 Tri Dao
|
65f723bb9a
Split bwd into more .cu files to speed up compilation
|
há 4 meses atrás |
|
Tri Dao
|
751c762c9c
Don't specialize for hdim 224 to speed up compilation
|
há 4 meses atrás |
 rocking
|
d8f104e97a
Support AMD ROCm on FlashAttention 2 (#1010)
|
há 4 meses atrás |
|
Tri Dao
|
844912dca0
[CI] Switch from CUDA 12.2 to 12.3
|
há 5 meses atrás |
|
Tri Dao
|
908511b2b6
Split into more .cu files to speed up compilation
|
há 5 meses atrás |
|
Tri Dao
|
beb2bf2a32
Drop support for pytorch 1.12, 1.13, and python 3.7
|
há 5 meses atrás |
 Nicolas Patry
|
8f873cc6ac
Implement softcapping. (#1025)
|
há 5 meses atrás |
 Corey James Levinson
|
beb8b8ba9f
add exception to Timeout Error (#963)
|
há 6 meses atrás |
 Wei Ji
|
9c0e9ee86d
Move packaging and ninja from install_requires to setup_requires (#937)
|
há 7 meses atrás |
|
Tri Dao
|
2aea958f89
[CI] Compile with torch 2.3.0.dev20240207
|
há 8 meses atrás |
 Arvind Sundararajan
|
26c9e82743
Support ARM builds (#757)
|
há 9 meses atrás |
 Chirag Jain
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
há 9 meses atrás |
 Qubitium
|
f45bbb4c94
Optimize compile to 1: avoid oom 2: minimize swap usage 3: avoid threads starvation when letting ninja decide how many workers to spawn or manual MAX_JOBS "guesses". Logic is to take the min value of MAX_JOBS auto-calculated by two metrics: 1: cpu cores 2: free memory. This should allow flash-attn to compile close to the most efficient manner under any consumer/server env. (#832)
|
há 10 meses atrás |
|
Tri Dao
|
d4a7c8ffbb
[CI] Only compile for CUDA 11.8 & 12.2, MAX_JOBS=2,add torch-nightly
|
há 1 ano atrás |
|
Tri Dao
|
5e525a8dc8
[CI] Use official Pytorch 2.1, add CUDA 11.8 for Pytorch 2.1
|
há 1 ano atrás |
|
Tri Dao
|
1879e089c7
Reduce number of templates for headdim > 128
|
há 1 ano atrás |
|
Tri Dao
|
bff3147175
Re-enable compilation for Hopper
|
há 1 ano atrás |
|
Tri Dao
|
dfe29f5e2b
[Gen] Don't use ft_attention, use flash_attn_with_kvcache instead
|
há 1 ano atrás |
 Federico Berto
|
fa3ddcbaaa
[Minor] add nvcc note on bare_metal_version `RuntimeError` (#552)
|
há 1 ano atrás |
|
Tri Dao
|
799f56fa90
Don't compile for Pytorch 2.1 on CUDA 12.1 due to nvcc segfaults
|
há 1 ano atrás |
|
Tri Dao
|
bb9beb3645
Remove some unused headers
|
há 1 ano atrás |
|
Tri Dao
|
0c04943fa2
Require CUDA 11.6+, clean up setup.py
|
há 1 ano atrás |
|
Tri Dao
|
b1fbbd8337
Implement splitKV attention
|
há 1 ano atrás |
|
Tri Dao
|
cbb4cf5f46
Don't need to set TORCH_CUDA_ARCH_LIST in setup.py
|
há 1 ano atrás |
 Aman Gupta Karmani
|
aab603af4f
fix binary wheel installation when nvcc is not available (#448)
|
há 1 ano atrás |
|
Tri Dao
|
9c531bdc0a
Use single thread compilation for cuda12.1, torch2.1 to avoid OOM CI
|
há 1 ano atrás |
|
Tri Dao
|
2ddeaa406c
Fix wheel building
|
há 1 ano atrás |
|
Tri Dao
|
3c458cff77
Merge branch 'feature/demo-wheels' of https://github.com/piercefreeman/flash-attention into piercefreeman-feature/demo-wheels
|
há 1 ano atrás |
|
Tri Dao
|
1c41d2b0e5
Fix race condition in bwd (overwriting sK)
|
há 1 ano atrás |
|
Tri Dao
|
4f285b3547
FlashAttention-2 release
|
há 1 ano atrás |
