diff --git a/README.md b/README.md index ce84f188..ff02fb83 100644 --- a/README.md +++ b/README.md @@ -153,6 +153,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa | Whisper | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/whisper) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/whisper) | | Phi-1_5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/phi-1_5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/phi-1_5) | | Flan-t5 | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/flan-t5) | [link](python/llm/example/GPU/HF-Transformers-AutoModels/Model/flan-t5) | +| Qwen-VL | [link](python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | | ***For more details, please refer to the `bigdl-llm` [Document](https://test-bigdl-llm.readthedocs.io/en/main/doc/LLM/index.html), [Readme](python/llm), [Tutorial](https://github.com/intel-analytics/bigdl-llm-tutorial) and [API Doc](https://bigdl.readthedocs.io/en/latest/doc/PythonAPI/LLM/index.html).*** diff --git a/python/llm/README.md b/python/llm/README.md index 4fcab1e4..6032a992 100644 --- a/python/llm/README.md +++ b/python/llm/README.md @@ -60,6 +60,7 @@ Over 20 models have been optimized/verified on `bigdl-llm`, including *LLaMA/LLa | Whisper | [link](example/CPU/HF-Transformers-AutoModels/Model/whisper) | [link](example/GPU/HF-Transformers-AutoModels/Model/whisper) | | Phi-1_5 | [link](example/CPU/HF-Transformers-AutoModels/Model/phi-1_5) | [link](example/GPU/HF-Transformers-AutoModels/Model/phi-1_5) | | Flan-t5 | [link](example/CPU/HF-Transformers-AutoModels/Model/flan-t5) | [link](example/GPU/HF-Transformers-AutoModels/Model/flan-t5) | +| Qwen-VL | [link](example/CPU/HF-Transformers-AutoModels/Model/qwen-vl) | | ### Working with `bigdl-llm` diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md index 92cb9103..f7bd3b55 100644 --- a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/README.md @@ -25,6 +25,8 @@ You can use BigDL-LLM to run any Huggingface Transformer models with INT4 optimi | Replit | [link](replit) | | Mistral | [link](mistral) | | Flan-t5 | [link](flan-t5) | +| Phi-1_5 | [link](phi-1_5) | +| Qwen-VL | [link](qwen-vl) | ## Recommended Requirements To run the examples, we recommend using Intel® Xeon® processors (server), or >= 12th Gen Intel® Core™ processor (client). diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl/README.md b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl/README.md new file mode 100644 index 00000000..8c04bbb6 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl/README.md @@ -0,0 +1,91 @@ +# Qwen-VL +In this directory, you will find examples on how you could apply BigDL-LLM INT4 optimizations on Qwen-VL models. 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, 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. +### 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 + +pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option + +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. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```powershell +python ./chat.py +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.2 Server +For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket. + +E.g. on Linux, +```bash +# set BigDL-Nano env variables +source bigdl-nano-init + +# e.g. for a server with 48 cores per socket +export OMP_NUM_THREADS=48 +numactl -C 0-47 -m 0 python ./chat.py +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.3 Arguments Info +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path`: str, argument defining the huggingface repo id for the Qwen-VL model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen-VL-Chat'`. +- `--n-predict`: int, 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. + +#### 2.4 Sample Chat +#### [Qwen/Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat) + +```log +-------------------- Session 1 -------------------- + Please input a picture: https://images.unsplash.com/photo-1533738363-b7f9aef128ce?auto=format&fit=crop&q=60&w=500&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxzZWFyY2h8NHx8Y2F0fGVufDB8fDB8fHwy + Please enter the text: 这是什么 +---------- Response ---------- +图中是一只戴着墨镜的酷炫猫咪，正坐在窗边，看着窗外。 + +-------------------- Session 2 -------------------- + Please input a picture: + Please enter the text: 这只猫猫多大了？ +---------- Response ---------- +由于只猫猫戴着太阳镜，无法判断年龄，但可以猜测它应该是一只成年猫猫，已经成年。 + +-------------------- Session 3 -------------------- + Please input a picture: + Please enter the text: 在图中检测框出猫猫的墨镜 +---------- Response ---------- +猫猫的墨镜(398,313),(994,506) + +-------------------- Session 4 -------------------- + Please input a picture: exit +``` +The sample input image in Session 1 is (which is fetched from [here](https://images.unsplash.com/photo-1533738363-b7f9aef128ce?auto=format&fit=crop&q=60&w=500&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxzZWFyY2h8NHx8Y2F0fGVufDB8fDB8fHwy)): + + + +The sample output image in Session 3 is: + + + + + diff --git a/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl/chat.py b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl/chat.py new file mode 100644 index 00000000..6c017755 --- /dev/null +++ b/python/llm/example/CPU/HF-Transformers-AutoModels/Model/qwen-vl/chat.py @@ -0,0 +1,85 @@ +# +# 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. +# + +from bigdl.llm.transformers import AutoModel, AutoModelForCausalLM +from transformers import AutoTokenizer, LlamaTokenizer +from transformers.generation import GenerationConfig +import torch +import time +import os +import argparse +from bigdl.llm import optimize_model +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, + device_map="cpu", + trust_remote_code=True, + modules_to_not_convert=['c_fc', 'out_proj'] ) + + # 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) + + 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/CPU/PyTorch-Models/Model/README.md b/python/llm/example/CPU/PyTorch-Models/Model/README.md index 3cee8c45..40288dd4 100644 --- a/python/llm/example/CPU/PyTorch-Models/Model/README.md +++ b/python/llm/example/CPU/PyTorch-Models/Model/README.md @@ -11,6 +11,8 @@ You can use `optimize_model` API to accelerate general PyTorch models on Intel s | Bark | [link](bark) | | Mistral | [link](mistral) | | Flan-t5 | [link](flan-t5) | +| Phi-1_5 | [link](phi-1_5) | +| Qwen-VL | [link](qwen-vl) | ## Recommended Requirements To run the examples, we recommend using Intel® Xeon® processors (server), or >= 12th Gen Intel® Core™ processor (client). diff --git a/python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/README.md b/python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/README.md new file mode 100644 index 00000000..444929ff --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/README.md @@ -0,0 +1,90 @@ +# Qwen-VL +In this directory, you will find examples on how you could use BigDL-LLM `optimize_model` API to accelerate Qwen-VL models. 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, 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. +### 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 + +pip install --pre --upgrade bigdl-llm[all] # install the latest bigdl-llm nightly build with 'all' option + +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. Run +After setting up the Python environment, you could run the example by following steps. + +#### 2.1 Client +On client Windows machines, it is recommended to run directly with full utilization of all cores: +```powershell +python ./chat.py +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.2 Server +For optimal performance on server, it is recommended to set several environment variables (refer to [here](../README.md#best-known-configuration-on-linux) for more information), and run the example with all the physical cores of a single socket. + +E.g. on Linux, +```bash +# set BigDL-Nano env variables +source bigdl-nano-init + +# e.g. for a server with 48 cores per socket +export OMP_NUM_THREADS=48 +numactl -C 0-47 -m 0 python ./chat.py +``` +More information about arguments can be found in [Arguments Info](#23-arguments-info) section. The expected output can be found in [Sample Output](#24-sample-output) section. + +#### 2.3 Arguments Info +In the example, several arguments can be passed to satisfy your requirements: + +- `--repo-id-or-model-path`: str, argument defining the huggingface repo id for the Qwen-VL model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be `'Qwen/Qwen-VL-Chat'`. +- `--n-predict`: int, 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. + +#### 2.4 Sample Chat +#### [Qwen/Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat) + +```log +-------------------- Session 1 -------------------- + Please input a picture: https://images.unsplash.com/photo-1533738363-b7f9aef128ce?auto=format&fit=crop&q=60&w=500&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxzZWFyY2h8NHx8Y2F0fGVufDB8fDB8fHwy + Please enter the text: 这是什么 +---------- Response ---------- +图中是一只戴着墨镜的酷炫猫咪，正坐在窗边，看着窗外。 + +-------------------- Session 2 -------------------- + Please input a picture: + Please enter the text: 这只猫猫多大了？ +---------- Response ---------- +由于只猫猫戴着太阳镜，无法判断年龄，但可以猜测它应该是一只成年猫猫，已经成年。 + +-------------------- Session 3 -------------------- + Please input a picture: + Please enter the text: 在图中检测框出猫猫的墨镜 +---------- Response ---------- +猫猫的墨镜(398,313),(994,506) + +-------------------- Session 4 -------------------- + Please input a picture: exit +``` + +The sample input image in Session 1 is (which is fetched from [here](https://images.unsplash.com/photo-1533738363-b7f9aef128ce?auto=format&fit=crop&q=60&w=500&ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxzZWFyY2h8NHx8Y2F0fGVufDB8fDB8fHwy)): + + + +The sample output image in Session 3 is: + + + diff --git a/python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/chat.py b/python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/chat.py new file mode 100644 index 00000000..5502a697 --- /dev/null +++ b/python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/chat.py @@ -0,0 +1,85 @@ +# +# 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. +# + +from transformers import AutoModelForCausalLM, AutoTokenizer +from transformers.generation import GenerationConfig +import torch +import time +import os +import argparse +from bigdl.llm import optimize_model +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']) + + # 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) + + 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