Add custom op declaration for all_reduce (#3473)

* add custom op declaration * roll back try except
2025-10-05 08:37:06 +08:00 · 2025-08-20 20:29:58 +08:00
parent 33ff0bfe38
commit bcdfc1d6b9
4 changed files with 16 additions and 4 deletions
--- a/fastdeploy/model_executor/layers/attention/mla_attention_backend.py
+++ b/fastdeploy/model_executor/layers/attention/mla_attention_backend.py
@@ -89,6 +89,9 @@ class MLAAttentionMetadata(AttentionMetadata):
    kv_signal_metadata: Optional[paddle.Tensor] = None
    kv_signal_data_list: List[Optional[paddle.Tensor]] = field(default_factory=list)

+    max_enc_len_this_time: Optional[paddle.Tensor] = None
+    max_dec_len_this_time: Optional[paddle.Tensor] = None
+

 class MLAAttentionBackend(AttentionBackend):
    """