 AlpinDale
|
22305c91e9
refactor _prepare_model_input_tensor and attn metadata builder for most backends
|
il y a 5 mois |
|
AlpinDale
|
9d7beaa5b9
chore: separate kv_scale into k_scale and v_scale
|
il y a 5 mois |
|
AlpinDale
|
2105e4fd6b
feat: correctly invoke prefill & decode kernels for cross-attention
|
il y a 5 mois |
 AlpinDale
|
405bb74612
Control plane comms refactor (#573)
|
il y a 6 mois |
|
AlpinDale
|
696f2cd59c
add phi3_small support with blocksparse attention
|
il y a 6 mois |
|
AlpinDale
|
a94de94c44
refactor: combine the prefill and decode into a single API (#553)
|
il y a 6 mois |
|
AlpinDale
|
50b7c13db0
refactor: attention selector (#552)
|
il y a 6 mois |
|
AlpinDale
|
8b56dc4347
dict -> torch.Tensor for blocks_to_swap
|
il y a 6 mois |
|
AlpinDale
|
3a0d1c7705
add get_name method to attention backends
|
il y a 6 mois |
|
AlpinDale
|
21ce19b3ea
blocks_to_copy dict -> torch.Tensor
|
il y a 6 mois |
|
AlpinDale
|
2351a0e2cd
feat: FlashInfer backend for decoding phase (#548)
|
il y a 6 mois |
|
AlpinDale
|
fca911ee0a
vLLM Upstream Sync (#526)
|
il y a 7 mois |
|
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
il y a 9 mois |
