| .. |
|
layers
|
8c20cfef49
[Rotary] Support qkv block layout from GQA
|
3 months ago |
|
losses
|
c7f32a8409
[CrossEntropy] Support precomputed LSE
|
3 months ago |
|
models
|
30e1ef0f79
minify torch.torch.int32 to torch.int32 (#1237)
|
2 months ago |
|
modules
|
3f1b4d38e7
Fix: check the type of max_seqlen_k instead of checking max_seqlen twice (#1127)
|
4 months ago |
|
ops
|
8c20cfef49
[Rotary] Support qkv block layout from GQA
|
3 months ago |
|
utils
|
320fb59487
Update citation
|
6 months ago |
|
__init__.py
|
418d677192
Bump to v2.6.3
|
4 months ago |
|
bert_padding.py
|
30e1ef0f79
minify torch.torch.int32 to torch.int32 (#1237)
|
2 months ago |
|
flash_attn_interface.py
|
83e41b3ca4
Add custom ops for compatibility with PT Compile (#1139)
|
2 months ago |
|
flash_attn_triton.py
|
f1a73d0740
Run isort and black on python files
|
1 year ago |
|
flash_attn_triton_og.py
|
f1a73d0740
Run isort and black on python files
|
1 year ago |
|
flash_blocksparse_attention.py
|
cdbbe844b1
minor changes to unpad_input test util func
|
2 months ago |
|
flash_blocksparse_attn_interface.py
|
f1a73d0740
Run isort and black on python files
|
1 year ago |
|
fused_softmax.py
|
f1a73d0740
Run isort and black on python files
|
1 year ago |
|
pyproject.toml
|
73bd3f3bbb
Move pyproject.toml to flash-attn and tests dir to avoid PEP 517
|
1 year ago |
Zhihao Shen
30e1ef0f79
minify torch.torch.int32 to torch.int32 (#1237)