diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/generate.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/generate.py
index 4a8314d09d2..ebef3dae4c2 100644
--- a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/generate.py
+++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/generate.py
@@ -41,6 +41,8 @@
 
     # Load model in 4 bit,
     # which convert the relevant layers in the model into INT4 format
+    # When running LLMs on Intel iGPUs for Windows users, we recommend setting `cpu_embedding=True` in the from_pretrained function.
+    # This will allow the memory-intensive embedding layer to utilize the CPU instead of iGPU.
     model = AutoModel.from_pretrained(model_path,
                                       load_in_4bit=True,
                                       optimize_model=True,
@@ -62,10 +64,7 @@
                                 max_new_tokens=args.n_predict)
 
         st = time.time()
-        # if your selected model is capable of utilizing previous key/value attentions
-        # to enhance decoding speed, but has `"use_cache": false` in its model config,
-        # it is important to set `use_cache=True` explicitly in the `generate` function
-        # to obtain optimal performance with IPEX-LLM INT4 optimizations
+
         output = model.generate(input_ids,
                                 max_new_tokens=args.n_predict)
         
diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/streamchat.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/streamchat.py
index 90c9a7da975..31be35d9639 100644
--- a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/streamchat.py
+++ b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/glm4/streamchat.py
@@ -39,11 +39,14 @@
 
     # Load model in 4 bit,
     # which convert the relevant layers in the model into INT4 format
+    # When running LLMs on Intel iGPUs for Windows users, we recommend setting `cpu_embedding=True` in the from_pretrained function.
+    # This will allow the memory-intensive embedding layer to utilize the CPU instead of iGPU.
     model = AutoModel.from_pretrained(model_path,
                                       trust_remote_code=True,
                                       load_in_4bit=True,
                                       optimize_model=True,
-                                      use_cache=True)
+                                      use_cache=True,
+                                      cpu_embedding=True)
     
     model = model.to('xpu')
 
@@ -64,4 +67,3 @@
             for response, history in model.stream_chat(tokenizer, args.question, history=[]):
                 print(response.replace(response_, ""), end="")
                 response_ = response
-                
\ No newline at end of file
diff --git a/python/llm/example/GPU/PyTorch-Models/Model/glm4/generate.py b/python/llm/example/GPU/PyTorch-Models/Model/glm4/generate.py
index 246e52ef635..3433ba29167 100644
--- a/python/llm/example/GPU/PyTorch-Models/Model/glm4/generate.py
+++ b/python/llm/example/GPU/PyTorch-Models/Model/glm4/generate.py
@@ -39,6 +39,8 @@
     model_path = args.repo_id_or_model_path
 
     # Load model
+    # When running LLMs on Intel iGPUs for Windows users, we recommend setting `cpu_embedding=True` in the from_pretrained function.
+    # This will allow the memory-intensive embedding layer to utilize the CPU instead of iGPU.
     model = AutoModel.from_pretrained(model_path,
                                       trust_remote_code=True,
                                       torch_dtype = 'auto',