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:
@@ -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):
|
||||
|
||||
Reference in New Issue
Block a user