From c75f3dd8748642a25b5b9cb55ade776c5cf4a618 Mon Sep 17 00:00:00 2001
From: "Wang, Jian4" <61138589+hzjane@users.noreply.github.com>
Date: Wed, 11 Sep 2024 13:44:40 +0800
Subject: [PATCH] vllm no padding glm4 to avoid nan error (#12062)

* no padding glm4

* add codegeex
---
 python/llm/src/ipex_llm/vllm/xpu/model_convert.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/python/llm/src/ipex_llm/vllm/xpu/model_convert.py b/python/llm/src/ipex_llm/vllm/xpu/model_convert.py
index 89a8fa67..d8ece358 100644
--- a/python/llm/src/ipex_llm/vllm/xpu/model_convert.py
+++ b/python/llm/src/ipex_llm/vllm/xpu/model_convert.py
@@ -89,7 +89,8 @@ def get_load_function(low_bit):
             )
             if "qwen" in self.model_config.model.lower() or \
                     "baichuan" in self.model_config.model.lower() or \
-                    "glm" in self.model_config.model.lower():
+                    "codegeex4-all" in self.model_config.model.lower() or \
+                    "chatglm" in self.model_config.model.lower():
                 self.model.apply(padding_mlp)
             from ipex_llm import optimize_model
             import os