mirror of
https://github.com/TabbyML/tabby
synced 2024-11-23 10:05:08 +00:00
a0b438da06
* Add python backend * Split docker-compose.triton.yml * update makefile |
||
---|---|---|
.github/workflows | ||
admin | ||
converter | ||
deployment | ||
preprocess | ||
server | ||
testdata/tiny-70M | ||
.dockerignore | ||
.gitattributes | ||
.gitignore | ||
.pre-commit-config.yaml | ||
LICENSE | ||
Makefile | ||
poetry.lock | ||
pyproject.toml | ||
README.md |
🐾 Tabby
Warning
This repository is undering heavy construction, everything changes fast.
An opensource / on-prem alternative to GitHub Copilot
Contents
admin
: Admin panel for monitoring / settings purpose.server
: API server for completion requests. It also logs users' selections (as feedback to model's quality).deployment
: Container related deployment configs.converter
: Converts a transformers causal LM model into TensorRT / FasterTransformer serving formats.preprocess
: Preprocess files into datasetstabformer
: NOT RELEASED Trainer(PEFT w/RLHF) for tabby models.
Development
Assuming Linux workstation with:
- docker
- docker w/ gpu driver
- python 3.10
Use make setup-development-environment
to setup basic dev environment, and make dev
to start local development server.