Awni Hannun e142aaf8a1 Option for precise softmax (#953) ...

* precise softmax

* Add an equivalency check

* Make the threadgroup memory definition fixed

* precise cpu softmax

* precise option on cpu

* remove print

---------

Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>

2024-04-04 08:32:35 -07:00

..

3rdparty

jagrit's commit files

2023-11-29 10:52:08 -08:00

backend

Option for precise softmax (#953)

2024-04-04 08:32:35 -07:00

io

fix a couple bugs (#952)

2024-04-02 12:07:41 -07:00

types

Do not use static constexpr in header (#863)

2024-03-20 21:28:05 -07:00

allocator.cpp

Memory allocation (#292)

2024-01-02 11:59:19 -08:00

allocator.h

Memory allocation (#292)

2024-01-02 11:59:19 -08:00

array.cpp

Reduce a little overhead (#871)

2024-03-22 17:29:36 -07:00

array.h

Fix race in multi-stream eval (#911)

2024-03-26 16:36:36 -07:00

CMakeLists.txt

Custom primitive + RoPE fat op (#676)

2024-02-14 14:04:25 -08:00

compile_impl.h

Ios compile (#784)

2024-03-04 20:02:26 -08:00

compile.cpp

Fix compile fusion for multi-output edge cases (#950)

2024-04-02 08:42:31 -07:00

compile.h

Shapeless compilation for some graphs (#687)

2024-02-19 21:43:54 -08:00

device.cpp

copyright + ack

2023-11-30 11:12:53 -08:00

device.h

copyright + ack

2023-11-30 11:12:53 -08:00

dtype.cpp

Remove unnecessary string copies (#891)

2024-03-28 13:14:59 -07:00

dtype.h

Remove unnecessary string copies (#891)

2024-03-28 13:14:59 -07:00

fast_primitives.h

Implement vjps for some primitives in the fast namespace (#883)

2024-03-26 16:35:34 -07:00

fast.cpp

Option for precise softmax (#953)

2024-04-04 08:32:35 -07:00

fast.h

Adds mx.fast.layer_norm (#870)

2024-03-21 13:55:51 -07:00

fft.cpp

Don't use make_unique to create shared_ptr (#902)

2024-03-27 06:13:29 -07:00

fft.h

Remove duplicate defines of StreamOrDevice and is_big_endian (#892)

2024-03-26 15:15:11 -07:00

graph_utils.cpp

Remove unnecessary string copies (#891)

2024-03-28 13:14:59 -07:00

graph_utils.h

Add missing && when forwarding args (#894)

2024-03-25 14:55:54 -07:00

io.h

Remove unnecessary string copies (#891)

2024-03-28 13:14:59 -07:00

linalg.cpp

Remove unnecessary string copies (#891)

2024-03-28 13:14:59 -07:00

linalg.h

Remove unnecessary string copies (#891)

2024-03-28 13:14:59 -07:00

mlx.h

Custom primitive + RoPE fat op (#676)

2024-02-14 14:04:25 -08:00

ops.cpp

Option for precise softmax (#953)

2024-04-04 08:32:35 -07:00

ops.h

Option for precise softmax (#953)

2024-04-04 08:32:35 -07:00

primitives.cpp

Option for precise softmax (#953)

2024-04-04 08:32:35 -07:00

primitives.h

Option for precise softmax (#953)

2024-04-04 08:32:35 -07:00

random.cpp

Fixing random.normal for half-precision dtype #642 (#904)

2024-03-26 09:58:27 -07:00

random.h

Add loc and scale to random.normal (#638)

2024-02-07 11:49:59 -08:00

scheduler.cpp

copyright + ack

2023-11-30 11:12:53 -08:00

scheduler.h

move all ObjC (via metal-cpp) interaction until post static initializers (#370)

2024-01-04 16:12:00 -08:00

stream.h

copyright + ack

2023-11-30 11:12:53 -08:00

transforms_impl.h

Use uintptr_t instead of size_t to store funtion id (#916)

2024-03-28 06:37:59 -07:00

transforms.cpp

Add Metal debug option and capture functions (#707)

2024-03-28 09:40:31 -07:00

transforms.h

Add missing && when forwarding args (#894)

2024-03-25 14:55:54 -07:00

utils.cpp

Add a SliceUpdate op and primitive (#850)

2024-03-20 10:39:25 -07:00

utils.h

Remove duplicate defines of StreamOrDevice and is_big_endian (#892)

2024-03-26 15:15:11 -07:00