[V1][Spec Decode] Ngram Spec Decode (#12193)

Signed-off-by: LiuXiaoxuanPKU <lilyliupku@gmail.com>
This commit is contained in:
Lily Liu
2025-02-15 18:05:11 -08:00
committed by GitHub
parent 367cb8ce8c
commit 80f63a3966
21 changed files with 1023 additions and 82 deletions

View File

@@ -12,6 +12,8 @@ class SamplingMetadata:
temperature: torch.Tensor
all_greedy: bool
all_random: bool
rejection_sampling: bool
spec_token_ids: List[List[int]]
top_p: torch.Tensor
top_k: torch.Tensor