Check at runtime if the CPU supports native bfloat16 (FEAT_BF16). This allows us to compile once for all Macs but still enable the feature if it is s

Return true if we are in a dynamic (shapeless) trace used for compiling or * exporting graphs with dynamic shapes. */

Return the submodules that do not contain other modules.

https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html#1202

Helper like below but only merges the two provided arrays. If the src has siblings then these won't be merged to the dst.

MLX loop-of-matmuls baseline.

The number of bytes in the array. */

Helper for the ndimensional strided loop

Quantize the sub-modules of a module according to a predicate. By default all layers that define a ``to_quantized()`` method will be quantize

Computes: O = softmax(Q @ K.T) @ V **/

* Get the size of the corresponding dimension. * * This function supports negative indexing and provides * bounds checking. */

Fill a pytree (recursive dict or list of dict or list) in place with the given arrays Non dict or list nodes are ignored

Applies a reduction to the leaves of a Python tree. This function reduces Python trees into an accumulated result by applying the provided fu

Check if the event is valid