seanreynoldscs / tabby

Self-hosted AI coding assistant

🐾 Tabby

Self-hosted AI coding assistant. An opensource / on-prem alternative to GitHub Copilot.

Warning Tabby is still in the alpha phrase

Features

• Self-contained, with no need for a DBMS or cloud service
• Web UI for visualizing and configuration models and MLOps.
• OpenAPI interface, easy to integrate with existing infrastructure (e.g Cloud IDE).
• Consumer level GPU supports (FP-16 weight loading with various optimization).

Demo

Get started

Docker

The easiest way of getting started is using the docker image:

# Create data dir and grant owner to 1000 (Tabby run as uid 1000 in container)
mkdir -p data/hf_cache && chown -R 1000 data

docker run \
  -it --rm \
  -v ./data:/data \
  -v ./data/hf_cache:/home/app/.cache/huggingface \
  -p 5000:5000 \
  -e MODEL_NAME=TabbyML/J-350M \
  tabbyml/tabby

To use the GPU backend (triton) for a faster inference speed:

docker run \
  --gpus all \
  -it --rm \
  -v ./data:/data \
  -v ./data/hf_cache:/home/app/.cache/huggingface \
  -p 5000:5000 \
  -e MODEL_NAME=TabbyML/J-350M \
  -e MODEL_BACKEND=triton \
  tabbyml/tabby

Note: To use GPUs, you need to install the NVIDIA Container Toolkit. We also recommend using NVIDIA drivers with CUDA version 11.8 or higher.

You can then query the server using /v1/completions endpoint:

curl -X POST http://localhost:5000/v1/completions -H 'Content-Type: application/json' --data '{
    "prompt": "def binarySearch(arr, left, right, x):\n    mid = (left +"
}'

We also provides an interactive playground in admin panel localhost:5000/_admin

Skypilot

See deployment/skypilot/README.md

API documentation

Tabby opens an FastAPI server at localhost:5000, which embeds an OpenAPI documentation of the HTTP API.

Development

Go to development directory.

make dev

or

make dev-triton # Turn on triton backend (for cuda env developers)