diff --git a/README.md b/README.md index db3bf58c..39ba3c7c 100644 --- a/README.md +++ b/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) | diff --git a/python/llm/README.md b/python/llm/README.md index 0a21375f..4877bf79 100644 --- a/python/llm/README.md +++ b/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) | diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl/README.md b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl/README.md new file mode 100644 index 00000000..19ca669f --- /dev/null +++ b/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 ---------- +玩具熊(330,267),(603,869) + +-------------------- 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)): + + + +The sample output image in Session 3 is: + + diff --git a/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl/chat.py b/python/llm/example/GPU/HF-Transformers-AutoModels/Model/qwen-vl/chat.py new file mode 100644 index 00000000..55d1eb47 --- /dev/null +++ b/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 diff --git a/python/llm/example/GPU/PyTorch-Models/Model/qwen-vl/README.md b/python/llm/example/GPU/PyTorch-Models/Model/qwen-vl/README.md new file mode 100644 index 00000000..9b7f2606 --- /dev/null +++ b/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 ---------- +玩具熊(330,267),(603,869) + +-------------------- 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)): + + + +The sample output image in Session 3 is: + + diff --git a/python/llm/example/GPU/PyTorch-Models/Model/qwen-vl/chat.py b/python/llm/example/GPU/PyTorch-Models/Model/qwen-vl/chat.py new file mode 100644 index 00000000..9e9220f0 --- /dev/null +++ b/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