[MTP][V1] Adapt mtp with graph mode in v1. (#1023)

Adapts deepseek mtp with torch air graph mode in v1. --------- Signed-off-by: whx-sjtu <2952154980@qq.com>
2025-06-09 22:21:42 +08:00
parent 5ac4872f5e
commit cd2f14a1b3
4 changed files with 87 additions and 24 deletions
--- a/vllm_ascend/attention/attention_v1.py
+++ b/vllm_ascend/attention/attention_v1.py
@@ -100,6 +100,7 @@ class AscendAttentionState(Enum):
    PrefillCacheHit = 1
    DecodeOnly = 2
    ChunkedPrefill = 3
+    SpecDecoding = 4


@dataclass