 Driss Guessous
|
bc482cbf91
Add a macro for namespace (#1419)
|
vor 3 Wochen |
 Lu Fang
|
74aed78373
Replace c10::optional with std::optional in flash_attn
|
vor 3 Wochen |
 Kirthi Shankar Sivamani
|
bbb21d6533
Generalize arch checks for A100 and above (#1427)
|
vor 4 Wochen |
 Tri Dao
|
9375ac9322
[CI] Don't include <ATen/cuda/CUDAGraphsUtils.cuh>
|
vor 2 Monaten |
|
Tri Dao
|
e782d28692
[CI] Change torch #include to make it work with torch 2.1 Philox
|
vor 2 Monaten |
|
Tri Dao
|
073afd5931
[CI] Use torch 2.6.0.dev20241001, reduce torch #include
|
vor 2 Monaten |
 sclarkson
|
1feb711f46
Fix compilation with clang on ARM64 (#1285)
|
vor 2 Monaten |
 Antoni Viros
|
83e41b3ca4
Add custom ops for compatibility with PT Compile (#1139)
|
vor 4 Monaten |
|
Tri Dao
|
65f723bb9a
Split bwd into more .cu files to speed up compilation
|
vor 6 Monaten |
|
Tri Dao
|
751c762c9c
Don't specialize for hdim 224 to speed up compilation
|
vor 6 Monaten |
|
Driss Guessous
|
1c275eb070
Fix ima for split-kv kernel (#1085)
|
vor 6 Monaten |
 Jorge António
|
4df62e1440
catch typo (#1058)
|
vor 6 Monaten |
|
Tri Dao
|
40e534a7f6
Implement cache_leftpad
|
vor 7 Monaten |
|
Tri Dao
|
dca6d89da4
Don't support softcap and dropout at the same time
|
vor 7 Monaten |
|
Tri Dao
|
908511b2b6
Split into more .cu files to speed up compilation
|
vor 7 Monaten |
|
Tri Dao
|
1d536d7de5
Minor cleanup of softcapping
|
vor 7 Monaten |
 Nicolas Patry
|
8f873cc6ac
Implement softcapping. (#1025)
|
vor 7 Monaten |
|
Nicolas Patry
|
5bf201966a
Fixing argument checking when using `seqlenq_ngroups_swapped`. (#976)
|
vor 7 Monaten |
 Grigory Sizov
|
f816dee63c
Support unpadded LSE layout (#970)
|
vor 7 Monaten |
|
Tri Dao
|
9eb3d099c1
Transpose out when swapping seqlen_q and num_groups
|
vor 10 Monaten |
|
Driss Guessous
|
4a73e903da
Add in, macrosf for defining __grid_constant__ (#852)
|
vor 10 Monaten |
|
Grigory Sizov
|
2a15840f09
Enable paged attention in varlen forward (#831)
|
vor 10 Monaten |
|
Tri Dao
|
2406f28805
Enable headdim 256 backward on consumer GPUs (Ampere, Ada)
|
vor 11 Monaten |
|
Tri Dao
|
d9a5cb291c
Fix dv = torch::empty_like(k) for mha_bwd_varlen as well
|
vor 1 Jahr |
 Brian Hirsh
|
2423cca3ad
fix backward for when query and key have different contiguity (#818)
|
vor 1 Jahr |
|
Grigory Sizov
|
4687936413
Fix Windows build (#816)
|
vor 1 Jahr |
 Jeremy Reizenstein
|
0658e320f6
Preprocessor switches to control functionality (#788)
|
vor 1 Jahr |
|
Tri Dao
|
54e80a3829
Implement page KV cache
|
vor 1 Jahr |
|
Tri Dao
|
ea8a25ca38
Remove configure in bwd kernel launch
|
vor 1 Jahr |
 Grigory Sizov
|
af01244ddd
Add split-kv and M<->H swap to varlen forward decoding attention (#754)
|
vor 1 Jahr |
