Tri Dao
|
45c48afb2b
Add option for WG1 to use RS MMA but WG2 using SS MMA
|
vor 1 Tag |
Tri Dao
|
5458c78e6d
Remove sink token
|
vor 3 Tagen |
Tri Dao
|
6752d62aa4
Add dynamic splits
|
vor 1 Woche |
Ted Zadouri
|
06e34f62d1
Enable MLA flag in FA3 (rope=64, latent=512) (#1504)
|
vor 1 Woche |
Tri Dao
|
b36ad4ef76
Use split for super long sequences that don't fit into L2
|
vor 2 Wochen |
Tri Dao
|
5378bc3204
Tile fwd_combine kernel along headdim, don't need kBlockM > 128
|
vor 3 Wochen |
Tri Dao
|
5fc5ebf82b
Fix num_splits heuristic again when PackGQA
|
vor 3 Wochen |
Tri Dao
|
5fab938555
Fix num_splits heuristic being called before get_pack_gqa
|
vor 3 Wochen |
Tri Dao
|
4e8496a781
Support Qv
|
vor 3 Wochen |
Tri Dao
|
ed53b5fc4c
Fix compilation for FP8 when kHeadDimV != kHeadDim
|
vor 3 Wochen |
Tri Dao
|
2a204125ae
Implement HeadDim_V != HeadDim_QK, support hdimQK=192, hdimV=128
|
vor 1 Monat |
Lu Fang
|
74aed78373
Replace c10::optional with std::optional in flash_attn
|
vor 1 Monat |
Tri Dao
|
2ac6c986be
Fix Sm80 tile_count_semaphore, adjust test tolerance
|
vor 1 Monat |
Tri Dao
|
e94f7e89dc
Always enable PackGQA is Split to reduce compilation and binary size
|
vor 1 Monat |
Tri Dao
|
40fa35acd8
Always enable PackGQA if PagedKV to reduce compilation and bin size
|
vor 1 Monat |
Tri Dao
|
a84a237d2a
Split bwd softcap compilation units for Sm80
|
vor 1 Monat |
Tri Dao
|
1e3208566a
Tune tile sizes for compilation
|
vor 1 Monat |
Tri Dao
|
84f1287e42
Rename bool_constant<true> to true_type, same w bool_constant<false>
|
vor 1 Monat |
Tri Dao
|
df5fe55264
Change tile sizes for Sm8x to reduce stack frame
|
vor 1 Monat |
Tri Dao
|
8dd0b479d5
Always enable PackGQA for Sm8x to reduce compilation and binary size
|
vor 1 Monat |
Tri Dao
|
7bc3f031a4
Compile for both Sm80 and Sm90
|
vor 1 Monat |
Tri Dao
|
7a802796e1
Big refactor and update
|
vor 1 Monat |
sclarkson
|
1feb711f46
Fix compilation with clang on ARM64 (#1285)
|
vor 3 Monaten |
Kai Londenberg
|
b443207c1f
Paged Attention support for FA3 (#1268)
|
vor 3 Monaten |
jayhshah
|
a5a75274bc
FA3 kvcache + split kv + gqa parallelization (#1236)
|
vor 4 Monaten |
Ying Zhang
|
dff976a84a
fixes
|
vor 6 Monaten |
Ying Zhang
|
7b4e68e04f
hopper local attention
|
vor 6 Monaten |
Ying Zhang
|
db80387343
Add seqused_q in fwd / bwd and seqused_k in bwd.
|
vor 6 Monaten |
jayhshah
|
c92ca63268
FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173)
|
vor 6 Monaten |
Tri Dao
|
5d5bfbb619
Remove contiguous checks
|
vor 7 Monaten |