 Tri Dao
|
0c49ac9a07
Implement rotary non-interleaved
|
hai 3 semanas |
|
Tri Dao
|
9f82a326ad
Implement rotary for attn decode
|
hai 4 semanas |
|
Tri Dao
|
4d00645c76
Implement appending new KV to KV cache
|
hai 1 mes |
|
Tri Dao
|
df96486c31
Decode: varlen, paged KV, leftpad
|
hai 1 mes |
|
Tri Dao
|
6e8b25e426
Refactor
|
hai 2 meses |
 Ying Zhang
|
dff976a84a
fixes
|
hai 3 meses |
|
Ying Zhang
|
7b4e68e04f
hopper local attention
|
hai 3 meses |
|
Ying Zhang
|
db80387343
Add seqused_q in fwd / bwd and seqused_k in bwd.
|
hai 3 meses |
 Charlene Yang
|
bdf733be55
Add q, k, v descales to FA3 interface (#1210)
|
hai 3 meses |
 jayhshah
|
c92ca63268
FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173)
|
hai 3 meses |
|
Tri Dao
|
bafe253042
[FA3] Bwd
|
hai 4 meses |
|
Tri Dao
|
3aae9c18c1
Revert "Changes For FP8 (#1075)"
|
hai 4 meses |
 ganeshcolfax
|
1899c970c8
Changes For FP8 (#1075)
|
hai 4 meses |
 Ying Zhang
|
dfe1a59e4b
Add var-seq-len to FA3 fp16 / bf16 fwd (#1072)
|
hai 4 meses |
 youkaichao
|
ef3e358a25
remove lambda (#1056)
|
hai 4 meses |
|
Tri Dao
|
7f67966cc7
FA3 initial code release
|
hai 5 meses |
