further fix

intel-analytics · Dec 13, 2024 · d2199db · d2199db
1 parent 509bdb4
commit d2199db
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/python/llm/src/ipex_llm/transformers/npu_models/mp_models_base.py b/python/llm/src/ipex_llm/transformers/npu_models/mp_models_base.py
@@ -213,6 +213,7 @@ def attention(self,
                 value_states = new_value_states
         else:
             value_states = self.transpose(value_states, [0, 2, 1, 3])
+            new_value_states = value_states
 
         query_states, key_states = self.apply_rotary_pos_emb(
             q=query_states,
@@ -225,7 +226,6 @@ def attention(self,
             head_dim=head_dim,
         )
         new_key_states = key_states
-        new_value_states = value_states
 
         if mode == "decode":
             key_states = self.concat(past_key, key_states, axis=-2)