From 278b191dc16991f86064e5b2e150e1edb0738e8e Mon Sep 17 00:00:00 2001
From: Guancheng Fu <110874468+gc-fu@users.noreply.github.com>
Date: Thu, 22 Aug 2024 17:45:26 +0800
Subject: [PATCH] Fix optimize lm head error (#11899)

---
 python/llm/src/ipex_llm/transformers/convert.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/python/llm/src/ipex_llm/transformers/convert.py b/python/llm/src/ipex_llm/transformers/convert.py
index 4f44b058..a36d668e 100644
--- 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(