hotfix attn alibi wo head mapping (#496)

Co-authored-by: oliveryuan <oliveryuan@basemind.com>
2023-07-19 02:31:48 +08:00
parent 453bafb96f
commit bda41c70dd
2 changed files with 3 additions and 0 deletions
--- a/vllm/model_executor/layers/attention.py
+++ b/vllm/model_executor/layers/attention.py
@@ -408,6 +408,7 @@ class PagedAttentionWithALiBi(PagedAttention):
            query,
            key_cache,
            value_cache,
+            self.head_mapping,
            self.scale,
            input_metadata.block_tables,
            input_metadata.context_lens,