Added support for Docker with GPU inference.

[dengsgo]

Added support for Docker with GPU inference.

Build Image

# cd your/path/to/ChatGLM2-6B
$ docker build -t chatglm2:v1 . 

Output similar execution results：

[+] Building 475.0s (9/9) FINISHED
 => [internal] load build definition from Dockerfile                                                               0.1s
 => => transferring dockerfile: 755B                                                                               0.0s
 => [internal] load .dockerignore                                                                                  0.1s
 => => transferring context: 2B                                                                                    0.0s
 => [internal] load metadata for docker.io/pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime                           0.0s
 => [internal] load build context                                                                                  0.1s
 => => transferring context: 11.00MB                                                                               0.1s
 => CACHED [1/4] FROM docker.io/pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime                                      0.0s
 => [2/4] COPY . .                                                                                                 0.1s
 => [3/4] RUN apt update && apt install -y git gcc                                                                93.8s
 => [4/4] RUN pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple/ && pip install icetk  372.3s
 => exporting to image                                                                                             8.6s
 => => exporting layers                                                                                            8.5s
 => => writing image sha256:828bdcc8d9b5c8537dc2f243633497f573b41518eae9e2dc11539d7f8d864eb5                       0.0s
 => => naming to docker.io/library/chatglm2:v1                                                                     0.0s

You will get the image of chatglm2:v1：

$ docker images
REPOSITORY   TAG              IMAGE ID            CREATED          SIZE
chatglm2       v1                828bdcc8d9b5   18 minutes ago   9.8GB

Use

The first time I need to download a model project, for example, I want to use chatglm2-6b-int4 and execute it in the path you think is suitable (such as /data/models):

$ cd  /data/models
# Make sure you have git-lfs installed (https://git-lfs.com)
$ git lfs install
$ git clone [email protected]:THUDM/chatglm2-6b-int4

Now you can start the docker：

$ docker run --rm -it -v /data/models/chatglm2-6b-int4:/workspace/THUDM/chatglm2-6b --gpus=all -e NVIDIA_DRIVER_CAPABILITIES=compute,utility -e NVIDIA_VISIBLE_DEVICES=all -p 7860:7860 chatglm2:v1

Output similar execution results：

/opt/conda/lib/python3.10/site-packages/gradio/components/textbox.py:259: UserWarning: The `style` method is deprecated. Please set these arguments in the constructor instead.
  warnings.warn(
Running on local URL:  http://0.0.0.0:7860

To create a public link, set `share=True` in `launch()`.

Open a local browser and enterhttp://localhost:7860/You can now open webUI.

Enjoy!

[nlpchen]

可以打包成支持gpu的镜像分享一下吗？多谢。[email protected]