English | 简体中文

Papilot

Papilot - an open-source GitHub Copilot server based PaddleNLP

Support

	localhost	Docker
CPU	✅	✅
GPU	✅	❌
GPUS	❌	❌

Uses

Windows

First, install the dependencies:

pip install -r requirements.txt

Then, run the python setup.py

██████╗  █████╗ ██████╗ ██╗██╗      ██████╗ ████████╗
██╔══██╗██╔══██╗██╔══██╗██║██║     ██╔═══██╗╚══██╔══╝
██████╔╝███████║██████╔╝██║██║     ██║   ██║   ██║
██╔═══╝ ██╔══██║██╔═══╝ ██║██║     ██║   ██║   ██║
██║     ██║  ██║██║     ██║███████╗╚██████╔╝   ██║
╚═╝     ╚═╝  ╚═╝╚═╝     ╚═╝╚══════╝ ╚═════╝    ╚═╝

An open-source GitHub Copilot server based PaddleNLP


                 Models available

  ID   Model                VRAM   Languages
 ─────────────────────────────────────────────────
  1    codegen-350M-mono    2GB    Python-only
  2    codegen-350M-multi   2GB    multi-language
  3    codegen-2B-mono      7GB    Python-only
  4    codegen-2B-multi     7GB    multi-language
  5    codegen-6B-mono      13GB   Python-only
  6    codegen-6B-multi     13GB   multi-language
  7    codegen-16B-mono     32GB   Python-only
  8    codegen-16B-multi    32GB   multi-language

Select model [default: 1]:
Enter number of GPUs [default: 1]:
Too long tokens will affect the speed.
Enter lock token length [default: 32]: 16
  Deploy method

  ID   Method
 ────────────────
  1    Docker
  2    localhost

Where do you want to deploy the model? [default: localhost]:
Enter port [default: 8000]:
Configuration completed...
                 Configuration

  Key             Value
 ──────────────────────────────────────────────
  MODEL           Salesforce/codegen-350M-mono
  NUM_GPUS        1
  DEPLOY_METHOD   localhost
  PORT            8000
  TOKEN_LENGTH    16

Enter to start deployment:

Under the windows terminal, only localhost can be deployed, you should run in wsl

Linux

Run the setup.sh script or run the python scriptpython setup.py(recommend). The inference model will be configured.

Models available:
[1] codegen-350M-mono (2GB total VRAM required; Python-only)
[2] codegen-350M-multi (2GB total VRAM required; multi-language)
[3] codegen-2B-mono (7GB total VRAM required; Python-only)
[4] codegen-2B-multi (7GB total VRAM required; multi-language)
[5] codegen-6B-mono (13GB total VRAM required; Python-only)
[6] codegen-6B-multi (13GB total VRAM required; multi-language)
[7] codegen-16B-mono (32GB total VRAM required; Python-only)
[8] codegen-16B-multi (32GB total VRAM required; multi-language)
Enter your choice [1]:
Enter number of GPUs [1]:
Deployment method:
[1] Deploy to Docker
[2] Deploy to localhost
Where do you want to deploy the Papilot [localhost]?
Port [8000]:
lock_max_tokens [16]:  LOCK_MAX_TOKENS
Deploying to localhost
install dependencies......
...
INFO:     Started server process [1116]
INFO 2022-08-05 18:52:50,595 server.py:75] Started server process [1116]
INFO:     Waiting for application startup.
INFO 2022-08-05 18:52:50,596 on.py:47] Waiting for application startup.
INFO:     Application startup complete.
INFO 2022-08-05 18:52:50,596 on.py:61] Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
INFO 2022-08-05 18:52:50,599 server.py:207] Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)

This will configure the runtime environment variables and deploy a FastAPI backend service.

The backend can be started via python main.py if config.env exists

API

After the above is completed, the listening service will be started at http://0.0.0.0:8000. You can test (partially supported) through the standard OpenAI API, for example:

$ ipython

Python 3.9.12 (main, Jun  1 2022, 11:38:51)
Type 'copyright', 'credits' or 'license' for more information
IPython 8.4.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import openai

In [2]: openai.api_key = 'dummy'

In [3]: openai.api_base = 'http://127.0.0.1:8000/v1'

In [4]: result = openai.Completion.create(engine='codegen', prompt='def hello', max_tokens=16, temperature=0.1)

In [5]: result
Out[5]:
<OpenAIObject text_completion id=cmpl-dmhoeHmcw9DJ4NeqOJDQVKv3iivJ0 at 0x7fe7a81d42c0> JSON: {
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null,
      "text": "_world():\n    print(\"Hello World!\")\n\n\n#"
    }
  ],
  "created": 1659699508,
  "id": "cmpl-dmhoeHmcw9DJ4NeqOJDQVKv3iivJ0",
  "model": "codegen",
  "object": "text_completion",
  "usage": {
    "completion_tokens": null,
    "prompt_tokens": null,
    "total_tokens": null
  }
}

Or test via swagger at http://127.0.0.1:8000/docs#

Copilot Plugin

Just like fauxpilot we can set settings.json to modify the backend address of the Copilot plugin.

Unfortunately, no relevant documentation was found. Only this discuss is relevant

    "github.copilot.advanced": {
        "debug.overrideEngine": "codegen",
        "debug.testOverrideProxyUrl": "http://127.0.0.1:8000",
        "debug.overrideProxyUrl": "http://127.0.0.1:8000"
    }

Papilot-extension

In addition, Papilot has also open sourced the corresponding vscode extension, you can check the details at the link below.

Papilot-extension

Notes

The inference speed of the current inference service is relatively slow, so there is a TOKEN_LENGTH environment variable in setup.sh to lock the maximum length to improve the inference speed. If you want to experience more powerful completion, you can increase the value of this variable.

Acknowledge

This repo references the FauxPilot repo, without his work there would be no following. Thank.