LLM: Add qwen-vl gpu example (#9290)

* create qwen-vl gpu example.

* add readme.

* fix.

* change input figure and update outputs.

* add qwen-vl pytorch model gpu example.

* fix.

* add readme.

README.md

@@ -149,6 +149,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
 | Baichuan2  | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/baichuan2) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/baichuan2)  |
 | InternLM   | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/internlm)  | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/internlm)   |
 | Qwen       | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen)      | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen)       |
+| Qwen-VL    | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl)   | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl)    |
 | Aquila     | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/aquila)    | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/aquila)     |
 | MOSS       | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/moss)      |    | 
 | Whisper    | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/whisper)   | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/whisper)    |

python/llm/README.md

@@ -56,6 +56,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa
 | Baichuan2  | [link](example/CPU/HF-Transformers-AutoModels/Model/baichuan2) | [link](example/GPU/HF-Transformers-AutoModels/Model/baichuan2)  |
 | InternLM   | [link](example/CPU/HF-Transformers-AutoModels/Model/internlm)  | [link](example/GPU/HF-Transformers-AutoModels/Model/internlm)   |
 | Qwen       | [link](example/CPU/HF-Transformers-AutoModels/Model/qwen)      | [link](example/GPU/HF-Transformers-AutoModels/Model/qwen)       |
+| Qwen-VL    | [link](example/CPU/HF-Transformers-AutoModels/Model/qwen-vl)   | [link](example/GPU/HF-Transformers-AutoModels/Model/qwen-vl)    |
 | Aquila     | [link](example/CPU/HF-Transformers-AutoModels/Model/aquila)    | [link](example/GPU/HF-Transformers-AutoModels/Model/aquila)     |
 | MOSS       | [link](example/CPU/HF-Transformers-AutoModels/Model/moss)      |    | 
 | Whisper    | [link](example/CPU/HF-Transformers-AutoModels/Model/whisper)   | [link](example/GPU/HF-Transformers-AutoModels/Model/whisper)    |

python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl/README.md

@@ -0,0 +1,78 @@
+# Qwen-VL
+In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Qwen-VL models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [Qwen/Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat) as a reference Qwen-VL model.
+
+## Requirements
+To run these examples with BigDL-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Multimodal chat using `chat()` API
+In the example [chat.py](./chat.py), we show a basic use case for a Qwen-VL model to start a multimodal chat using `chat()` API, with BigDL-LLM INT4 optimizations on Intel GPUs.
+### 1. Install
+We suggest using conda to manage the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#).
+
+After installing conda, create a Python environment for BigDL-LLM:
+```bash
+conda create -n llm python=3.9 # recommend to use Python 3.9
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.0.110+xpu as default
+# you can install specific ipex/torch version for your need
+pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
+pip install accelerate tiktoken einops transformers_stream_generator==0.0.4 scipy torchvision pillow tensorboard matplotlib # additional package required for Qwen-VL-Chat to conduct generation
+```
+
+### 2. Configures OneAPI environment variables
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Run
+
+For optimal performance on Arc, it is recommended to set several environment variables.
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+```
+```
+python ./chat.py
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen-VL model (e.g `Qwen/Qwen-VL-Chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen-VL-Chat'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+  
+In every session, image and text can be entered into cmd (user can skip the input by type **'Enter'**) ; please type **'exit'** anytime you want to quit the dialouge.
+
+Every image output will be named as the round of session and placed under the current directory.
+
+#### Sample Chat
+#### [Qwen/Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat)
+
+```log
+-------------------- Session 1 --------------------
+ Please input a picture: http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg
+ Please enter the text: 这是什么？
+---------- Response ----------
+这是一张图片，展现了一个穿着粉色条纹连衣裙的小女孩，她手持一只穿粉色裙子的小熊。这个场景发生在一个户外环境，有砖块背景墙和花朵。
+
+-------------------- Session 2 --------------------
+ Please input a picture:
+ Please enter the text: 这个小女孩多大了？
+---------- Response ----------
+根据图片中的描述，这个小女孩应该是年龄较小的孩子，但具体年龄难以确定。从她的外表来看，可能是在5岁左右。。 
+
+-------------------- Session 3 --------------------
+ Please input a picture: 
+ Please enter the text: 在图中检测框出玩具熊
+---------- Response ----------
+<ref>玩具熊</ref><box>(330,267),(603,869)</box>
+
+-------------------- Session 4 --------------------
+ Please input a picture: exit
+```
+The sample input image in Session 1 is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=264959)):
+
+<a href="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg"><img width=400px src="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg" ></a>
+
+The sample output image in Session 3 is:
+
+<a href="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output-gpu.png"><img width=400px src="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output-gpu.png" ></a>

python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl/chat.py

@@ -0,0 +1,98 @@
+#
+# 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 argparse
+import os
+
+import torch
+from transformers import AutoTokenizer
+from transformers.generation import GenerationConfig
+
+from bigdl.llm.transformers import AutoModelForCausalLM
+import intel_extension_for_pytorch as ipex
+
+torch.manual_seed(1234)
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `chat()` API for Qwen-VL model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen-VL-Chat",
+                        help='The huggingface repo id for the Qwen-VL model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--n-predict', type=int, default=32, help='Max tokens to predict')
+    
+    current_path = os.path.dirname(os.path.abspath(__file__))
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path  
+        
+    # Load model
+    # For successful BigDL-LLM optimization on Qwen-VL-Chat, skip the 'c_fc' and 'out_proj' modules during optimization
+    model = AutoModelForCausalLM.from_pretrained(model_path, 
+                                                 load_in_4bit=True, 
+                                                 trust_remote_code=True, 
+                                                 modules_to_not_convert=['c_fc', 'out_proj'])
+    model = model.to('xpu')
+    # Due to issue https://github.com/intel/intel-extension-for-pytorch/issues/454,
+    # currently put interpolation execution into cpu
+    def to_cpu(module, input, output):
+        return output.to("cpu")
+
+    def to_xpu(module, input):
+        return (input[0].to("xpu"),)
+
+    model.transformer.visual.ln_pre.register_forward_hook(to_cpu)
+    model.transformer.visual.transformer.register_forward_pre_hook(to_xpu)
+
+    # Specify hyperparameters for generation (No need to do this if you are using transformers>=4.32.0)
+    model.generation_config = GenerationConfig.from_pretrained(model_path, trust_remote_code=True)
+    
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+
+    # Session ID
+    session_id = 1
+
+    while True:
+      print('-'*20, 'Session %d' % session_id, '-'*20)
+      image_input = input(f' Please input a picture: ')
+      if image_input.lower() == 'exit' : # type 'exit' to quit the dialouge
+         break
+
+      text_input = input(f' Please enter the text: ')
+      if text_input.lower() == 'exit' : # type 'exit' to quit the dialouge
+         break
+      
+      if session_id == 1:
+         history = None
+
+      all_input = [{'image': image_input}, {'text': text_input}]
+      input_list = [_input for _input in all_input if list(_input.values())[0] != '']
+
+      if len(input_list) == 0:
+         print("Input list should not be empty. Please try again with valid input.")
+         continue
+      
+      query = tokenizer.from_list_format(input_list)
+      response, history = model.chat(tokenizer, query = query, history = history)
+      torch.xpu.synchronize()
+
+      print('-'*10, 'Response', '-'*10)
+      print(response, '\n')
+
+      image = tokenizer.draw_bbox_on_latest_picture(response, history)
+      if image is not None:
+         image.save(os.path.join(current_path, f'Session_{session_id}.png'), )
+
+      session_id += 1

python/llm/example/GPU/PyTorch-Models/Model/qwen-vl/README.md

@@ -0,0 +1,78 @@
+# Qwen-VL
+In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Qwen-VL models on [Intel GPUs](../README.md). For illustration purposes, we utilize the [Qwen/Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat) as a reference Qwen-VL model.
+
+## Requirements
+To run these examples with BigDL-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../README.md#recommended-requirements) for more information.
+
+## Example: Multimodal chat using `chat()` API
+In the example [chat.py](./chat.py), we show a basic use case for a Qwen-VL model to start a multimodal chat using `chat()` API, with BigDL-LLM 'optimize_model' API on Intel GPUs.
+### 1. Install
+We suggest using conda to manage the Python environment. For more information about conda installation, please refer to [here](https://docs.conda.io/en/latest/miniconda.html#).
+
+After installing conda, create a Python environment for BigDL-LLM:
+```bash
+conda create -n llm python=3.9 # recommend to use Python 3.9
+conda activate llm
+# below command will install intel_extension_for_pytorch==2.0.110+xpu as default
+# you can install specific ipex/torch version for your need
+pip install --pre --upgrade bigdl-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
+pip install accelerate tiktoken einops transformers_stream_generator==0.0.4 scipy torchvision pillow tensorboard matplotlib # additional package required for Qwen-VL-Chat to conduct generation
+```
+
+### 2. Configures OneAPI environment variables
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+
+### 3. Run
+
+For optimal performance on Arc, it is recommended to set several environment variables.
+
+```bash
+export USE_XETLA=OFF
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+```
+```
+python ./chat.py
+```
+
+Arguments info:
+- `--repo-id-or-model-path REPO_ID_OR_MODEL_PATH`: argument defining the huggingface repo id for the Qwen-VL model (e.g `Qwen/Qwen-VL-Chat`) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen-VL-Chat'`.
+- `--n-predict N_PREDICT`: argument defining the max number of tokens to predict. It is default to be `32`.
+  
+In every session, image and text can be entered into cmd (user can skip the input by type **'Enter'**) ; please type **'exit'** anytime you want to quit the dialouge.
+
+Every image output will be named as the round of session and placed under the current directory.
+
+#### Sample Chat
+#### [Qwen/Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat)
+
+```log
+-------------------- Session 1 --------------------
+ Please input a picture: http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg
+ Please enter the text: 这是什么？
+---------- Response ----------
+这是一张图片，展现了一个穿着粉色条纹连衣裙的小女孩，她手持一只穿粉色裙子的小熊。这个场景发生在一个户外环境，有砖块背景墙和花朵。
+
+-------------------- Session 2 --------------------
+ Please input a picture:
+ Please enter the text: 这个小女孩多大了？
+---------- Response ----------
+根据图片中的描述，这个小女孩应该是年龄较小的孩子，但具体年龄难以确定。从她的外表来看，可能是在5岁左右。。 
+
+-------------------- Session 3 --------------------
+ Please input a picture: 
+ Please enter the text: 在图中检测框出玩具熊
+---------- Response ----------
+<ref>玩具熊</ref><box>(330,267),(603,869)</box>
+
+-------------------- Session 4 --------------------
+ Please input a picture: exit
+```
+The sample input image in Session 1 is (which is fetched from [COCO dataset](https://cocodataset.org/#explore?id=264959)):
+
+<a href="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg"><img width=400px src="http://farm6.staticflickr.com/5268/5602445367_3504763978_z.jpg" ></a>
+
+The sample output image in Session 3 is:
+
+<a href="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output-gpu.png"><img width=400px src="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output-gpu.png" ></a>

python/llm/example/GPU/PyTorch-Models/Model/qwen-vl/chat.py

@@ -0,0 +1,100 @@
+#
+# 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 argparse
+import os
+
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers.generation import GenerationConfig
+
+from bigdl.llm import optimize_model
+import intel_extension_for_pytorch as ipex
+
+torch.manual_seed(1234)
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Predict Tokens using `chat()` API for Qwen-VL model')
+    parser.add_argument('--repo-id-or-model-path', type=str, default="Qwen/Qwen-VL-Chat",
+                        help='The huggingface repo id for the Qwen-VL model to be downloaded'
+                             ', or the path to the huggingface checkpoint folder')
+    parser.add_argument('--n-predict', type=int, default=32, help='Max tokens to predict')
+    
+    current_path = os.path.dirname(os.path.abspath(__file__))
+    args = parser.parse_args()
+    model_path = args.repo_id_or_model_path  
+        
+    # Load model
+    model = AutoModelForCausalLM.from_pretrained(model_path, device_map="cpu",  trust_remote_code=True)
+
+    # With only one line to enable BigDL-LLM optimization on model
+    # For successful BigDL-LLM optimization on Qwen-VL-Chat, skip the 'c_fc' and 'out_proj' modules during optimization
+    model = optimize_model(model, 
+                           low_bit='sym_int4', 
+                           modules_to_not_convert=['c_fc', 'out_proj'])
+    model = model.to('xpu')
+    # Due to issue https://github.com/intel/intel-extension-for-pytorch/issues/454,
+    # currently put interpolation execution into cpu
+    def to_cpu(module, input, output):
+        return output.to("cpu")
+
+    def to_xpu(module, input):
+        return (input[0].to("xpu"),)
+
+    model.transformer.visual.ln_pre.register_forward_hook(to_cpu)
+    model.transformer.visual.transformer.register_forward_pre_hook(to_xpu)
+
+    # Specify hyperparameters for generation (No need to do this if you are using transformers>=4.32.0)
+    model.generation_config = GenerationConfig.from_pretrained(model_path, trust_remote_code=True)
+    
+    # Load tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+
+    # Session ID
+    session_id = 1
+
+    while True:
+      print('-'*20, 'Session %d' % session_id, '-'*20)
+      image_input = input(f' Please input a picture: ')
+      if image_input.lower() == 'exit' : # type 'exit' to quit the dialouge
+         break
+
+      text_input = input(f' Please enter the text: ')
+      if text_input.lower() == 'exit' : # type 'exit' to quit the dialouge
+         break
+      
+      if session_id == 1:
+         history = None
+
+      all_input = [{'image': image_input}, {'text': text_input}]
+      input_list = [_input for _input in all_input if list(_input.values())[0] != '']
+
+      if len(input_list) == 0:
+         print("Input list should not be empty. Please try again with valid input.")
+         continue
+      
+      query = tokenizer.from_list_format(input_list)
+      response, history = model.chat(tokenizer, query = query, history = history)
+      torch.xpu.synchronize()
+
+      print('-'*10, 'Response', '-'*10)
+      print(response, '\n')
+
+      image = tokenizer.draw_bbox_on_latest_picture(response, history)
+      if image is not None:
+         image.save(os.path.join(current_path, f'Session_{session_id}.png'), )
+
+      session_id += 1