Jay Shah
|
50cb90aea6
comment out unimplemented kwargs from flash_attn_with_kvcache
|
4 months ago |
Jay Shah
|
b5cac6d586
rebase with Is_local disabled temporarily
|
5 months ago |
Jay Shah
|
2472e5e0b4
add 'in principle' fp8 kv cache support
|
5 months ago |
Jay Shah
|
e36e004cb3
change fp8 code path to allow for split kernel and kv cache without perf regression
|
5 months ago |
Jay Shah
|
f53703b98d
add split kv heuristic modifications
|
5 months ago |
Ganesh Bikshandi
|
099ca28488
add gqa decoding logic.
|
5 months ago |
Jay Shah
|
a06f1f970b
add to kv cache api
|
6 months ago |
Jay Shah
|
535b8279bb
complete gqa parallel changes for non-causal
|
6 months ago |
Ganesh Bikshandi
|
84e31c22d6
adding python interface.
|
6 months ago |
Ying Zhang
|
dff976a84a
fixes
|
6 months ago |
Ying Zhang
|
7b4e68e04f
hopper local attention
|
6 months ago |
Ying Zhang
|
db80387343
Add seqused_q in fwd / bwd and seqused_k in bwd.
|
6 months ago |
Charlene Yang
|
bdf733be55
Add q, k, v descales to FA3 interface (#1210)
|
5 months ago |
jayhshah
|
c92ca63268
FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173)
|
6 months ago |
Tri Dao
|
bafe253042
[FA3] Bwd
|
7 months ago |
Tri Dao
|
3aae9c18c1
Revert "Changes For FP8 (#1075)"
|
7 months ago |
ganeshcolfax
|
1899c970c8
Changes For FP8 (#1075)
|
7 months ago |
Ying Zhang
|
dfe1a59e4b
Add var-seq-len to FA3 fp16 / bf16 fwd (#1072)
|
7 months ago |
youkaichao
|
ef3e358a25
remove lambda (#1056)
|
7 months ago |
Tri Dao
|
7f67966cc7
FA3 initial code release
|
7 months ago |