Fix optimize lm head error (#11899)

intel-analytics · Aug 22, 2024 · 278b191 · 278b191
1 parent c5b51d4
commit 278b191
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/python/llm/src/ipex_llm/transformers/convert.py b/python/llm/src/ipex_llm/transformers/convert.py
@@ -225,6 +225,8 @@ def convert_vllm(module, qtype, in_features, out_features, mp_group, cur_qtype,
     from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
     from ipex_llm.transformers.low_bit_linear import LowBitLinear, \
         FP16Linear, BF16Linear, vLLMLowBitLinear, vLLMFP16Linear, vLLMBF16Linear
+    # Currently, vLLM does not support optimize_lm_head = True
+    optimize_lm_head = False
     if isinstance(module, ParallelLMHead):
         if qtype == ggml_tensor_qtype["fp16"]:
             new_linear = FP16Linear(