 AlpinDale
|
696f2cd59c
add phi3_small support with blocksparse attention
|
vor 6 Monaten |
|
AlpinDale
|
0c15965621
fix fp8 kv
|
vor 6 Monaten |
 AlpinDale
|
a94de94c44
refactor: combine the prefill and decode into a single API (#553)
|
vor 6 Monaten |
|
AlpinDale
|
50b7c13db0
refactor: attention selector (#552)
|
vor 6 Monaten |
|
AlpinDale
|
d11d68f4e6
switch to vllm-flash-attn
|
vor 6 Monaten |
|
AlpinDale
|
8b56dc4347
dict -> torch.Tensor for blocks_to_swap
|
vor 6 Monaten |
|
AlpinDale
|
3a0d1c7705
add get_name method to attention backends
|
vor 6 Monaten |
|
AlpinDale
|
21ce19b3ea
blocks_to_copy dict -> torch.Tensor
|
vor 6 Monaten |
|
AlpinDale
|
2351a0e2cd
feat: FlashInfer backend for decoding phase (#548)
|
vor 6 Monaten |
