fix: 修复KV缓存槽位索引错位、版本校验缺失与注意力掩码问题,合并预填充方法

This commit is contained in:
2026-05-06 19:51:14 +08:00
parent 520de3ebe8
commit 123f25e339
2 changed files with 86 additions and 114 deletions
-2
View File
@@ -186,8 +186,6 @@ class InferenceEngine:
dtype=dtype,
)
self.kv_cache = self.scheduler.kv_cache
self.seq_mask = self.scheduler.seq_mask
self.scheduler.start()
def __enter__(self):