From c9e6d42ad10598193e14d175cb5d1b01bd33232e Mon Sep 17 00:00:00 2001
From: "Wang, Jian4" <61138589+hzjane@users.noreply.github.com>
Date: Wed, 10 Apr 2024 11:24:06 +0800
Subject: [PATCH] LLM: Fix chatglm3-6b-32k error (#10719)

* fix chatglm3-6b-32k

* update style
---
 python/llm/src/ipex_llm/transformers/convert.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/python/llm/src/ipex_llm/transformers/convert.py b/python/llm/src/ipex_llm/transformers/convert.py
index 6d81c677..ff133b35 100644
--- a/python/llm/src/ipex_llm/transformers/convert.py
+++ b/python/llm/src/ipex_llm/transformers/convert.py
@@ -889,7 +889,8 @@ def _optimize_post(model, lightweight_bmm=False):
 
     if model.config.architectures is not None \
        and model.config.architectures[0] in ["ChatGLMModel", "ChatGLMForConditionalGeneration"]:
-        if model.config.num_layers == 28 and hasattr(model.config, 'rope_ratio'):
+        if (model.config.num_layers == 28 and hasattr(model.config, 'rope_ratio')
+                and model.config.rope_ratio == 16):
             # chatglm2-6b-32k
             modeling_module_name = model.__class__.__module__
             module = importlib.import_module(modeling_module_name)