Add qwen vl CPU example (#9221)
* eee * add examples on CPU and GPU * fix * fix * optimize model examples * add Qwen-VL-Chat CPU example * Add Qwen-VL CPU example * fix optimize problem * fix error * Have updated, benchmark fix removed from this PR * add generate API example * Change formats in qwen-vl example * Add CPU transformer int4 example for qwen-vl * fix repo-id problem and add Readme * change picture url * Remove unnecessary file --------- Co-authored-by: Yuwen Hu <yuwen.hu@intel.com>
This commit is contained in:
Yining Wang
GitHub
parent
f597a9d4f5
commit
a6a8afc47e
8 changed files with 357 additions and 0 deletions
|
|
@ -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).***
|
||||
|
||||
|
|
|
|||
|
|
@ -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`
|
||||
|
||||
|
|
|
|||
|
|
@ -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).
|
||||
|
|
|
|||
|
|
@ -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 ----------
|
||||
<ref>猫猫的墨镜</ref><box>(398,313),(994,506)</box>
|
||||
|
||||
-------------------- 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)):
|
||||
|
||||
<a href="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-input.jpg"><img width=250px src="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-input.jpg" ></a>
|
||||
|
||||
The sample output image in Session 3 is:
|
||||
|
||||
<a href="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output.png"><img width=250px src="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output.png" ></a>
|
||||
|
||||
|
||||
|
||||
|
|
@ -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
|
||||
|
|
@ -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).
|
||||
|
|
|
|||
|
|
@ -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 ----------
|
||||
<ref>猫猫的墨镜</ref><box>(398,313),(994,506)</box>
|
||||
|
||||
-------------------- 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)):
|
||||
|
||||
<a href="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-input.jpg"><img width=250px src="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-input.jpg" ></a>
|
||||
|
||||
The sample output image in Session 3 is:
|
||||
|
||||
<a href="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output.png"><img width=250px src="https://llm-assets.readthedocs.io/en/latest/_images/qwen-vl-example-output.png" ></a>
|
||||
|
||||
85
python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/chat.py
Normal file
85
python/llm/example/CPU/PyTorch-Models/Model/qwen-vl/chat.py
Normal file
|
|
@ -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
|
||||
Loading…
Reference in a new issue