale and bias. Default: 64.w in the returned quantized matrix. Default: 4.
-
+
+
@@ -370,6 +370,7 @@
Perform the matrix multiplication with the quantized matrix w. The
quantization uses one floating point scale and bias per group_size of
elements. Each element in w takes bits bits and is packed in an
@@ -883,12 +885,12 @@ unsigned 32 bit integer.
w (array) – Quantized matrix packed in unsigned integers
scales (array) – The scales to use per group_size elements of w
biases (array) – The biases to use per group_size elements of w
transpose (bool, optional) – Defines whether to multiply with the +
transpose (bool, optional) – Defines whether to multiply with the
transposed w or not, namely whether we are performing
x @ w.T or x @ w. Default: True.
group_size (int, optional) – The size of the group in w that
+
group_size (int, optional) – The size of the group in w that
shares a scale and bias. Default: 64.
bits (int, optional) – The number of bits occupied by each element in +
bits (int, optional) – The number of bits occupied by each element in
w. Default: 4.
-
+
+
@@ -370,6 +370,7 @@
Convert angles from degrees to radians.
-
+
+
@@ -370,6 +370,7 @@
Generate Bernoulli random values.
The values are sampled from the bernoulli distribution with parameter
-p. The parameter p can be a float or array and
+p. The parameter p can be a float or array and
must be broadcastable to shape.
p (float or array, optional) – Parameter of the Bernoulli
distribution. Default: 0.5.
shape (list(int), optional) – Shape of the output.
Default: p.shape.
key (array, optional) – A PRNG key. Default: None.
-
+
+
@@ -370,6 +370,7 @@
Sample from a categorical distribution.
The values are sampled from the categorical distribution specified by
the unnormalized values in logits (array) – The unnormalized categorical distribution(s). axis (int, optional) – The axis which specifies the distribution.
+ axis (int, optional) – The axis which specifies the distribution.
Default: shape (list(int), optional) – The shape of the output. This must
+ shape (list(int), optional) – The shape of the output. This must
be broadcast compatable with num_samples (int, optional) – The number of samples to draw from each
+ num_samples (int, optional) – The number of samples to draw from each
of the categorical distributions in key (array, optional) – A PRNG key. Default: logits. Note, at most one of shape
@@ -881,12 +883,12 @@ has the same shape as Parameters:
-1.logits.shape with the axis
dimension removed. Default: Nonelogits. The output will have
num_samples in the last dimension. Default: None.None.
-
+
+
@@ -370,6 +370,7 @@
Sample from the standard Gumbel distribution.
The values are sampled from a standard Gumbel distribution
which CDF exp(-exp(-x)).
-
+
+
@@ -370,6 +370,7 @@
Get a PRNG key from a seed.
seed (int) – Seed for the PRNG.
+seed (int) – Seed for the PRNG.
The PRNG key array.
diff --git a/docs/build/html/python/_autosummary/mlx.core.random.laplace.html b/docs/build/html/python/_autosummary/mlx.core.random.laplace.html index fc5d00cde..179bfe062 100644 --- a/docs/build/html/python/_autosummary/mlx.core.random.laplace.html +++ b/docs/build/html/python/_autosummary/mlx.core.random.laplace.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Sample numbers from a Laplace distribution.
shape (list(int), optional) – Shape of the output. Default: ().
shape (list(int), optional) – Shape of the output. Default: ().
dtype (Dtype, optional) – Type of the output. Default: float32.
loc (float, optional) – Mean of the distribution. Default: 0.0.
scale (float, optional) – The scale “b” of the Laplace distribution. +
loc (float, optional) – Mean of the distribution. Default: 0.0.
scale (float, optional) – The scale “b” of the Laplace distribution.
Default:1.0.
key (array, optional) – A PRNG key. Default: None.
1.0<
diff --git a/docs/build/html/python/_autosummary/mlx.core.random.multivariate_normal.html b/docs/build/html/python/_autosummary/mlx.core.random.multivariate_normal.html
index 042f9cbad..c1026cb3c 100644
--- a/docs/build/html/python/_autosummary/mlx.core.random.multivariate_normal.html
+++ b/docs/build/html/python/_autosummary/mlx.core.random.multivariate_normal.html
@@ -8,7 +8,7 @@
- mlx.core.random.multivariate_normal — MLX 0.18.0 documentation
+ mlx.core.random.multivariate_normal — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
Generate jointly-normal random samples given a mean and covariance.
The matrix cov must be positive semi-definite. The behavior is
undefined if it is not. The only supported dtype is float32.
cov (array) – array of shape (..., n, n), the covariance
matrix of the distribution. The batch shape ... must be
broadcast-compatible with that of mean.
shape (list(int), optional) – The output shape must be
broadcast-compatible with mean.shape[:-1] and cov.shape[:-2].
If empty, the result shape is determined by broadcasting the batch
shapes of mean and cov. Default: [].
-
+
+
@@ -370,6 +370,7 @@
Generate normally distributed random numbers.
shape (list(int), optional) – Shape of the output. Default is ().
shape (list(int), optional) – Shape of the output. Default is ().
dtype (Dtype, optional) – Type of the output. Default is float32.
loc (float, optional) – Mean of the distribution. Default is 0.0.
scale (float, optional) – Standard deviation of the distribution. Default is 1.0.
loc (float, optional) – Mean of the distribution. Default is 0.0.
scale (float, optional) – Standard deviation of the distribution. Default is 1.0.
key (array, optional) – A PRNG key. Default: None.
+
+
+
+Generate a random permutation or permute the entries of an array.
+The generated random permutation or randomly permuted input array.
+
-
+
+
@@ -370,6 +370,7 @@
Generate random integers from the given interval.
The values are sampled with equal probability from the integers in
half-open interval low (scalar or array) – Lower bound of the interval. high (scalar or array) – Upper bound of the interval. shape (list(int), optional) – Shape of the output. Default: shape (list(int), optional) – Shape of the output. Default: dtype (Dtype, optional) – Type of the output. Default: key (array, optional) – A PRNG key. Default: [low, high). The lower and upper bound can be
@@ -881,7 +883,7 @@ scalars or arrays and must be roadcastable to
diff --git a/docs/build/html/python/_autosummary/mlx.core.random.seed.html b/docs/build/html/python/_autosummary/mlx.core.random.seed.html
index 0dd92d2f8..e0d394381 100644
--- a/docs/build/html/python/_autosummary/mlx.core.random.seed.html
+++ b/docs/build/html/python/_autosummary/mlx.core.random.seed.html
@@ -8,7 +8,7 @@
- ().().int32.None.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Split a PRNG key into sub keys.
-
+
+
@@ -370,6 +370,7 @@
Generate values from a truncated normal distribution.
The values are sampled from the truncated normal distribution
on the domain lower (scalar or array) – Lower bound of the domain. upper (scalar or array) – Upper bound of the domain. shape (list(int), optional) – The shape of the output.
Default: dtype (Dtype, optional) – The data type of the output.
Default: (lower, upper). The bounds lower and upper
@@ -881,7 +883,7 @@ can be scalars or arrays and must be broadcastable to
().float32.
-
+
+
@@ -370,6 +370,7 @@
Generate uniformly distributed random numbers.
The values are sampled uniformly in the half-open interval [low, high).
The lower and upper bound can be scalars or arrays and must be
@@ -883,7 +885,7 @@ broadcastable to sh
Default: 0.
high (scalar or array, optional) – Upper bound of the distribution.
Default: 1.
shape (list(int), optional) – Shape of the output. Default:().
shape (list(int), optional) – Shape of the output. Default:().
dtype (Dtype, optional) – Type of the output. Default: float32.
key (array, optional) – A PRNG key. Default: None.
-
+
+
@@ -370,6 +370,7 @@
Element-wise reciprocal.
-
+
+
@@ -370,6 +370,7 @@
Element-wise remainder of division.
Computes the remainder of dividing a with b with numpy-style broadcasting semantics. Either or both input arrays can also be diff --git a/docs/build/html/python/_autosummary/mlx.core.repeat.html b/docs/build/html/python/_autosummary/mlx.core.repeat.html index b5cad8fc7..02523905a 100644 --- a/docs/build/html/python/_autosummary/mlx.core.repeat.html +++ b/docs/build/html/python/_autosummary/mlx.core.repeat.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Repeat an array along a specified axis.
array (array) – Input array.
repeats (int) – The number of repetitions for each element.
axis (int, optional) – The axis in which to repeat the array along. If +
repeats (int) – The number of repetitions for each element.
axis (int, optional) – The axis in which to repeat the array along. If unspecified it uses the flattened array of the input and repeats along axis 0.
stream (Stream, optional) – Stream or device. Defaults to None.
-
+
+
@@ -370,6 +370,7 @@
Reshape an array while preserving the size.
a (array) – Input array.
stream (Stream, optional) – Stream or device. Defaults to None
in which case the default stream of the default device is used.
-
+
+
@@ -370,6 +370,7 @@
Element-wise right shift.
Shift the bits of the first input to the right by the second using numpy-style broadcasting semantics. Either or both input arrays can @@ -915,11 +917,11 @@ also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.roll.html b/docs/build/html/python/_autosummary/mlx.core.roll.html new file mode 100644 index 000000000..95b93796f --- /dev/null +++ b/docs/build/html/python/_autosummary/mlx.core.roll.html @@ -0,0 +1,998 @@ + + + + + + + + + + +
+
+
+
+Roll array elements along a given axis.
+Elements that are rolled beyond the end of the array are introduced at +the beggining and vice-versa.
+If the axis is not provided the array is flattened, rolled and then the +shape is restored.
+a (array) – Input array
shift (int or tuple(int)) – The number of places by which elements +are shifted. If positive the array is rolled to the right, if +negative it is rolled to the left. If an int is provided but the +axis is a tuple then the same value is used for all axes.
axis (int or tuple(int), optional) – The axis or axes along which to +roll the elements.
-
+
+
@@ -370,6 +370,7 @@
Round to the given number of decimals.
Basically performs:
s = 10**decimals
@@ -882,7 +884,7 @@ document.write(`
-
+
+
Element-wise reciprocal and square root.
-
+
+
Save the array to a binary file in .npy format.
-
+
+
Save array(s) to a binary file in .gguf format.
See the GGUF documentation for more information on the format.
file (file, str) – File in which the array is saved.
arrays (dict(str, array)) – The dictionary of names to arrays to +
file (file, str) – File in which the array is saved.
arrays (dict(str, array)) – The dictionary of names to arrays to be saved.
metadata (dict(str, Union[array, str, list(str)])) – The dictionary +
metadata (dict(str, Union[array, str, list(str)])) – The dictionary
of metadata to be saved. The values can be a scalar or 1D
-obj:array, a str, or a list of str.
str, or a list of str.
-
+
+
@@ -370,6 +370,7 @@
Save array(s) to a binary file in .safetensors format.
See the Safetensors documentation for more information on the format.
file (file, str) – File in which the array is saved.
arrays (dict(str, array)) – The dictionary of names to arrays to
file (file, str) – File in which the array is saved.
arrays (dict(str, array)) – The dictionary of names to arrays to
metadata (be saved.) – The dictionary of
saved. (metadata to be) –
-
+
+
@@ -370,6 +370,7 @@
Save several arrays to a binary file in uncompressed .npz
format.
import mlx.core as mx
@@ -890,7 +892,7 @@ format.
file (file, str) – Path to file to which the arrays are saved.
file (file, str) – Path to file to which the arrays are saved.
args (arrays) – Arrays to be saved.
kwargs (arrays) – Arrays to be saved. Each array will be saved with the associated keyword as the output file name.
-
+
+
Save several arrays to a binary file in compressed .npz format.
file (file, str) – Path to file to which the arrays are saved.
file (file, str) – Path to file to which the arrays are saved.
args (arrays) – Arrays to be saved.
kwargs (arrays) – Arrays to be saved. Each array will be saved with the associated keyword as the output file name.
-
+
+
Set the default device.
-
+
+
@@ -370,6 +370,7 @@
Set the default stream.
This will make the given stream the default for the streams device. It will not change the default device.
diff --git a/docs/build/html/python/_autosummary/mlx.core.sigmoid.html b/docs/build/html/python/_autosummary/mlx.core.sigmoid.html index be55a1b46..8fcf3d580 100644 --- a/docs/build/html/python/_autosummary/mlx.core.sigmoid.html +++ b/docs/build/html/python/_autosummary/mlx.core.sigmoid.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Element-wise logistic sigmoid.
The logistic sigmoid function is:
-
+
+
Element-wise sign.
-
+
+
@@ -370,6 +370,7 @@
Element-wise sine.
-
+
+
@@ -370,6 +370,7 @@
Element-wise hyperbolic sine.
-
+
+
@@ -370,6 +370,7 @@
Perform the softmax along the given axis.
This operation is a numerically stable version of:
exp(a) / sum(exp(a), axis, keepdims=True)
@@ -881,7 +883,7 @@ document.write(`
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to compute +
axis (int or list(int), optional) – Optional axis or axes to compute the softmax over. If unspecified this performs the softmax over the full array.
-
+
+
Returns a sorted copy of the array.
a (array) – Input array.
axis (int or None, optional) – Optional axis to sort over. +
axis (int or None, optional) – Optional axis to sort over.
If None, this sorts over the flattened array.
If unspecified, it defaults to -1 (sorting over the last axis).
-
+
+
Split an array along a given axis.
a (array) – Input array.
indices_or_sections (int or list(int)) – If indices_or_sections
+
indices_or_sections (int or list(int)) – If indices_or_sections
is an integer the array is split into that many sections of equal
size. An error is raised if this is not possible. If indices_or_sections
is a list, the list contains the indices of the start of each subarray
along the given axis.
axis (int, optional) – Axis to split along, defaults to 0.
axis (int, optional) – Axis to split along, defaults to 0.
A list of split arrays.
-
+
+
@@ -370,6 +370,7 @@
Element-wise square root.
-
+
+
@@ -370,6 +370,7 @@
Element-wise square.
-
+
+
@@ -370,6 +370,7 @@
Remove length one axes from an array.
-
+
+
@@ -370,6 +370,7 @@
Stacks the arrays along a new axis.
axis (int, optional) – The axis in the result array along which the +
axis (int, optional) – The axis in the result array along which the
input arrays are stacked. Defaults to 0.
stream (Stream, optional) – Stream or device. Defaults to None.
-
+
+
@@ -370,6 +370,7 @@
Compute the standard deviation(s) over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
ddof (int, optional) – The divisor to compute the variance +
ddof (int, optional) – The divisor to compute the variance
is N - ddof, defaults to 0.
-
+
+
@@ -370,6 +370,7 @@
Stop gradients from being computed.
The operation is the identity but it prevents gradients from flowing through the array.
diff --git a/docs/build/html/python/_autosummary/mlx.core.stream.html b/docs/build/html/python/_autosummary/mlx.core.stream.html index 06d14fe75..6e72c0efa 100644 --- a/docs/build/html/python/_autosummary/mlx.core.stream.html +++ b/docs/build/html/python/_autosummary/mlx.core.stream.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Create a context manager to set the default device and stream.
-
+
+
@@ -370,6 +370,7 @@
Element-wise subtraction.
Subtract one array from another with numpy-style broadcasting semantics. Either or both input arrays can also be scalars.
diff --git a/docs/build/html/python/_autosummary/mlx.core.sum.html b/docs/build/html/python/_autosummary/mlx.core.sum.html index 962370ac1..cfc43d60e 100644 --- a/docs/build/html/python/_autosummary/mlx.core.sum.html +++ b/docs/build/html/python/_autosummary/mlx.core.sum.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Sum reduce the array over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
-
+
+
@@ -370,6 +370,7 @@
Swap two axes of an array.
-
+
+
@@ -370,6 +370,7 @@
Synchronize with the given stream.
-
+
+
@@ -370,6 +370,7 @@
Take elements along an axis.
The elements are taken from indices along the specified axis.
If the axis is not specified the array is treated as a flattened
@@ -881,8 +883,8 @@ If the axis is not specified the array is treated as a flattened
a (array) – Input array.
indices (int or array) – Integer index or input array with integral type.
axis (int, optional) – Axis along which to perform the take. If unspecified +
indices (int or array) – Integer index or input array with integral type.
axis (int, optional) – Axis along which to perform the take. If unspecified the array is treated as a flattened 1-D vector.
-
+
+
@@ -370,6 +370,7 @@
Take values along an axis at the specified indices.
a (array) – Input array.
indices (array) – Indices array. These should be broadcastable with the input array excluding the axis dimension.
axis (int or None) – Axis in the input to take the values from. If +
axis (int or None) – Axis in the input to take the values from. If
axis == None the array is flattened to 1D prior to the indexing
operation.
-
+
+
@@ -370,6 +370,7 @@
Element-wise tangent.
-
+
+
@@ -370,6 +370,7 @@
Element-wise hyperbolic tangent.
-
+
+
@@ -370,6 +370,7 @@
Compute the tensor dot product along the specified axes.
a (array) – Input array
b (array) – Input array
axes (int or list(list(int)), optional) – The number of dimensions to +
axes (int or list(list(int)), optional) – The number of dimensions to
sum over. If an integer is provided, then sum over the last
axes dimensions of a and the first axes dimensions of
b. If a list of lists is provided, then sum over the
diff --git a/docs/build/html/python/_autosummary/mlx.core.tile.html b/docs/build/html/python/_autosummary/mlx.core.tile.html
index b3c5300f7..f6c2f24ce 100644
--- a/docs/build/html/python/_autosummary/mlx.core.tile.html
+++ b/docs/build/html/python/_autosummary/mlx.core.tile.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Construct an array by repeating a the number of times given by reps.
-
+
+
@@ -370,6 +370,7 @@
Returns the k largest elements from the input along a given axis.
The elements will not necessarily be in sorted order.
a (array) – Input array.
k (int) – k top elements to be returned
axis (int or None, optional) – Optional axis to select over. +
k (int) – k top elements to be returned
axis (int or None, optional) – Optional axis to select over.
If None, this selects the top k elements over the
flattened array. If unspecified, it defaults to -1.
-
+
+
@@ -370,6 +370,7 @@
Return the sum along a specified diagonal in the given array.
a (array) – Input array
offset (int, optional) – Offset of the diagonal from the main diagonal. +
offset (int, optional) – Offset of the diagonal from the main diagonal.
Can be positive or negative. Default: 0.
axis1 (int, optional) – The first axis of the 2-D sub-arrays from which +
axis1 (int, optional) – The first axis of the 2-D sub-arrays from which
the diagonals should be taken. Default: 0.
axis2 (int, optional) – The second axis of the 2-D sub-arrays from which +
axis2 (int, optional) – The second axis of the 2-D sub-arrays from which
the diagonals should be taken. Default: 1.
dtype (Dtype, optional) – Data type of the output array. If unspecified the output type is inferred from the input array.
-
+
+
@@ -370,6 +370,7 @@
Transpose the dimensions of the array.
-
+
+
@@ -370,6 +370,7 @@
An array with ones at and below the given diagonal and zeros elsewhere.
n (int) – The number of rows in the output.
m (int, optional) – The number of cols in the output. Defaults to None.
k (int, optional) – The diagonal of the 2-D array. Defaults to 0.
n (int) – The number of rows in the output.
m (int, optional) – The number of cols in the output. Defaults to None.
k (int, optional) – The diagonal of the 2-D array. Defaults to 0.
dtype (Dtype, optional) – Data type of the output array. Defaults to float32.
stream (Stream, optional) – Stream or device. Defaults to None.
-
+
+
@@ -370,6 +370,7 @@
Zeros the array above the given diagonal.
-
+
+
@@ -370,6 +370,7 @@
Zeros the array below the given diagonal.
-
+
+
@@ -370,6 +370,7 @@
Returns a function which computes the value and gradient of fun.
The function passed to value_and_grad() should return either
a scalar loss or a tuple in which the first element is a scalar
@@ -905,12 +907,12 @@ loss and the remaining elements can be anything.
array or trees of array and returns
a scalar output array or a tuple the first element
of which should be a scalar array.argnums (int or list(int), optional) – Specify the index (or indices) +
argnums (int or list(int), optional) – Specify the index (or indices)
of the positional arguments of fun to compute the gradient
with respect to. If neither argnums nor argnames are
provided argnums defaults to 0 indicating fun’s first
argument.
argnames (str or list(str), optional) – Specify keyword arguments of +
argnames (str or list(str), optional) – Specify keyword arguments of
fun to compute gradients with respect to. It defaults to [] so
no gradients for keyword arguments by default.
-
+
+
@@ -370,6 +370,7 @@
Compute the variance(s) over the given axes.
a (array) – Input array.
axis (int or list(int), optional) – Optional axis or axes to reduce over. If unspecified this defaults to reducing over the entire array.
keepdims (bool, optional) – Keep reduced axes as +
keepdims (bool, optional) – Keep reduced axes as singleton dimensions, defaults to False.
ddof (int, optional) – The divisor to compute the variance +
ddof (int, optional) – The divisor to compute the variance
is N - ddof, defaults to 0.
-
+
+
@@ -370,6 +370,7 @@
View the array as a different type.
The output shape changes along the last axis if the input array’s
type and the input dtype do not have the same size.
-
+
+
@@ -370,6 +370,7 @@
Compute the vector-Jacobian product.
Computes the product of the cotangents with the Jacobian of a
function fun evaluated at primals.
fun
fun (Callable) – A function which takes a variable number of array
and returns a single array or list of array.
primals (list(array)) – A list of array at which to
evaluate the Jacobian.
cotangents (list(array)) – A list of array which are the
“vector” in the vector-Jacobian product. The cotangents should be the
same in number, shape, and type as the outputs of fun.
@@ -892,7 +894,7 @@ same in number, shape, and type as the outputs of fun.
- Return type:
--
+
-
-
+
+
@@ -370,6 +370,7 @@
Returns a vectorized version of fun.
fun (Callable) – A function which takes a variable number of
array or a tree of array and returns
a variable number of array or a tree of array.
in_axes (int, optional) – An integer or a valid prefix tree of the +
in_axes (int, optional) – An integer or a valid prefix tree of the
inputs to fun where each node specifies the vmapped axis. If
the value is None then the corresponding input(s) are not vmapped.
Defaults to 0.
out_axes (int, optional) – An integer or a valid prefix tree of the +
out_axes (int, optional) – An integer or a valid prefix tree of the
outputs of fun where each node specifies the vmapped axis. If
the value is None then the corresponding outputs(s) are not vmapped.
Defaults to 0.
-
+
+
@@ -370,6 +370,7 @@
Select from x or y according to condition.
The condition and input arrays must be the same shape or broadcastable with each another.
diff --git a/docs/build/html/python/_autosummary/mlx.core.zeros.html b/docs/build/html/python/_autosummary/mlx.core.zeros.html index 1e2a7180d..b738e57a5 100644 --- a/docs/build/html/python/_autosummary/mlx.core.zeros.html +++ b/docs/build/html/python/_autosummary/mlx.core.zeros.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Construct an array of zeros.
dtype (Dtype, optional) – Data type of the output array. If
unspecified the output type defaults to float32.
-
+
+
@@ -370,6 +370,7 @@
An array of zeros like the input.
-
+
+
@@ -370,6 +370,7 @@
Quantize the sub-modules of a module according to a predicate.
By default all layers that define a to_quantized(group_size, bits)
method will be quantized. Both Linear and Embedding layers
@@ -880,9 +882,9 @@ will be quantized. Note also, the module is updated in-place.
model (Module) – The model whose leaf modules may be quantized.
group_size (int) – The quantization group size (see +
group_size (int) – The quantization group size (see
mlx.core.quantize()). Default: 64.
bits (int) – The number of bits per parameter (see +
bits (int) – The number of bits per parameter (see
mlx.core.quantize()). Default: 4.
class_predicate (Optional[Callable]) – A callable which receives the
Module path and Module itself and returns True if
diff --git a/docs/build/html/python/_autosummary/mlx.nn.value_and_grad.html b/docs/build/html/python/_autosummary/mlx.nn.value_and_grad.html
index 059bae60e..b24b1df49 100644
--- a/docs/build/html/python/_autosummary/mlx.nn.value_and_grad.html
+++ b/docs/build/html/python/_autosummary/mlx.nn.value_and_grad.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Transform the passed function fn to a function that computes the
gradients of fn wrt the model’s trainable parameters and also its
value.
-
+
+
@@ -370,6 +370,7 @@
max_no
- Parameters:
-
- Returns:
@@ -895,7 +897,7 @@ greater than max_no
gradient norm.
- Return type:
--
+
-
diff --git a/docs/build/html/python/_autosummary/mlx.utils.tree_flatten.html b/docs/build/html/python/_autosummary/mlx.utils.tree_flatten.html
index 1f43e7af2..6496f16bd 100644
--- a/docs/build/html/python/_autosummary/mlx.utils.tree_flatten.html
+++ b/docs/build/html/python/_autosummary/mlx.utils.tree_flatten.html
@@ -8,7 +8,7 @@
- mlx.utils.tree_flatten — MLX 0.18.0 documentation
+ mlx.utils.tree_flatten — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
Flattens a Python tree to a list of key, value tuples.
The keys are using the dot notation to define trees of arbitrary depth and complexity.
@@ -892,7 +894,7 @@ complexity.tree (Any) – The Python tree to be flattened.
prefix (str) – A prefix to use for the keys. The first character is +
prefix (str) – A prefix to use for the keys. The first character is always discarded.
is_leaf (callable) – An optional callable that returns True if the passed object is considered a leaf or False otherwise.
The flat representation of the Python tree.
List[Tuple[str, Any]]
+List[Tuple[str, Any]]
-
+
+
@@ -370,6 +370,7 @@
Applies fn to the leaves of the Python tree tree and
returns a new collection with the results.
If rest is provided, every item is assumed to be a superset of tree
and the corresponding leaves are provided as extra positional arguments to
-fn. In that respect, tree_map() is closer to itertools.starmap()
-than to map().
fn. In that respect, tree_map() is closer to itertools.starmap()
+than to map().
The keyword argument is_leaf decides what constitutes a leaf from
tree similar to tree_flatten().
import mlx.nn as nn
@@ -896,7 +898,7 @@ than to
fn (callable) – The function that processes the leaves of the tree.
tree (Any) – The main Python tree that will be iterated upon.
rest (tuple[Any]) – Extra trees to be iterated together with tree.
rest (tuple[Any]) – Extra trees to be iterated together with tree.
is_leaf (callable, optional) – An optional callable that returns True
if the passed object is considered a leaf or False otherwise.
diff --git a/docs/build/html/python/_autosummary/mlx.utils.tree_map_with_path.html b/docs/build/html/python/_autosummary/mlx.utils.tree_map_with_path.html
index e43503de8..4dd8c5e21 100644
--- a/docs/build/html/python/_autosummary/mlx.utils.tree_map_with_path.html
+++ b/docs/build/html/python/_autosummary/mlx.utils.tree_map_with_path.html
@@ -8,7 +8,7 @@
- mlx.utils.tree_map_with_path — MLX 0.18.0 documentation
+ mlx.utils.tree_map_with_path — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
Applies fn to the path and leaves of the Python tree tree and
returns a new collection with the results.
This function is the same tree_map() but the fn takes the path as
@@ -881,7 +883,7 @@ the first argument followed by the remaining tree nodes.
fn (callable) – The function that processes the leaves of the tree.
tree (Any) – The main Python tree that will be iterated upon.
rest (tuple[Any]) – Extra trees to be iterated together with tree.
rest (tuple[Any]) – Extra trees to be iterated together with tree.
is_leaf (callable, optional) – An optional callable that returns True
if the passed object is considered a leaf or False otherwise.
-
+
+
-
+
+
@@ -370,6 +370,7 @@
Recreate a Python tree from its flat representation.
from mlx.utils import tree_unflatten
@@ -882,7 +884,7 @@ document.write(`
tree (list[tuple[str, Any]]) – The flat representation of a Python tree. +
tree (list[tuple[str, Any]]) – The flat representation of a Python tree.
For instance as returned by tree_flatten().
-
+
+
A stream for running operations on a given device.
Methods
diff --git a/docs/build/html/python/array.html b/docs/build/html/python/array.html index aa281dd68..1d9d8f96b 100644 --- a/docs/build/html/python/array.html +++ b/docs/build/html/python/array.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
Access the value of a scalar array.
array.tolist(self)
Convert the array to a Python list.
Convert the array to a Python list.
The array's Dtype.
See reciprocal().
array.reshape(self, *shape[, stream])
Equivalent to reshape() but the shape can be passed either as a tuple or as separate arguments.
Equivalent to reshape() but the shape can be passed either as a tuple or as separate arguments.
array.round(self[, decimals, stream])
See round().
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies 1-dimensional average pooling.
Assuming an input of shape \((N, L, C)\) and kernel_size is
\(k\), the output is a tensor of shape \((N, L_{out}, C)\), given
@@ -887,10 +889,10 @@ by:
kernel_size (int or tuple(int)) – The size of the pooling window kernel.
stride (int or tuple(int), optional) – The stride of the pooling window. +
kernel_size (int or tuple(int)) – The size of the pooling window kernel.
stride (int or tuple(int), optional) – The stride of the pooling window.
Default: kernel_size.
padding (int or tuple(int), optional) – How much zero padding to apply to +
padding (int or tuple(int), optional) – How much zero padding to apply to
the input. The padding amount is applied to both sides of the spatial
axis. Default: 0.
-
+
+
@@ -371,6 +371,7 @@
Applies 2-dimensional average pooling.
Assuming an input of shape \((N, H, W, C)\) and kernel_size (int or tuple(int, int)) – The size of the pooling window. stride (int or tuple(int, int), optional) – The stride of the pooling
+ kernel_size (int or tuple(int, int)) – The size of the pooling window. stride (int or tuple(int, int), optional) – The stride of the pooling
window. Default: padding (int or tuple(int, int), optional) – How much zero
+ padding (int or tuple(int, int), optional) – How much zero
padding to apply to the input. The padding is applied on both sides
of the height and width axis. Default: kernel_size is
\((k_H, k_W)\), the output is a tensor of shape \((N, H_{out},
@@ -897,10 +899,10 @@ used for the height axis, the second
-
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.BatchNorm.html b/docs/build/html/python/nn/_autosummary/mlx.nn.BatchNorm.html
index b5a7c1b82..69e322de8 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.BatchNorm.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.BatchNorm.html
@@ -8,7 +8,7 @@
- kernel_size.0.
-
+
+
@@ -371,6 +371,7 @@
Applies Batch Normalization over a 2D or 3D input.
Computes
num_features (int) – The feature dimension to normalize over.
eps (float, optional) – A small additive constant for numerical +
num_features (int) – The feature dimension to normalize over.
eps (float, optional) – A small additive constant for numerical
stability. Default: 1e-5.
momentum (float, optional) – The momentum for updating the running +
momentum (float, optional) – The momentum for updating the running
mean and variance. Default: 0.1.
affine (bool, optional) – If True, apply a learned affine
+
affine (bool, optional) – If True, apply a learned affine
transformation after the normalization. Default: True.
track_running_stats (bool, optional) – If True, track the
+
track_running_stats (bool, optional) – If True, track the
running mean and variance. Default: True.
-
+
+
-
+
+
@@ -370,6 +370,7 @@
Applies a 1-dimensional convolution over the multi-channel input sequence.
The channels are expected to be last i.e. the input shape should be NLC where:
in_channels (int) – The number of input channels
out_channels (int) – The number of output channels
kernel_size (int) – The size of the convolution filters
stride (int, optional) – The stride when applying the filter. +
in_channels (int) – The number of input channels
out_channels (int) – The number of output channels
kernel_size (int) – The size of the convolution filters
stride (int, optional) – The stride when applying the filter.
Default: 1.
padding (int, optional) – How many positions to 0-pad the input with. +
padding (int, optional) – How many positions to 0-pad the input with.
Default: 0.
dilation (int, optional) – The dilation of the convolution.
groups (int, optional) – The number of groups for the convolution. +
dilation (int, optional) – The dilation of the convolution.
groups (int, optional) – The number of groups for the convolution.
Default: 1.
bias (bool, optional) – If True add a learnable bias to the output.
+
bias (bool, optional) – If True add a learnable bias to the output.
Default: True
-
+
+
@@ -370,6 +370,7 @@
Applies a 2-dimensional convolution over the multi-channel input image.
The channels are expected to be last i.e. the input shape should be NHWC where:
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when +
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when
applying the filter. Default: 1.
padding (int or tuple, optional) – How many positions to 0-pad +
padding (int or tuple, optional) – How many positions to 0-pad
the input with. Default: 0.
dilation (int or tuple, optional) – The dilation of the convolution.
bias (bool, optional) – If True add a learnable bias to the
+
dilation (int or tuple, optional) – The dilation of the convolution.
bias (bool, optional) – If True add a learnable bias to the
output. Default: True
-
+
+
@@ -370,6 +370,7 @@
Applies a 3-dimensional convolution over the multi-channel input image.
The channels are expected to be last i.e. the input shape should be NDHWC where:
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when +
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when
applying the filter. Default: 1.
padding (int or tuple, optional) – How many positions to 0-pad +
padding (int or tuple, optional) – How many positions to 0-pad
the input with. Default: 0.
bias (bool, optional) – If True add a learnable bias to the
+
bias (bool, optional) – If True add a learnable bias to the
output. Default: True
-
+
+
@@ -370,6 +370,7 @@
Applies a 1-dimensional transposed convolution over the multi-channel input sequence.
The channels are expected to be last i.e. the input shape should be NLC where:
in_channels (int) – The number of input channels
out_channels (int) – The number of output channels
kernel_size (int) – The size of the convolution filters
stride (int, optional) – The stride when applying the filter. +
in_channels (int) – The number of input channels
out_channels (int) – The number of output channels
kernel_size (int) – The size of the convolution filters
stride (int, optional) – The stride when applying the filter.
Default: 1.
padding (int, optional) – How many positions to 0-pad the input with. +
padding (int, optional) – How many positions to 0-pad the input with.
Default: 0.
dilation (int, optional) – The dilation of the convolution.
bias (bool, optional) – If True add a learnable bias to the output.
+
dilation (int, optional) – The dilation of the convolution.
bias (bool, optional) – If True add a learnable bias to the output.
Default: True
-
+
+
@@ -370,6 +370,7 @@
Applies a 2-dimensional transposed convolution over the multi-channel input image.
The channels are expected to be last i.e. the input shape should be NHWC where:
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when +
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when
applying the filter. Default: 1.
padding (int or tuple, optional) – How many positions to 0-pad +
padding (int or tuple, optional) – How many positions to 0-pad
the input with. Default: 0.
dilation (int or tuple, optional) – The dilation of the convolution.
bias (bool, optional) – If True add a learnable bias to the
+
dilation (int or tuple, optional) – The dilation of the convolution.
bias (bool, optional) – If True add a learnable bias to the
output. Default: True
-
+
+
@@ -370,6 +370,7 @@
Applies a 3-dimensional transposed convolution over the multi-channel input image.
The channels are expected to be last i.e. the input shape should be NDHWC where:
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when +
in_channels (int) – The number of input channels.
out_channels (int) – The number of output channels.
kernel_size (int or tuple) – The size of the convolution filters.
stride (int or tuple, optional) – The size of the stride when
applying the filter. Default: 1.
padding (int or tuple, optional) – How many positions to 0-pad +
padding (int or tuple, optional) – How many positions to 0-pad
the input with. Default: 0.
bias (bool, optional) – If True add a learnable bias to the
+
bias (bool, optional) – If True add a learnable bias to the
output. Default: True
-
+
+
@@ -371,6 +371,7 @@
Randomly zero a portion of the elements during training.
The remaining elements are multiplied with \(\frac{1}{1-p}\) where \(p\) is the probability of zeroing an element. This is done so the expected value of a given element will remain the same.
p (float) – The probability to zero an element
+p (float) – The probability to zero an element
Methods
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout2d.html b/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout2d.html index a104638e4..8302ccf1e 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout2d.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout2d.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Apply 2D channel-wise dropout during training.
Randomly zero out entire channels independently with probability \(p\). This layer expects the channels to be last, i.e. the input shape should be @@ -889,7 +891,7 @@ regularize activations. For more details, see [1].
Efficient Object Localization Using Convolutional Networks. CVPR 2015.p (float) – Probability of zeroing a channel during training.
+p (float) – Probability of zeroing a channel during training.
Methods
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout3d.html b/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout3d.html index e57654b0d..cfc3e8440 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout3d.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.Dropout3d.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Apply 3D channel-wise dropout during training.
Randomly zero out entire channels independently with probability \(p\). This layer expects the channels to be last, i.e., the input shape should be @@ -886,7 +888,7 @@ often beneficial for convolutional layers processing 3D data, like in medical imaging or video processing.
p (float) – Probability of zeroing a channel during training.
+p (float) – Probability of zeroing a channel during training.
Methods
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.ELU.html b/docs/build/html/python/nn/_autosummary/mlx.nn.ELU.html index 873fa41c3..84a995f5b 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.ELU.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.ELU.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
Implements a simple lookup table that maps each input integer to a high-dimensional vector.
Typically used to embed discrete tokens for processing by neural networks.
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies the gated linear unit function.
This function splits the axis dimension of the input into two halves
(\(a\) and \(b\)) and applies \(a * \sigma(b)\).
axis (int) – The dimension to split along. Default: -1
axis (int) – The dimension to split along. Default: -1
Methods
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.GRU.html b/docs/build/html/python/nn/_autosummary/mlx.nn.GRU.html index 70707e2d1..d247d6cc6 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.GRU.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.GRU.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
A gated recurrent unit (GRU) RNN layer.
The input has shape NLD or LD where:
- Parameters:
-input_size (int) – Dimension of the input, D.
hidden_size (int) – Dimension of the hidden state, H.
bias (bool) – Whether to use biases or not. Default: True.
input_size (int) – Dimension of the input, D.
hidden_size (int) – Dimension of the hidden state, H.
bias (bool) – Whether to use biases or not. Default: True.
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.GroupNorm.html b/docs/build/html/python/nn/_autosummary/mlx.nn.GroupNorm.html
index 9b84bfb15..55b81a012 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.GroupNorm.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.GroupNorm.html
@@ -8,7 +8,7 @@
- mlx.nn.GroupNorm — MLX 0.18.0 documentation
+ mlx.nn.GroupNorm — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -132,8 +132,8 @@
-
-
+
+
@@ -371,6 +371,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -426,6 +427,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
Applies Group Normalization [1] to the inputs.
Computes the same normalization as layer norm, namely
num_groups (int) – Number of groups to separate the features into
dims (int) – The feature dimensions of the input to normalize over
eps (float) – A small additive constant for numerical stability
affine (bool) – If True learn an affine transform to apply after the +
num_groups (int) – Number of groups to separate the features into
dims (int) – The feature dimensions of the input to normalize over
eps (float) – A small additive constant for numerical stability
affine (bool) – If True learn an affine transform to apply after the normalization.
pytorch_compatible (bool) – If True perform the group normalization in +
pytorch_compatible (bool) – If True perform the group normalization in the same order/grouping as PyTorch.
-
+
+
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies instance normalization [1] on the inputs.
Computes
- Parameters:
-dims (int) – The number of features of the input.
eps (float) – A value added to the denominator for numerical stability. Default: 1e-5.
affine (bool) – Default: False.
dims (int) – The number of features of the input.
eps (float) – A value added to the denominator for numerical stability. Default: 1e-5.
affine (bool) – Default: False.
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.LSTM.html b/docs/build/html/python/nn/_autosummary/mlx.nn.LSTM.html
index e6797f4dc..c1aeca511 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.LSTM.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.LSTM.html
@@ -8,7 +8,7 @@
- mlx.nn.LSTM — MLX 0.18.0 documentation
+ mlx.nn.LSTM — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -132,8 +132,8 @@
-
-
+
+
An LSTM recurrent layer.
The input has shape NLD or LD where:
- Parameters:
-input_size (int) – Dimension of the input, D.
hidden_size (int) – Dimension of the hidden state, H.
bias (bool) – Whether to use biases or not. Default: True.
input_size (int) – Dimension of the input, D.
hidden_size (int) – Dimension of the hidden state, H.
bias (bool) – Whether to use biases or not. Default: True.
-
+
+
@@ -371,6 +371,7 @@
Applies layer normalization [1] on the inputs.
Computes
dims (int) – The feature dimension of the input to normalize over
eps (float) – A small additive constant for numerical stability
affine (bool) – If True learn an affine transform to apply after the +
dims (int) – The feature dimension of the input to normalize over
eps (float) – A small additive constant for numerical stability
affine (bool) – If True learn an affine transform to apply after the normalization
bias (bool) – If True include a translation to the affine +
bias (bool) – If True include a translation to the affine transformation. If set to False the transformation is not really affine just scaling.
-
+
+
-
+
+
@@ -371,6 +371,7 @@
Applies an affine transformation to the input.
Concretely:
input_dims (int) – The dimensionality of the input features
output_dims (int) – The dimensionality of the output features
bias (bool, optional) – If set to False then the layer will
+
input_dims (int) – The dimensionality of the input features
output_dims (int) – The dimensionality of the output features
bias (bool, optional) – If set to False then the layer will
not use a bias. Default is True.
-
+
+
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies 1-dimensional max pooling.
Assuming an input of shape \((N, L, C)\) and kernel_size is
\(k\), the output is a tensor of shape \((N, L_{out}, C)\), given
@@ -887,10 +889,10 @@ by:
kernel_size (int or tuple(int)) – The size of the pooling window kernel.
stride (int or tuple(int), optional) – The stride of the pooling window. +
kernel_size (int or tuple(int)) – The size of the pooling window kernel.
stride (int or tuple(int), optional) – The stride of the pooling window.
Default: kernel_size.
padding (int or tuple(int), optional) – How much negative infinity +
padding (int or tuple(int), optional) – How much negative infinity
padding to apply to the input. The padding amount is applied to
both sides of the spatial axis. Default: 0.
-
+
+
@@ -371,6 +371,7 @@
Applies 2-dimensional max pooling.
Assuming an input of shape \((N, H, W, C)\) and kernel_size (int or tuple(int, int)) – The size of the pooling window. stride (int or tuple(int, int), optional) – The stride of the pooling
+ kernel_size (int or tuple(int, int)) – The size of the pooling window. stride (int or tuple(int, int), optional) – The stride of the pooling
window. Default: padding (int or tuple(int, int), optional) – How much negative infinity
+ padding (int or tuple(int, int), optional) – How much negative infinity
padding to apply to the input. The padding is applied on both sides
of the height and width axis. Default: kernel_size is
\((k_H, k_W)\), the output is a tensor of shape \((N, H_{out},
@@ -897,10 +899,10 @@ used for the height axis, the second
-
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.Mish.html b/docs/build/html/python/nn/_autosummary/mlx.nn.Mish.html
index 5bf528f0f..60d0e4a31 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.Mish.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.Mish.html
@@ -8,7 +8,7 @@
- kernel_size.0.
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
Map all the parameters using the provided map_fn and immediately
update the module with the mapped parameters.
For instance running model.apply(lambda x: x.astype(mx.float16))
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.Module.apply_to_modules.html b/docs/build/html/python/nn/_autosummary/mlx.nn.Module.apply_to_modules.html
index b5a2712f8..892b71b62 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.Module.apply_to_modules.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.Module.apply_to_modules.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Apply a function to all the modules in this instance (including this instance).
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Recursively filter the contents of the module using filter_fn,
namely only select keys and values where filter_fn returns true.
This is used to implement parameters() and trainable_parameters()
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.Module.freeze.html b/docs/build/html/python/nn/_autosummary/mlx.nn.Module.freeze.html
index d7c558b94..bac2441ca 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.Module.freeze.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.Module.freeze.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Freeze the Module’s parameters or some of them. Freezing a parameter means not computing gradients for it.
This function is idempotent i.e. freezing a frozen model is a no-op.
@@ -885,13 +887,13 @@ computing gradients for it.recurse (bool, optional) – If True then freeze the parameters of the +
recurse (bool, optional) – If True then freeze the parameters of the
submodules as well. Default: True.
keys (str or list[str], optional) – If provided then only these +
keys (str or list[str], optional) – If provided then only these
parameters will be frozen otherwise all the parameters of a
module. For instance freeze all biases by calling
module.freeze(keys="bias").
strict (bool, optional) – If set to True validate that the passed keys exist.
+
strict (bool, optional) – If set to True validate that the passed keys exist.
Default: False.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Update the model’s weights from a .npz, a .safetensors file, or a list.
file_or_weights (str or list(tuple(str, mx.array))) – The path to +
file_or_weights (str or list(tuple(str, mx.array))) – The path to
the weights .npz file (.npz or .safetensors) or a list
of pairs of parameter names and arrays.
strict (bool, optional) – If True then checks that the provided
+
strict (bool, optional) – If True then checks that the provided
weights exactly match the parameters of the model. Otherwise,
only the weights actually contained in the model are loaded and
shapes are not checked. Default: True.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Save the model’s weights to a file. The saving method is determined by the file extension:
- .npz will use mx.savez()
- .safetensors will use mx.save_safetensors()
-
+
+
@@ -370,6 +370,7 @@
dtype (Dtype) – The new dtype.
predicate (Callable, optional) – A predicate to select +
predicate (Callable, optional) – A predicate to select
parameters to cast. By default, only parameters of type
floating will be updated to avoid casting integer
parameters to the new dtype.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Set the model in or out of training mode.
Training mode only applies to certain layers. For example
Dropout applies a random mask in training mode, but is the
identity in evaluation mode.
mode (bool) – Indicate if the model should be in training or +
mode (bool) – Indicate if the model should be in training or
evaluation mode. Default: True.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Unfreeze the Module’s parameters or some of them.
This function is idempotent ie unfreezing a model that is not frozen is a noop.
@@ -885,13 +887,13 @@ a noop.recurse (bool, optional) – If True then unfreeze the parameters of the +
recurse (bool, optional) – If True then unfreeze the parameters of the
submodules as well. Default: True.
keys (str or list[str], optional) – If provided then only these +
keys (str or list[str], optional) – If provided then only these
parameters will be unfrozen otherwise all the parameters of a
module. For instance unfreeze all biases by calling
module.unfreeze(keys="bias").
strict (bool, optional) – If set to True validate that the passed keys exist.
+
strict (bool, optional) – If set to True validate that the passed keys exist.
Default: False.
-
+
+
@@ -370,6 +370,7 @@
Replace the parameters of this Module with the provided ones in the dict of dicts and lists.
Commonly used by the optimizer to change the model to the updated
@@ -882,7 +884,7 @@ similar to
parameters (dict) – A complete or partial dictionary of the modules
+ parameters (dict) – A complete or partial dictionary of the modules
parameters.
-
+
+
@@ -370,6 +370,7 @@
Replace the child modules of this Module instance with the
provided ones in the dict of dicts and lists.
It is the equivalent of modules (dict) – A complete or partial dictionary of the modules
+ modules (dict) – A complete or partial dictionary of the modules
submodules.Module.update() but for modules instead
@@ -882,7 +884,7 @@ similar to
-
+
+
@@ -370,6 +370,7 @@
Implements the scaled dot product attention with multiple heads.
Given inputs for queries, keys and values the MultiHeadAttention
produces new values by aggregating information from the input values
@@ -885,20 +887,20 @@ that should not be attended to.
dims (int) – The model dimensions. This is also the default +
dims (int) – The model dimensions. This is also the default value for the queries, keys, values, and the output.
num_heads (int) – The number of attention heads to use.
query_input_dims (int, optional) – The input dimensions of the queries. +
num_heads (int) – The number of attention heads to use.
query_input_dims (int, optional) – The input dimensions of the queries.
Default: dims.
key_input_dims (int, optional) – The input dimensions of the keys. +
key_input_dims (int, optional) – The input dimensions of the keys.
Default: dims.
value_input_dims (int, optional) – The input dimensions of the values. +
value_input_dims (int, optional) – The input dimensions of the values.
Default: key_input_dims.
value_dims (int, optional) – The dimensions of the values after the +
value_dims (int, optional) – The dimensions of the values after the
projection. Default: dims.
value_output_dims (int, optional) – The dimensions the new values will +
value_output_dims (int, optional) – The dimensions the new values will
be projected to. Default: dims.
bias (bool, optional) – Whether or not to use a bias in the projections. +
bias (bool, optional) – Whether or not to use a bias in the projections.
Default: False.
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
The same as Embedding but with a quantized weight matrix.
QuantizedEmbedding also provides a from_embedding()
classmethod to convert embedding layers to QuantizedEmbedding
@@ -879,12 +881,12 @@ layers.
num_embeddings (int) – How many possible discrete tokens can we embed. +
num_embeddings (int) – How many possible discrete tokens can we embed. Usually called the vocabulary size.
dims (int) – The dimensionality of the embeddings.
group_size (int, optional) – The group size to use for the quantized +
dims (int) – The dimensionality of the embeddings.
group_size (int, optional) – The group size to use for the quantized
weight. See quantize(). Default: 64.
bits (int, optional) – The bit width to use for the quantized weight. +
bits (int, optional) – The bit width to use for the quantized weight.
See quantize(). Default: 4.
-
+
+
@@ -370,6 +370,7 @@
Applies an affine transformation to the input using a quantized weight matrix.
It is the quantized equivalent of input_dims (int) – The dimensionality of the input features. output_dims (int) – The dimensionality of the output features. bias (bool, optional) – If set to input_dims (int) – The dimensionality of the input features. output_dims (int) – The dimensionality of the output features. bias (bool, optional) – If set to group_size (int, optional) – The group size to use for the quantized
+ group_size (int, optional) – The group size to use for the quantized
weight. See bits (int, optional) – The bit width to use for the quantized weight.
+ bits (int, optional) – The bit width to use for the quantized weight.
See mlx.nn.Linear. For now its
parameters are frozen and will not be included in any gradient computation
@@ -881,13 +883,13 @@ convert linear layers to
-
False then the layer will not use
+False then the layer will not use
a bias. Default: True.quantize(). Default: 64.quantize(). Default: 4.
-
+
+
@@ -371,6 +371,7 @@
Applies Root Mean Square normalization [1] to the inputs.
Computes
-
+
+
An Elman recurrent layer.
The input is a sequence of shape NLD or LD where:
- Parameters:
-input_size (int) – Dimension of the input, D.
hidden_size (int) – Dimension of the hidden state, H.
bias (bool, optional) – Whether to use a bias. Default: True.
input_size (int) – Dimension of the input, D.
hidden_size (int) – Dimension of the hidden state, H.
bias (bool, optional) – Whether to use a bias. Default: True.
nonlinearity (callable, optional) – Non-linearity to use. If None,
then func:tanh is used. Default: None.
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.ReLU.html b/docs/build/html/python/nn/_autosummary/mlx.nn.ReLU.html
index fbec908e7..1ede8c4cf 100644
--- a/docs/build/html/python/nn/_autosummary/mlx.nn.ReLU.html
+++ b/docs/build/html/python/nn/_autosummary/mlx.nn.ReLU.html
@@ -8,7 +8,7 @@
- mlx.nn.ReLU — MLX 0.18.0 documentation
+ mlx.nn.ReLU — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Implements the rotary positional encoding.
The traditional implementation rotates consecutive pairs of elements in the feature dimension while the default implementation rotates pairs with @@ -881,13 +883,13 @@ Embedding.
dims (int) – The feature dimensions to be rotated. If the input feature +
dims (int) – The feature dimensions to be rotated. If the input feature is larger than dims then the rest is left unchanged.
traditional (bool, optional) – If set to True choose the traditional
+
traditional (bool, optional) – If set to True choose the traditional
implementation which is slightly less efficient. Default: False.
base (float, optional) – The base used to compute angular frequency for +
base (float, optional) – The base used to compute angular frequency for
each dimension in the positional encodings. Default: 10000.
scale (float, optional) – The scale used to scale the positions. Default: 1.0.
scale (float, optional) – The scale used to scale the positions. Default: 1.0.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
nn.Module instances.
modules (tuple of Callables) – The modules to call in order
+modules (tuple of Callables) – The modules to call in order
Methods
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.SiLU.html b/docs/build/html/python/nn/_autosummary/mlx.nn.SiLU.html index c249b3a59..e917299ac 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.SiLU.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.SiLU.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
Implements sinusoidal positional encoding.
For more details see the paper Attention Is All You Need.
dims (int) – The dimensionality of the resulting positional embeddings.
min_freq (float, optional) – The minimum frequency expected. Default: +
dims (int) – The dimensionality of the resulting positional embeddings.
min_freq (float, optional) – The minimum frequency expected. Default:
0.0001.
max_freq (float, optional) – The maximum frequency expected. Default: +
max_freq (float, optional) – The maximum frequency expected. Default:
1.
scale (float, optional) – A multiplicative scale for the embeddings. +
scale (float, optional) – A multiplicative scale for the embeddings.
Default: sqrt(2/dims).
cos_first (bool, optional) – If True embed using [cos(x); sin(x)]
+
cos_first (bool, optional) – If True embed using [cos(x); sin(x)]
instead of the reverse. Default: False.
full_turns (bool, optional) – If True multiply the frequencies with
+
full_turns (bool, optional) – If True multiply the frequencies with
\(2\pi\). Default: False.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies the Step Activation Function.
This function implements a binary step activation, where the output is set to 1 if the input is greater than a specified threshold, and 0 otherwise.
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.Tanh.html b/docs/build/html/python/nn/_autosummary/mlx.nn.Tanh.html index 77192cc6f..527e4cbf1 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.Tanh.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.Tanh.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
dims (int, optional) – The number of expected features in the +
dims (int, optional) – The number of expected features in the
encoder/decoder inputs. Default: 512.
num_heads (int, optional) – The number of attention heads. Default: +
num_heads (int, optional) – The number of attention heads. Default:
8.
num_encoder_layers (int, optional) – The number of encoder layers in the +
num_encoder_layers (int, optional) – The number of encoder layers in the
Transformer encoder. Default: 6.
num_decoder_layers (int, optional) – The number of decoder layers in the +
num_decoder_layers (int, optional) – The number of decoder layers in the
Transformer decoder. Default: 6.
mlp_dims (int, optional) – The hidden dimension of the MLP block in each +
mlp_dims (int, optional) – The hidden dimension of the MLP block in each
Transformer layer. Defaults to 4*dims if not provided. Default:
None.
dropout (float, optional) – The dropout value for the Transformer +
dropout (float, optional) – The dropout value for the Transformer
encoder and decoder. Dropout is used after each attention layer and
the activation in the MLP layer. Default: 0.0.
activation (function, optional) – the activation function for the MLP @@ -901,10 +903,10 @@ hidden layer. Default: None.
custom_decoder (Module, optional) – A custom decoder to replace the
standard Transformer decoder. Default: None.
norm_first (bool, optional) – if True, encoder and decoder layers
+
norm_first (bool, optional) – if True, encoder and decoder layers
will perform layer normalization before attention and MLP
operations, otherwise after. Default: True.
checkpoint (bool, optional) – if True perform gradient checkpointing
+
checkpoint (bool, optional) – if True perform gradient checkpointing
to reduce the memory usage at the expense of more computation.
Default: False.
-
+
+
@@ -370,6 +370,7 @@
Upsample the input signal spatially.
The spatial dimensions are by convention dimensions 1 to x.ndim -
2. The first is the batch dimension and the last is the feature
@@ -893,13 +895,13 @@ output will be matching as will the bottom right edge.
scale_factor (float or tuple) – The multiplier for the spatial size. +
scale_factor (float or tuple) – The multiplier for the spatial size.
If a float is provided, it is the multiplier for all spatial dimensions.
Otherwise, the number of scale factors provided must match the
number of spatial dimensions.
mode (str, optional) – The upsampling algorithm, either "nearest",
+
mode (str, optional) – The upsampling algorithm, either "nearest",
"linear" or "cubic". Default: "nearest".
align_corners (bool, optional) – Changes the way the corners are treated +
align_corners (bool, optional) – Changes the way the corners are treated
during "linear" and "cubic" upsampling. See the note above and the
examples below for more details. Default: False.
-
+
+
@@ -370,6 +370,7 @@
An initializer that returns an array filled with value.
value (float) – The value to fill the array with.
value (float) – The value to fill the array with.
dtype (Dtype, optional) – The data type of the array. Default:
float32.
-
+
+
@@ -371,6 +371,7 @@
A Glorot normal initializer.
This initializer samples from a normal distribution with a standard
deviation computed from the number of input (fan_in) and output
@@ -891,7 +893,7 @@ with the same shape as the input, filled with samples from the Glorot
normal distribution.
Callable[[array, float], array]
+Example
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.init.glorot_uniform.html b/docs/build/html/python/nn/_autosummary/mlx.nn.init.glorot_uniform.html index e9d477d82..438fefdc2 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.init.glorot_uniform.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.init.glorot_uniform.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
A Glorot uniform initializer.
This initializer samples from a uniform distribution with a range
computed from the number of input (fan_in) and output (fan_out)
@@ -891,7 +893,7 @@ with the same shape as the input, filled with samples from the Glorot
uniform distribution.
Callable[[array, float], array]
+Example
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_normal.html b/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_normal.html index 95f34e469..45eff6fbe 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_normal.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_normal.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
Build a He normal initializer.
This initializer samples from a normal distribution with a standard
deviation computed from the number of input (fan_in) or output
@@ -894,7 +896,7 @@ array with the same shape as the input, filled with samples from the He
normal distribution.
Callable[[array, str, float], array]
+Example
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_uniform.html b/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_uniform.html index 219fc82ed..995be092a 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_uniform.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.init.he_uniform.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -371,6 +371,7 @@
A He uniform (Kaiming uniform) initializer.
This initializer samples from a uniform distribution with a range
computed from the number of input (fan_in) or output (fan_out)
@@ -894,7 +896,7 @@ array with the same shape as the input, filled with samples from the
He uniform distribution.
Callable[[array, str, float], array]
+Example
diff --git a/docs/build/html/python/nn/_autosummary/mlx.nn.init.identity.html b/docs/build/html/python/nn/_autosummary/mlx.nn.init.identity.html index b26ee973b..b65545f85 100644 --- a/docs/build/html/python/nn/_autosummary/mlx.nn.init.identity.html +++ b/docs/build/html/python/nn/_autosummary/mlx.nn.init.identity.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
An initializer that returns an identity matrix.
-
+
+
@@ -370,6 +370,7 @@
An initializer that returns samples from a normal distribution.
mean (float, optional) – Mean of the normal distribution. Default: +
mean (float, optional) – Mean of the normal distribution. Default:
0.0.
std (float, optional) – Standard deviation of the normal distribution. +
std (float, optional) – Standard deviation of the normal distribution.
Default: 1.0.
dtype (Dtype, optional) – The data type of the array. Default:
float32.
-
+
+
@@ -370,6 +370,7 @@
An initializer that returns samples from a uniform distribution.
low (float, optional) – The lower bound of the uniform distribution. +
low (float, optional) – The lower bound of the uniform distribution.
Default: 0.0.
high (float, optional) – The upper bound of the uniform distribution. +
high (float, optional) – The upper bound of the uniform distribution.
Default: 1.0
dtype (Dtype, optional) – The data type of the array. Default: float32.
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies the gated linear unit function.
This function splits the axis dimension of the input into two halves
(\(a\) and \(b\)) and applies \(a * \sigma(b)\).
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
Computes the binary cross entropy loss.
By default, this function takes the pre-sigmoid logits, which results in a faster
and more precise loss. For improved numerical stability when inputs (array) – The predicted values. If with_logits=False,
@@ -883,9 +885,9 @@ of -100with_logits is True, then
inputs are unnormalized logits. Otherwise, inputs are probabilities.
targets (array) – The binary target values in {0, 1}.
with_logits (bool, optional) – Whether inputs are logits. Default: True.
with_logits (bool, optional) – Whether inputs are logits. Default: True.
weights (array, optional) – Optional weights for each target. Default: None.
reduction (str, optional) – Specifies the reduction to apply to the output: +
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'mean'.
-
+
+
@@ -371,6 +371,7 @@
Computes the cosine similarity between the two inputs.
The cosine similarity loss is given by
x1 (mx.array) – The first set of inputs.
x2 (mx.array) – The second set of inputs.
axis (int, optional) – The embedding axis. Default: 1.
eps (float, optional) – The minimum value of the denominator used for +
axis (int, optional) – The embedding axis. Default: 1.
eps (float, optional) – The minimum value of the denominator used for
numerical stability. Default: 1e-8.
reduction (str, optional) – Specifies the reduction to apply to the output: +
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
Computes the cross entropy loss.
axistargets shape should be the same as
the logits shape.weights (array, optional) – Optional weights for each target. Default: None.
axis (int, optional) – The axis over which to compute softmax. Default: -1.
label_smoothing (float, optional) – Label smoothing factor. Default: 0.
reduction (str, optional) – Specifies the reduction to apply to the output: +
axis (int, optional) – The axis over which to compute softmax. Default: -1.
label_smoothing (float, optional) – Label smoothing factor. Default: 0.
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
@@ -371,6 +371,7 @@
Computes the negative log likelihood loss for a Gaussian distribution.
The loss is given by:
inputs (array) – The predicted expectation of the Gaussian distribution.
targets (array) – The target values (samples from the Gaussian distribution).
vars (array) – The predicted variance of the Gaussian distribution.
full (bool, optional) – Whether to include the constant term in the loss calculation. +
full (bool, optional) – Whether to include the constant term in the loss calculation.
Default: False.
eps (float, optional) – Small positive constant for numerical stability. +
eps (float, optional) – Small positive constant for numerical stability.
Default: 1e-6.
reduction (str, optional) – Specifies the reduction to apply to the output: +
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
Computes the hinge loss between inputs and targets.
-
+
+
@@ -371,6 +371,7 @@
Computes the Huber loss between inputs and targets.
inputs (array) – The predicted values.
targets (array) – The target values.
delta (float, optional) – The threshold at which to change between L1 and L2 loss. +
delta (float, optional) – The threshold at which to change between L1 and L2 loss.
Default: 1.0.
reduction (str, optional) – Specifies the reduction to apply to the output: +
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
Computes the Kullback-Leibler divergence loss.
Computes the following when reduction == 'none':
mx.exp(targets) * (targets - inputs).sum(axis)
@@ -882,8 +884,8 @@ document.write(`
inputs (array) – Log probabilities for the predicted distribution.
targets (array) – Log probabilities for the target distribution.
axis (int, optional) – The distribution axis. Default: -1.
reduction (str, optional) – Specifies the reduction to apply to the output: +
axis (int, optional) – The distribution axis. Default: -1.
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
Computes the L1 loss.
-
+
+
Computes the log cosh loss between inputs and targets.
Logcosh acts like L2 loss for small errors, ensuring stable gradients, and like the L1 loss for large errors, reducing sensitivity to outliers. This @@ -886,7 +888,7 @@ dual behavior offers a balanced, robust approach for regression tasks.
-
+
+
@@ -371,6 +371,7 @@
Calculate the margin ranking loss that loss given inputs \(x_1\), \(x_2\) and a label \(y\) (containing 1 or -1).
The loss is given by:
@@ -887,9 +889,9 @@ representsinputs2<
inputs2 (array) – Scores for the second input.
targets (array) – Labels indicating whether samples in inputs1 should be ranked higher
than samples in inputs2. Values should be 1 or -1.
margin (float, optional) – The margin by which the scores should be separated.
+
margin (float, optional) – The margin by which the scores should be separated.
Default: 0.0.
reduction (str, optional) – Specifies the reduction to apply to the output:
+
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
@@ -370,6 +370,7 @@
Computes the mean squared error loss.
-
+
+
@@ -370,6 +370,7 @@
Computes the negative log likelihood loss.
inputs (array) – The predicted distribution in log space.
targets (array) – The target values.
axis (int, optional) – The distribution axis. Default: -1.
reduction (str, optional) – Specifies the reduction to apply to the output: +
axis (int, optional) – The distribution axis. Default: -1.
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
@@ -371,6 +371,7 @@
Computes the smooth L1 loss.
The smooth L1 loss is a variant of the L1 loss which replaces the absolute
difference with a squared difference when the absolute difference is less
@@ -888,9 +890,9 @@ than predictions (array) – Predicted values. targets (array) – Ground truth values. beta (float, optional) – The threshold after which the loss changes
+ beta (float, optional) – The threshold after which the loss changes
from the squared to the absolute difference. Default: reduction (str, optional) – Specifies the reduction to apply to the output:
+ reduction (str, optional) – Specifies the reduction to apply to the output:
beta
1.0.'none' | 'mean' | 'sum'. Default: 'mean'.
-
+
+
@@ -371,6 +371,7 @@
Computes the triplet loss for a set of anchor, positive, and negative samples. Margin is represented with alpha in the math section.
anchors (array) – The anchor samples.
positives (array) – The positive samples.
negatives (array) – The negative samples.
axis (int, optional) – The distribution axis. Default: -1.
p (int, optional) – The norm degree for pairwise distance. Default: 2.
margin (float, optional) – Margin for the triplet loss. Defaults to 1.0.
eps (float, optional) – Small positive constant to prevent numerical instability. Defaults to 1e-6.
reduction (str, optional) – Specifies the reduction to apply to the output: +
axis (int, optional) – The distribution axis. Default: -1.
p (int, optional) – The norm degree for pairwise distance. Default: 2.
margin (float, optional) – Margin for the triplet loss. Defaults to 1.0.
eps (float, optional) – Small positive constant to prevent numerical instability. Defaults to 1e-6.
reduction (str, optional) – Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. Default: 'none'.
-
+
+
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -371,6 +371,7 @@
Applies the Softshrink activation function.
-
+
+
Applies the Step Activation Function.
This function implements a binary step activation, where the output is set to 1 if the input is greater than a specified threshold, and 0 otherwise.
diff --git a/docs/build/html/python/nn/_autosummary_functions/mlx.nn.tanh.html b/docs/build/html/python/nn/_autosummary_functions/mlx.nn.tanh.html index 5553aa66d..dba3564fa 100644 --- a/docs/build/html/python/nn/_autosummary_functions/mlx.nn.tanh.html +++ b/docs/build/html/python/nn/_autosummary_functions/mlx.nn.tanh.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
An or reduction over the given axes.
arange(-> array)
Generates ranges of numbers.
Overloaded function.
arccos(a, /, *[, stream])
Element-wise inverse cosine.
right_shift(a, b[, stream])
Element-wise right shift.
round(a, /[, decimals, stream])
roll(a, shift[, axis, stream])
Roll array elements along a given axis.
round(a, /[, decimals, stream])
Round to the given number of decimals.
rsqrt(a, /, *[, stream])
rsqrt(a, /, *[, stream])
Element-wise reciprocal and square root.
save(file, arr)
save(file, arr)
Save the array to a binary file in .npy format.
savez(file, *args, **kwargs)
savez(file, *args, **kwargs)
Save several arrays to a binary file in uncompressed .npz format.
savez_compressed(file, *args, **kwargs)
savez_compressed(file, *args, **kwargs)
Save several arrays to a binary file in compressed .npz format.
save_gguf(file, arrays, metadata)
save_gguf(file, arrays, metadata)
Save array(s) to a binary file in .gguf format.
save_safetensors(file, arrays[, metadata])
save_safetensors(file, arrays[, metadata])
Save array(s) to a binary file in .safetensors format.
sigmoid(a, /, *[, stream])
sigmoid(a, /, *[, stream])
Element-wise logistic sigmoid.
sign(a, /, *[, stream])
sign(a, /, *[, stream])
Element-wise sign.
sin(a, /, *[, stream])
sin(a, /, *[, stream])
Element-wise sine.
sinh(a, /, *[, stream])
sinh(a, /, *[, stream])
Element-wise hyperbolic sine.
softmax(a, /[, axis, stream])
softmax(a, /[, axis, stream])
Perform the softmax along the given axis.
sort(a, /[, axis, stream])
sort(a, /[, axis, stream])
Returns a sorted copy of the array.
split(a, /, indices_or_sections[, axis, stream])
split(a, /, indices_or_sections[, axis, stream])
Split an array along a given axis.
sqrt(a, /, *[, stream])
sqrt(a, /, *[, stream])
Element-wise square root.
square(a, /, *[, stream])
square(a, /, *[, stream])
Element-wise square.
squeeze(a, /[, axis, stream])
squeeze(a, /[, axis, stream])
Remove length one axes from an array.
stack(arrays[, axis, stream])
stack(arrays[, axis, stream])
Stacks the arrays along a new axis.
std(a, /[, axis, keepdims, ddof, stream])
std(a, /[, axis, keepdims, ddof, stream])
Compute the standard deviation(s) over the given axes.
stop_gradient(a, /, *[, stream])
stop_gradient(a, /, *[, stream])
Stop gradients from being computed.
subtract(a, b[, stream])
subtract(a, b[, stream])
Element-wise subtraction.
sum(a, /[, axis, keepdims, stream])
sum(a, /[, axis, keepdims, stream])
Sum reduce the array over the given axes.
swapaxes(a, /, axis1, axis2, *[, stream])
swapaxes(a, /, axis1, axis2, *[, stream])
Swap two axes of an array.
take(a, /, indices[, axis, stream])
take(a, /, indices[, axis, stream])
Take elements along an axis.
take_along_axis(a, /, indices[, axis, stream])
take_along_axis(a, /, indices[, axis, stream])
Take values along an axis at the specified indices.
tan(a, /, *[, stream])
tan(a, /, *[, stream])
Element-wise tangent.
tanh(a, /, *[, stream])
tanh(a, /, *[, stream])
Element-wise hyperbolic tangent.
tensordot(a, b, /[, axes, stream])
tensordot(a, b, /[, axes, stream])
Compute the tensor dot product along the specified axes.
tile(a, reps, /, *[, stream])
tile(a, reps, /, *[, stream])
Construct an array by repeating a the number of times given by reps.
topk(a, /, k[, axis, stream])
topk(a, /, k[, axis, stream])
Returns the k largest elements from the input along a given axis.
trace(a, /[, offset, axis1, axis2, dtype, ...])
trace(a, /[, offset, axis1, axis2, dtype, ...])
Return the sum along a specified diagonal in the given array.
transpose(a, /[, axes, stream])
transpose(a, /[, axes, stream])
Transpose the dimensions of the array.
tri(n, m, k[, dtype, stream])
tri(n, m, k[, dtype, stream])
An array with ones at and below the given diagonal and zeros elsewhere.
tril(x, k, *[, stream])
tril(x, k, *[, stream])
Zeros the array above the given diagonal.
triu(x, k, *[, stream])
triu(x, k, *[, stream])
Zeros the array below the given diagonal.
var(a, /[, axis, keepdims, ddof, stream])
var(a, /[, axis, keepdims, ddof, stream])
Compute the variance(s) over the given axes.
view(a, dtype[, stream])
view(a, dtype[, stream])
View the array as a different type.
where(condition, x, y, /, *[, stream])
where(condition, x, y, /, *[, stream])
Select from x or y according to condition.
zeros(shape[, dtype, stream])
zeros(shape[, dtype, stream])
Construct an array of zeros.
zeros_like(a, /, *[, stream])
zeros_like(a, /, *[, stream])
An array of zeros like the input.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -371,6 +371,7 @@
The AdaDelta optimizer with a learning rate [1].
Our AdaDelta implementation follows the original paper. In detail,
[1]: Zeiler, M.D., 2012. ADADELTA: an adaptive learning rate method. arXiv preprint arXiv:1212.5701.
@@ -884,10 +886,10 @@ w_{t+1} &= w_t - \lambda \Delta w_{t+1}\end{split}\]learning_rate (float or callable) – The learning rate \(\lambda\).
rho (float, optional) – The coefficient \(\rho\) used for computing a +
learning_rate (float or callable) – The learning rate \(\lambda\).
rho (float, optional) – The coefficient \(\rho\) used for computing a
running average of squared gradients. Default: 0.9
eps (float, optional) – The term \(\epsilon\) added to the denominator to improve +
eps (float, optional) – The term \(\epsilon\) added to the denominator to improve numerical stability. Default: 1e-8
-
+
+
@@ -371,6 +371,7 @@
The Adafactor optimizer.
Our Adafactor implementation follows the original paper: Adafactor: Adaptive Learning Rates with Sublinear Memory Cost
learning_rate (float or callable, optional) – The learning rate. +
learning_rate (float or callable, optional) – The learning rate.
Default: None.
eps (tuple(float, float), optional) – The first term \(\epsilon_1\) +
eps (tuple(float, float), optional) – The first term \(\epsilon_1\)
added to the square of the gradients to improve numerical
stability and the second term \(\epsilon_2\) is used for
parameter scaling if parameter_scale is set to True.
Default: (1e-30, 1e-3).
clip_threshold (float, optional) – Clips the unscaled update at +
clip_threshold (float, optional) – Clips the unscaled update at
clip_threshold. Default: 1.0.
decay_rate (float, optional) – Coefficient for the running average +
decay_rate (float, optional) – Coefficient for the running average
of the squared gradient. Default: -0.8.
beta_1 (float, optional) – If set to a value bigger than zero +
beta_1 (float, optional) – If set to a value bigger than zero
then first moment will be used. Default: None.
weight_decay (float, optional) – The weight decay \(\lambda\). +
weight_decay (float, optional) – The weight decay \(\lambda\).
Default: 0.0.
scale_parameter (bool, optional) – If set to True the learning rate
+
scale_parameter (bool, optional) – If set to True the learning rate
will be scaled by \(\max(\epsilon_1, \text{RMS}(w_{t-1}))\).
Default: True.
relative_step (bool, optional) – If set to True the learning_rate
+
relative_step (bool, optional) – If set to True the learning_rate
will be ignored and relative step size will be computed.
Default: True.
warmup_init (bool, optional) – If set to True then the relative
+
warmup_init (bool, optional) – If set to True then the relative
step size will be calculated by the current step. Default:
False.
-
+
+
@@ -371,6 +371,7 @@
The Adagrad optimizer [1].
Our Adagrad implementation follows the original paper. In detail,
[1]: Duchi, J., Hazan, E. and Singer, Y., 2011. Adaptive subgradient methods @@ -883,8 +885,8 @@ w_{t+1} &= w_t - \lambda \frac{g_t}{\sqrt{v_{t+1}} + \epsilon}\end{split}\]<
learning_rate (float or callable) – The learning rate \(\lambda\).
eps (float, optional) – The term \(\epsilon\) added to the +
learning_rate (float or callable) – The learning rate \(\lambda\).
eps (float, optional) – The term \(\epsilon\) added to the
denominator to improve numerical stability. Default: 1e-8
-
+
+
@@ -371,6 +371,7 @@
The Adam optimizer [1].
Our Adam implementation follows the original paper and omits the bias correction in the first and second moment estimates. In detail,
@@ -885,11 +887,11 @@ w_{t+1} &= w_t - \lambda \frac{m_{t+1}}{\sqrt{v_{t+1} + \epsilon}}\end{splitlearning_rate (float or callable) – The learning rate \(\lambda\).
learning_rate (float or callable) – The learning rate \(\lambda\).
betas (Tuple[float, float], optional) – The coefficients
\((\beta_1, \beta_2)\) used for computing running averages of the
gradient and its square. Default: (0.9, 0.999)
eps (float, optional) – The term \(\epsilon\) added to the +
eps (float, optional) – The term \(\epsilon\) added to the
denominator to improve numerical stability. Default: 1e-8
-
+
+
@@ -371,6 +371,7 @@
The AdamW optimizer [1].
Following the above convention, in contrast with [1], we do not use bias correction in the first and second moments for AdamW. We update the weights @@ -886,13 +888,13 @@ w_{t+1} &= w_t - \alpha (\frac{m_{t+1}}{\sqrt{v_{t+1} + \epsilon}} + \lambda
learning_rate (float or callable) – The learning rate \(\alpha\).
learning_rate (float or callable) – The learning rate \(\alpha\).
betas (Tuple[float, float], optional) – The coefficients
\((\beta_1, \beta_2)\) used for computing running averages of the
gradient and its square. Default: (0.9, 0.999)
eps (float, optional) – The term \(\epsilon\) added to the +
eps (float, optional) – The term \(\epsilon\) added to the
denominator to improve numerical stability. Default: 1e-8
weight_decay (float, optional) – The weight decay \(\lambda\). +
weight_decay (float, optional) – The weight decay \(\lambda\).
Default: 0.
-
+
+
@@ -371,6 +371,7 @@
The Adamax optimizer, a variant of Adam based on the infinity norm [1].
Our Adam implementation follows the original paper and omits the bias correction in the first and second moment estimates. In detail,
@@ -885,11 +887,11 @@ w_{t+1} &= w_t - \lambda \frac{m_{t+1}}{v_{t+1} + \epsilon}\end{split}\]learning_rate (float or callable) – The learning rate \(\lambda\).
learning_rate (float or callable) – The learning rate \(\lambda\).
betas (Tuple[float, float], optional) – The coefficients
\((\beta_1, \beta_2)\) used for computing running averages of the
gradient and its square. Default: (0.9, 0.999)
eps (float, optional) – The term \(\epsilon\) added to the +
eps (float, optional) – The term \(\epsilon\) added to the
denominator to improve numerical stability. Default: 1e-8
-
+
+
@@ -371,6 +371,7 @@
The Lion optimizer [1].
Since updates are computed through the sign operation, they tend to have larger norm than for other optimizers such as SGD and Adam. @@ -889,11 +891,11 @@ w_{t + 1} &= w_t - \eta (\text{sign}(c_t) + \lambda w_t)\end{split}\]
learning_rate (float or callable) – The learning rate \(\eta\).
learning_rate (float or callable) – The learning rate \(\eta\).
betas (Tuple[float, float], optional) – The coefficients
\((\beta_1, \beta_2)\) used for computing the gradient
momentum and update direction. Default: (0.9, 0.99)
weight_decay (float, optional) – The weight decay \(\lambda\). Default: 0.0
weight_decay (float, optional) – The weight decay \(\lambda\). Default: 0.0
-
+
+
@@ -370,6 +370,7 @@
Apply the gradients to the parameters and return the updated parameters.
Can be used to update a model via
gradients (dict) – A Python tree of gradients. parameters (dict) – A Python tree of parameters. It can be a
+ gradients (dict) – A Python tree of gradients. parameters (dict) – A Python tree of parameters. It can be a
superset of the gradients. In that case the returned python
tree will be of the same structure as the gradients.model.update(opt.apply_gradients(grads, model)) which is precisely
@@ -879,8 +881,8 @@ how
-
diff --git a/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.init.html b/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.init.html
index ee65ecd12..997b30e86 100644
--- a/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.init.html
+++ b/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.init.html
@@ -8,7 +8,7 @@
-
-
+
+
Initialize the optimizer’s state
This function can be used to initialize optimizers which have state
(like momentum in SGD). Using this method is optional as the
@@ -881,7 +883,7 @@ to have access to the Optimizer.update().
model (dict) – A Python tree of parameters.
+model (dict) – A Python tree of parameters.
Example
diff --git a/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.state.html b/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.state.html index aea35ba7c..c7a053c2f 100644 --- a/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.state.html +++ b/docs/build/html/python/optimizers/_autosummary/mlx.optimizers.Optimizer.state.html @@ -8,7 +8,7 @@ -
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
Apply the gradients to the parameters of the model and update the model with the new parameters.
model (Module) – An mlx module to be updated.
gradients (dict) – A Python tree of gradients, most likely computed +
gradients (dict) – A Python tree of gradients, most likely computed
via mlx.nn.value_and_grad().
-
+
+
@@ -371,6 +371,7 @@
The RMSprop optimizer [1].
[1]: Tieleman, T. and Hinton, G. 2012. Lecture 6.5-rmsprop, coursera: Neural networks for machine learning
learning_rate (float or callable) – The learning rate \(\lambda\).
alpha (float, optional) – The smoothing constant \(\alpha\). +
learning_rate (float or callable) – The learning rate \(\lambda\).
alpha (float, optional) – The smoothing constant \(\alpha\).
Default: 0.99
eps (float, optional) – The term \(\epsilon\) added to the denominator +
eps (float, optional) – The term \(\epsilon\) added to the denominator
to improve numerical stability. Default: 1e-8
-
+
+
The stochastic gradient descent optimizer.
Updates a parameter \(w\) with a gradient \(g\) as follows
learning_rate (float or callable) – The learning rate \(\lambda\).
momentum (float, optional) – The momentum strength \(\mu\). Default: 0
weight_decay (float, optional) – The weight decay (L2 penalty). Default: 0
dampening (float, optional) – Dampening for momentum \(\tau\). Default: 0
nesterov (bool, optional) – Enables Nesterov momentum. Default: False
learning_rate (float or callable) – The learning rate \(\lambda\).
momentum (float, optional) – The momentum strength \(\mu\). Default: 0
weight_decay (float, optional) – The weight decay (L2 penalty). Default: 0
dampening (float, optional) – Dampening for momentum \(\tau\). Default: 0
nesterov (bool, optional) – Enables Nesterov momentum. Default: False
-
+
+
@@ -370,6 +370,7 @@
Make a cosine decay scheduler.
init (float) – Initial value.
decay_steps (int) – Number of steps to decay over. The decayed +
init (float) – Initial value.
decay_steps (int) – Number of steps to decay over. The decayed
value is constant for steps beyond decay_steps.
end (float, optional) – Final value to decay to. Default: 0.
end (float, optional) – Final value to decay to. Default: 0.
-
+
+
@@ -370,6 +370,7 @@
Make an exponential decay scheduler.
-
+
+
@@ -371,6 +371,7 @@
Join multiple schedules to create a new schedule.
schedules (list(Callable)) – A list of schedules. Schedule \(i+1\) +
schedules (list(Callable)) – A list of schedules. Schedule \(i+1\) receives a step count indicating the number of steps since the \(i\)-th boundary.
boundaries (list(int)) – A list of integers of length len(schedules) - 1
+
boundaries (list(int)) – A list of integers of length len(schedules) - 1
that indicates when to transition between schedules.
-
+
+
@@ -370,6 +370,7 @@
Make a linear scheduler.
-
+
+
@@ -370,6 +370,7 @@
Make a step decay scheduler.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
laplace([shape, dtype, loc, scale, key, stream])
Sample numbers from a Laplace distribution.
permutation(x[, axis, key, stream])
Generate a random permutation or permute the entries of an array.
-
+
+
@@ -370,6 +370,7 @@
-
+
+
-
+
+
@@ -369,6 +369,7 @@
|
- MLX
-
- |
-
This is the complete list of members for mlx::core::ContiguousIterator< StrideT >, including all inherited members.
-| ContiguousIterator(const std::vector< int > &shape, const std::vector< StrideT > &strides, int dims) | mlx::core::ContiguousIterator< StrideT > | inlineexplicit |
| loc | mlx::core::ContiguousIterator< StrideT > | |
| step() | mlx::core::ContiguousIterator< StrideT > | inline |
1.10.0
-
-
-
diff --git a/docs/build/html/structmlx_1_1core_1_1_contiguous_iterator.html b/docs/build/html/structmlx_1_1core_1_1_contiguous_iterator.html
deleted file mode 100644
index 253c066ab..000000000
--- a/docs/build/html/structmlx_1_1core_1_1_contiguous_iterator.html
+++ /dev/null
@@ -1,191 +0,0 @@
-
-
-
-
-
-
-
-|
- MLX
-
- |
-
#include <utils.h>
-Public Member Functions | |
| void | step () |
| ContiguousIterator (const std::vector< int > &shape, const std::vector< StrideT > &strides, int dims) | |
-Public Attributes | |
| StrideT | loc {0} |
-
|
- -inlineexplicit | -
-
|
- -inline | -
| StrideT mlx::core::ContiguousIterator< StrideT >::loc {0} | -
1.10.0
-
-
-
diff --git a/docs/build/html/usage/compile.html b/docs/build/html/usage/compile.html
index 020f25cb7..9b9dd771d 100644
--- a/docs/build/html/usage/compile.html
+++ b/docs/build/html/usage/compile.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
-
+
+
@@ -370,6 +370,7 @@
list, tuple, or
-dict).
+containers of arrays (specifically any of list, tuple, or
+dict).
Suppose we wanted a weight and a bias parameter in the above example. A nice way to do that is the following:
def loss_fn(params, x, y):
diff --git a/docs/build/html/usage/indexing.html b/docs/build/html/usage/indexing.html
index cc8941daa..bd5dde991 100644
--- a/docs/build/html/usage/indexing.html
+++ b/docs/build/html/usage/indexing.html
@@ -8,7 +8,7 @@
- Indexing Arrays — MLX 0.18.0 documentation
+ Indexing Arrays — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
For the most part, indexing an MLX array works the same as indexing a
NumPy numpy.ndarray. See the NumPy documentation for more details on
how that works.
For example, you can use regular integers and slices (slice) to index arrays:
For example, you can use regular integers and slices (slice) to index arrays:
>>> arr = mx.arange(10)
>>> arr[3]
array(3, dtype=int32)
@@ -883,7 +885,7 @@ array(8, ([2, 4, 6], dtype=int32)
For multi-dimensional arrays, the ... or Ellipsis syntax works as in NumPy:
For multi-dimensional arrays, the ... or Ellipsis syntax works as in NumPy:
>>> arr = mx.arange(8).reshape(2, 2, 2)
>>> arr[:, :, 0]
array(3, dtype=int32)
@@ -909,7 +911,7 @@ array([[0,
array([5, 7], dtype=int32)
Mixing and matching integers, slice, ..., and array indices
+
Mixing and matching integers, slice, ..., and array indices
works just as in NumPy.
Other functions which may be useful for indexing arrays are take() and
take_along_axis().
-
+
+
An important behavior to be aware of is when the graph will be implicitly
evaluated. Anytime you print an array, convert it to an
-numpy.ndarray, or otherwise access it’s memory via memoryview,
+numpy.ndarray, or otherwise access it’s memory via memoryview,
the graph will be evaluated. Saving arrays via save() (or any other MLX
saving functions) will also evaluate the array.
Calling array.item() on a scalar array will also evaluate it. In the
diff --git a/docs/build/html/usage/numpy.html b/docs/build/html/usage/numpy.html
index 889a18ad8..156381858 100644
--- a/docs/build/html/usage/numpy.html
+++ b/docs/build/html/usage/numpy.html
@@ -8,7 +8,7 @@
-
-
+
+
@@ -370,6 +370,7 @@
Warning
-PyTorch Support for memoryview is experimental and can break for
+
PyTorch Support for memoryview is experimental and can break for
multi-dimensional arrays. Casting to NumPy first is advised for now.
PyTorch supports the buffer protocol, but it requires an explicit memoryview.
PyTorch supports the buffer protocol, but it requires an explicit memoryview.
import mlx.core as mx
import torch
@@ -953,7 +955,7 @@ multi-dimensional arrays. Casting to NumPy first is advised for now.
TensorFlow#
-TensorFlow supports the buffer protocol, but it requires an explicit memoryview.
+TensorFlow supports the buffer protocol, but it requires an explicit memoryview.
import mlx.core as mx
import tensorflow as tf
diff --git a/docs/build/html/usage/quick_start.html b/docs/build/html/usage/quick_start.html
index 60c37651f..7e427ee26 100644
--- a/docs/build/html/usage/quick_start.html
+++ b/docs/build/html/usage/quick_start.html
@@ -8,7 +8,7 @@
- Quick Start Guide — MLX 0.18.0 documentation
+ Quick Start Guide — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
- Transforms
diff --git a/docs/build/html/usage/saving_and_loading.html b/docs/build/html/usage/saving_and_loading.html
index 210a30b87..dd4ef34b8 100644
--- a/docs/build/html/usage/saving_and_loading.html
+++ b/docs/build/html/usage/saving_and_loading.html
@@ -8,7 +8,7 @@
- Saving and Loading Arrays — MLX 0.18.0 documentation
+ Saving and Loading Arrays — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
@@ -370,6 +370,7 @@
- mlx.core.repeat
- mlx.core.reshape
- mlx.core.right_shift
+- mlx.core.roll
- mlx.core.round
- mlx.core.rsqrt
- mlx.core.save
@@ -425,6 +426,7 @@
- mlx.core.random.truncated_normal
- mlx.core.random.uniform
- mlx.core.random.laplace
+- mlx.core.random.permutation
- Transforms
@@ -928,7 +930,7 @@ provided. This can be loaded with:
In this case load() returns a dictionary of names to arrays.
The functions save_safetensors() and save_gguf() are similar to
-savez(), but they take as input a dict of string names to arrays:
savez(), but they take as input a dict of string names to arrays:
>>> a = mx.array([1.0])
>>> b = mx.array([2.0])
>>> mx.save_safetensors("arrays", {"a": a, "b": b})
diff --git a/docs/build/html/usage/unified_memory.html b/docs/build/html/usage/unified_memory.html
index d1aa37ae7..35345369a 100644
--- a/docs/build/html/usage/unified_memory.html
+++ b/docs/build/html/usage/unified_memory.html
@@ -8,7 +8,7 @@
- Unified Memory — MLX 0.18.0 documentation
+ Unified Memory — MLX 0.18.1 documentation
@@ -36,7 +36,7 @@
-
+
@@ -131,8 +131,8 @@
-
-
+
+
-
+
+