Add --modelscope in GPU examples for glm4, codegeex2, qwen2 and qwen2.5 (#12561)

* Add --modelscope for more models

* imporve readme

---------

Co-authored-by: ATMxsp01 <shou.xu@intel.com>

python/llm/example/GPU/HuggingFace/LLM/chatglm3/README.md

@@ -108,7 +108,7 @@ python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROM
 ```
 
 Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `ZhipuAI/chatglm3-6b` for **ModelScope**.
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `'ZhipuAI/chatglm3-6b'` for **ModelScope**.
 - `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么？'`.
 - `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
 - `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
@@ -162,7 +162,7 @@ python ./streamchat.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --question
 ```
 
 Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `ZhipuAI/chatglm3-6b` for **ModelScope**.
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the ChatGLM3 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/chatglm3-6b'` for **Hugging Face** or `'ZhipuAI/chatglm3-6b'` for **ModelScope**.
 - `--question QUESTION`: argument defining the question to ask. It is default to be `"晚上睡不着应该怎么办"`.
 - `--disable-stream`: argument defining whether to stream chat. If include `--disable-stream` when running the script, the stream chat is disabled and `chat()` API is used.
 - `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.

python/llm/example/GPU/HuggingFace/LLM/codegeex2/README.md

@@ -1,6 +1,6 @@
 # CodeGeeX2
 
-In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on CodeGeeX2 models which is implemented based on the ChatGLM2 architecture trained on more code data on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex2-6b) as a reference CodeGeeX2 model.
+In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on CodeGeeX2 models which is implemented based on the ChatGLM2 architecture trained on more code data on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b) (or [ZhipuAI/codegeex2-6b](https://www.modelscope.cn/models/ZhipuAI/codegeex2-6b) for ModelScope) as a reference CodeGeeX2 model.
 
 ## 0. Requirements
 To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#requirements) for more information.
@@ -16,6 +16,9 @@ conda create -n llm python=3.11
 conda activate llm
 # below command will install intel_extension_for_pytorch==2.1.10+xpu as default
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 #### 1.2 Installation on Windows
@@ -26,10 +29,13 @@ conda activate llm
 
 # below command will install intel_extension_for_pytorch==2.1.10+xpu as default
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 ### 2. Download Model and Replace File
-If you select the codegeex2-6b model ([THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex2-6b)), please note that their code (`tokenization_chatglm.py`) initialized tokenizer after the call of `__init__` of its parent class, which may result in error during loading tokenizer. To address issue, we have provided an updated file ([tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py))
+If you select the codegeex2-6b model ([THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b) (for **Hugging Face**) or [ZhipuAI/codegeex2-6b](https://www.modelscope.cn/models/ZhipuAI/codegeex2-6b) (for **ModelScope**)), please note that their code (`tokenization_chatglm.py`) initialized tokenizer after the call of `__init__` of its parent class, which may result in error during loading tokenizer. To address issue, we have provided an updated file ([tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py))
 
 ```python
 def __init__(self, vocab_file, padding_side="left", clean_up_tokenization_spaces=False, **kwargs):
@@ -37,7 +43,7 @@ def __init__(self, vocab_file, padding_side="left", clean_up_tokenization_spaces
     super().__init__(padding_side=padding_side, clean_up_tokenization_spaces=clean_up_tokenization_spaces, **kwargs)
 ```
 
-You could download the model from [THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex2-6b), and replace the file  `tokenization_chatglm.py` with [tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py).
+You could download the model from [THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b) (for **Hugging Face**) or [ZhipuAI/codegeex2-6b](https://www.modelscope.cn/models/ZhipuAI/codegeex2-6b) (for **ModelScope**), and replace the file  `tokenization_chatglm.py` with [tokenization_chatglm.py](./codegeex2-6b/tokenization_chatglm.py).
 
 ### 3. Configures OneAPI environment variables for Linux
 
@@ -104,17 +110,22 @@ set SYCL_CACHE_PERSISTENT=1
 > For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile.
 
 ### 5. Running examples
-```
+```bash
+# for Hugging Face model hub
 python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+
+# for ModelScope model hub
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
 ```
 
 Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the CodeGeeX2 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/codegeex-6b'`.
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the CodeGeeX2 model to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/codegeex2-6b'` for **Hugging Face** or `'ZhipuAI/codegeex-6b'` for **ModelScope**.
 - `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'# language: Python\n# write a bubble sort function\n'`.
 - `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `128`.
+- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
 
 #### Sample Output
-#### [THUDM/codegeex-6b](https://huggingface.co/THUDM/codegeex-6b)
+#### [THUDM/codegeex2-6b](https://huggingface.co/THUDM/codegeex2-6b)
 ```log
 Inference time: xxxx s
 -------------------- Prompt --------------------

python/llm/example/GPU/HuggingFace/LLM/codegeex2/generate.py

@@ -28,18 +28,29 @@ CODEGEEX_PROMPT_FORMAT = "{prompt}"
 
 
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for ChatGLM2 model')
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for CodeGeeX2 model')
-    parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/codegeex2-6b",
+    parser.add_argument('--repo-id-or-model-path', type=str, 
-                        help='The huggingface repo id for the CodeGeeX2 model to be downloaded'
+                        help='The Hugging Face or ModelScope repo id for the CodeGeeX2 model to be downloaded'
-                             ', or the path to the huggingface checkpoint folder')
+                             ', or the path to the checkpoint folder')
     parser.add_argument('--prompt', type=str, default="# language: Python\n# write a bubble sort function\n",
                         help='Prompt to infer')
     parser.add_argument('--n-predict', type=int, default=128,
                         help='Max tokens to predict')
+    parser.add_argument('--modelscope', action="store_true", default=False, 
+                        help="Use models from modelscope")
 
     args = parser.parse_args()
-    model_path = args.repo_id_or_model_path
 
+    if args.modelscope:
+        from modelscope import AutoTokenizer
+        model_hub = 'modelscope'
+    else:
+        from transformers import AutoTokenizer
+        model_hub = 'huggingface'
+    
+    model_path = args.repo_id_or_model_path if args.repo_id_or_model_path else \
+        ("ZhipuAI/codegeex2-6b" if args.modelscope else "THUDM/codegeex2-6b")
+    
     # 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.
@@ -48,7 +59,8 @@ if __name__ == '__main__':
                                       load_in_4bit=True,
                                       optimize_model=True,
                                       trust_remote_code=True,
-                                      use_cache=True)
+                                      use_cache=True,
+                                      model_hub=model_hub)
     model = model.half().to('xpu')
 
     # Load tokenizer

python/llm/example/GPU/HuggingFace/LLM/glm4/README.md

@@ -1,5 +1,5 @@
 # GLM-4
-In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-4 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) as a reference InternLM model.
+In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on GLM-4 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize the [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat) (or [ZhipuAI/glm4-9b-chat](https://www.modelscope.cn/models/ZhipuAI/glm4-9b-chat) for ModelScope) as a reference InternLM model.
 
 ## 0. Requirements
 To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#requirements) for more information.
@@ -15,6 +15,9 @@ pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-exte
 
 # install packages required for GLM-4, it is recommended to use transformers>=4.44 for THUDM/glm-4-9b-chat updated after August 12, 2024
 pip install "tiktoken>=0.7.0" transformers==4.44 "trl<0.12.0"
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 ### 1.2 Installation on Windows
@@ -28,6 +31,9 @@ pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-exte
 
 # install packages required for GLM-4, it is recommended to use transformers>=4.44 for THUDM/glm-4-9b-chat updated after August 12, 2024
 pip install "tiktoken>=0.7.0" transformers==4.44 "trl<0.12.0"
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 ## 2. Configures OneAPI environment variables for Linux
@@ -98,14 +104,19 @@ set SYCL_CACHE_PERSISTENT=1
 ### Example 1: Predict Tokens using `generate()` API
 In the example [generate.py](./generate.py), we show a basic use case for a GLM-4 model to predict the next N tokens using `generate()` API, with IPEX-LLM INT4 optimizations on Intel GPUs.
 
-```
+```bash
+# for Hugging Face model hub
 python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+
+# for ModelScope model hub
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
 ```
 
 Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the GLM-4 model (e.g. `THUDM/glm-4-9b-chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'`.
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the GLM-4 model (e.g. `THUDM/glm-4-9b-chat`) to be downloaded, or the path to the checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'` for **Hugging Face** or `'ZhipuAI/glm-4-9b-chat'` for **ModelScope**.
 - `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么？'`.
 - `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
 
 #### Sample Output
 #### [THUDM/glm-4-9b-chat](https://huggingface.co/THUDM/glm-4-9b-chat)
@@ -134,21 +145,3 @@ What is AI?
 
 Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The term "art
 ```
-
-### Example 2: Stream Chat using `stream_chat()` API
-In the example [streamchat.py](./streamchat.py), we show a basic use case for a GLM-4 model to stream chat, with IPEX-LLM INT4 optimizations.
-
-**Stream Chat using `stream_chat()` API**:
-```
-python ./streamchat.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --question QUESTION
-```
-
-**Chat using `chat()` API**:
-```
-python ./streamchat.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --question QUESTION --disable-stream
-```
-
-Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the GLM-4 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'THUDM/glm-4-9b-chat'`.
-- `--question QUESTION`: argument defining the question to ask. It is default to be `"AI是什么？"`.
-- `--disable-stream`: argument defining whether to stream chat. If include `--disable-stream` when running the script, the stream chat is disabled and `chat()` API is used.

python/llm/example/GPU/HuggingFace/LLM/glm4/generate.py

@@ -20,7 +20,6 @@ import argparse
 import numpy as np
 
 from ipex_llm.transformers import AutoModel
-from transformers import AutoTokenizer
 
 # you could tune the prompt based on your own model,
 # here the prompt tuning refers to https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/tokenization_chatglm.py
@@ -28,16 +27,27 @@ GLM4_PROMPT_FORMAT = "<|user|>\n{prompt}\n<|assistant|>"
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for GLM-4 model')
-    parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/glm-4-9b-chat",
+    parser.add_argument('--repo-id-or-model-path', type=str, 
-                        help='The huggingface repo id for the GLM-4 model to be downloaded'
+                        help='The Hugging Face or ModelScope repo id for GLM-4 model model to be downloaded'
-                             ', or the path to the huggingface checkpoint folder')
+                             ', or the path to the checkpoint folder')
     parser.add_argument('--prompt', type=str, default="AI是什么？",
                         help='Prompt to infer')
     parser.add_argument('--n-predict', type=int, default=32,
                         help='Max tokens to predict')
+    parser.add_argument('--modelscope', action="store_true", default=False, 
+                        help="Use models from modelscope")
 
     args = parser.parse_args()
-    model_path = args.repo_id_or_model_path
+
+    if args.modelscope:
+        from modelscope import AutoTokenizer
+        model_hub = 'modelscope'
+    else:
+        from transformers import AutoTokenizer
+        model_hub = 'huggingface'
+    
+    model_path = args.repo_id_or_model_path if args.repo_id_or_model_path else \
+        ("ZhipuAI/glm-4-9b-chat" if args.modelscope else "THUDM/glm-4-9b-chat")
 
     # Load model in 4 bit,
     # which convert the relevant layers in the model into INT4 format
@@ -47,8 +57,9 @@ if __name__ == '__main__':
                                       load_in_4bit=True,
                                       optimize_model=True,
                                       trust_remote_code=True,
-                                      use_cache=True)
+                                      use_cache=True,
-    model = model.to("xpu")
+                                      model_hub=model_hub)
+    model = model.half().to("xpu")
 
     # Load tokenizer
     tokenizer = AutoTokenizer.from_pretrained(model_path,

python/llm/example/GPU/HuggingFace/LLM/glm4/streamchat.py

@@ -1,69 +0,0 @@
-#
-# Copyright 2016 The BigDL Authors.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-
-import torch
-import time
-import argparse
-import numpy as np
-
-from ipex_llm.transformers import AutoModel
-from transformers import AutoTokenizer
-
-
-if __name__ == '__main__':
-    parser = argparse.ArgumentParser(description='Stream Chat for GLM-4 model')
-    parser.add_argument('--repo-id-or-model-path', type=str, default="THUDM/glm-4-9b-chat",
-                        help='The huggingface repo id for the GLM-4 model to be downloaded'
-                             ', or the path to the huggingface checkpoint folder')
-    parser.add_argument('--question', type=str, default="晚上睡不着应该怎么办",
-                        help='Qustion you want to ask')
-    parser.add_argument('--disable-stream', action="store_true",
-                        help='Disable stream chat')
-
-    args = parser.parse_args()
-    model_path = args.repo_id_or_model_path
-    disable_stream = args.disable_stream
-
-    # 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,
-                                      cpu_embedding=True)
-    
-    model = model.to('xpu')
-
-    # Load tokenizer
-    tokenizer = AutoTokenizer.from_pretrained(model_path,
-                                              trust_remote_code=True)
-
-    with torch.inference_mode():
-        if disable_stream:
-            # Chat
-            response, history = model.chat(tokenizer, args.question, history=[])
-            print('-'*20, 'Chat Output', '-'*20)
-            print(response)
-        else:
-            # Stream chat
-            response_ = ""
-            print('-'*20, 'Stream Chat Output', '-'*20)
-            for response, history in model.stream_chat(tokenizer, args.question, history=[]):
-                print(response.replace(response_, ""), end="")
-                response_ = response

python/llm/example/GPU/HuggingFace/LLM/qwen2.5/README.md

@@ -1,5 +1,5 @@
 # Qwen2.5
-In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2.5 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct), [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) and [Qwen/Qwen2.5-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct) as reference Qwen2.5 models.
+In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2.5 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct), [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) and [Qwen/Qwen2.5-14B-Instruct](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct) (or [Qwen/Qwen2.5-3B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-3B-Instruct), [Qwen/Qwen2.5-7B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-7B-Instruct) and [Qwen/Qwen2.5-14B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-14B-Instruct) for ModelScope) as reference Qwen2.5 models.
 
 ## 0. Requirements
 To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#requirements) for more information.
@@ -14,6 +14,9 @@ conda create -n llm python=3.11
 conda activate llm
 # below command will install intel_extension_for_pytorch==2.1.10+xpu as default
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 #### 1.2 Installation on Windows
@@ -24,6 +27,9 @@ conda activate llm
 
 # below command will install intel_extension_for_pytorch==2.1.10+xpu as default
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 ### 2. Configures OneAPI environment variables for Linux
@@ -91,14 +97,19 @@ set SYCL_CACHE_PERSISTENT=1
 > For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile.
 ### 4. Running examples
 
-```
+```bash
+# for Hugging Face model hub
 python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+
+# for ModelScope model hub
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
 ```
 
 Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen2.5 model (e.g. `Qwen/Qwen2.5-7B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2.5-7B-Instruct'`.
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the **Hugging Face** or **ModelScope** repo id for the Qwen2.5 model (e.g. `Qwen/Qwen2.5-7B-Instruct`) to be downloaded, or the path to the checkpoint folder. It is default to be `'Qwen/Qwen2.5-7B-Instruct'`.
 - `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么？'`.
 - `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
 
 #### Sample Output
 ##### [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct)

python/llm/example/GPU/HuggingFace/LLM/qwen2.5/generate.py

@@ -18,20 +18,29 @@ import torch
 import time
 import argparse
 
-from transformers import AutoTokenizer
 
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Predict Tokens using generate() API for Qwen2.5 model')
     parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2.5-7B-Instruct",
-                        help='The huggingface repo id for the Qwen2.5 model to be downloaded'
+                        help='The Hugging Face or ModelScope repo id for the Qwen2.5 model to be downloaded'
                              ', or the path to the huggingface checkpoint folder')
     parser.add_argument('--prompt', type=str, default="AI是什么？",
                         help='Prompt to infer') 
     parser.add_argument('--n-predict', type=int, default=32,
                         help='Max tokens to predict')
+    parser.add_argument('--modelscope', action="store_true", default=False, 
+                        help="Use models from modelscope")
 
     args = parser.parse_args()
+    
+    if args.modelscope:
+        from modelscope import AutoTokenizer
+        model_hub = 'modelscope'
+    else:
+        from transformers import AutoTokenizer
+        model_hub = 'huggingface'
+    
     model_path = args.repo_id_or_model_path
 
 
@@ -42,7 +51,8 @@ if __name__ == '__main__':
                                                  load_in_4bit=True,
                                                  optimize_model=True,
                                                  trust_remote_code=True,
-                                                 use_cache=True)
+                                                 use_cache=True,
+                                                 model_hub=model_hub)
     model = model.half().to("xpu")
 
     # Load tokenizer

python/llm/example/GPU/HuggingFace/LLM/qwen2/README.md

@@ -1,5 +1,5 @@
 # Qwen2
-In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) and [Qwen/Qwen2-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2-1.5B-Instruct) as reference Qwen2 models.
+In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Qwen2 models on [Intel GPUs](../../../README.md). For illustration purposes, we utilize [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct) and [Qwen/Qwen2-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2-1.5B-Instruct) (or [Qwen/Qwen2-7B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2-7B-Instruct) and [Qwen/Qwen2-1.5B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2-1.5B-Instruct) for ModelScope) as reference Qwen2 models.
 
 ## 0. Requirements
 To run these examples with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../README.md#requirements) for more information.
@@ -16,6 +16,9 @@ conda activate llm
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
 
 pip install transformers==4.37.0 # install transformers which supports Qwen2
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 #### 1.2 Installation on Windows
@@ -28,6 +31,9 @@ conda activate llm
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
 
 pip install transformers==4.37.0 # install transformers which supports Qwen2
+
+# [optional] only needed if you would like to use ModelScope as model hub
+pip install modelscope==1.11.0
 ```
 
 ### 2. Configures OneAPI environment variables for Linux
@@ -95,14 +101,19 @@ set SYCL_CACHE_PERSISTENT=1
 > For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile.
 ### 4. Running examples
 
-```
+```bash
+# for Hugging Face model hub
 python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT
+
+# for ModelScope model hub
+python ./generate.py --repo-id-or-model-path REPO_ID_OR_MODEL_PATH --prompt PROMPT --n-predict N_PREDICT --modelscope
 ```
 
 Arguments info:
-- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen2 model (e.g. `Qwen/Qwen2-7B-Instruct`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`.
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the  **Hugging Face** or **ModelScope** repo id for the Qwen2 model (e.g. `Qwen/Qwen2-7B-Instruct`) to be downloaded, or the path to the checkpoint folder. It is default to be `'Qwen/Qwen2-7B-Instruct'`.
 - `--prompt PROMPT`: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be `'AI是什么？'`.
 - `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+- `--modelscope`: using **ModelScope** as model hub instead of **Hugging Face**.
 
 #### Sample Output
 ##### [Qwen/Qwen2-7B-Instruct](https://huggingface.co/Qwen/Qwen2-7B-Instruct)

python/llm/example/GPU/HuggingFace/LLM/qwen2/generate.py

@@ -18,21 +18,30 @@ import torch
 import time
 import argparse
 
-from transformers import AutoTokenizer
 import numpy as np
 
 
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser(description='Qwen2-7B-Instruct')
+    parser = argparse.ArgumentParser(description='Predict Tokens using `generate()` API for Qwen2 model')
     parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen2-7B-Instruct",
-                        help='The huggingface repo id for the Qwen2 model to be downloaded'
+                        help='The Hugging Face or ModelScope repo id for the Qwen2 model to be downloaded'
-                             ', or the path to the huggingface checkpoint folder')
+                             ', or the path to the checkpoint folder')
     parser.add_argument('--prompt', type=str, default="AI是什么？",
                         help='Prompt to infer') 
     parser.add_argument('--n-predict', type=int, default=32,
                         help='Max tokens to predict')
+    parser.add_argument('--modelscope', action="store_true", default=False, 
+                        help="Use models from modelscope")
 
     args = parser.parse_args()
+
+    if args.modelscope:
+        from modelscope import AutoTokenizer
+        model_hub = 'modelscope'
+    else:
+        from transformers import AutoTokenizer
+        model_hub = 'huggingface'
+    
     model_path = args.repo_id_or_model_path
 
 
@@ -43,7 +52,8 @@ if __name__ == '__main__':
                                                  load_in_4bit=True,
                                                  optimize_model=True,
                                                  trust_remote_code=True,
-                                                 use_cache=True)
+                                                 use_cache=True,
+                                                 model_hub=model_hub)
     model = model.half().to("xpu")
 
     # Load tokenizer