Tri Dao
|
20b84d6363
Don't use IntraWGOverlap for hdim 64,512
|
3 päivää sitten |
Tri Dao
|
6752d62aa4
Add dynamic splits
|
1 viikko sitten |
Tri Dao
|
74dfa43c8d
Fix divide by 0 in causal tile_scheduler for large seqlen
|
2 viikkoa sitten |
Tri Dao
|
7bc3f031a4
Compile for both Sm80 and Sm90
|
1 kuukausi sitten |
Tri Dao
|
7a802796e1
Big refactor and update
|
1 kuukausi sitten |
Son Nguyen
|
478ee666cc
Make namespace comment consistent (#1305)
|
4 kuukautta sitten |
jayhshah
|
a5a75274bc
FA3 kvcache + split kv + gqa parallelization (#1236)
|
4 kuukautta sitten |
Ying Zhang
|
dff976a84a
fixes
|
6 kuukautta sitten |
jayhshah
|
5018ac6ac5
Fp8 kernel with "in-kernel" transpose of V in producer (#1100)
|
7 kuukautta sitten |
Tri Dao
|
74b0761ff7
[FA3] BF16 forward
|
7 kuukautta sitten |
Tri Dao
|
7f67966cc7
FA3 initial code release
|
7 kuukautta sitten |