[NPU] Fix abnormal output with latest driver (#12530)

intel-analytics · Dec 12, 2024 · f36c236 · f36c236
1 parent ffce86d
commit f36c236
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
@@ -471,7 +471,7 @@ def layer_norm(self, hidden_states, layernorm_weight):
         )
         eps = self.constant(self.rms_norm_eps)
         hidden_states = self.eltwise_div(hidden_states, self.sqrt(self.eltwise_add(variance, eps)))
-        layernorm_weight = self.convert_to_fp32(layernorm_weight)
+        hidden_states = self.convert_to_fp16(hidden_states)
         hidden_states = self.eltwise_mul(layernorm_weight, hidden_states)
         hidden_states = self.convert_to_fp16(hidden_states)
         return hidden_states