Skip to content

localagi/oobabooga-docker

Repository files navigation

oobabooga-docker

Sophisticated docker builds for parent project oobabooga/text-generation-webui.

example workflow

Easy setup. Compatible. Tweakable. Scaleable.

TODO support different GPTQ-for-Llama's TODO fixp for compose mounts / dev env

Naming scheme

-cc-1.2.3 - NVidia compute capability, see NVidia CUDA GPUs

Supported platforms

amd64

Supported versions

main

Prerequisites

  • docker and docker compose are available on your system
NVIDIA card required

These containers require nvidia-container-toolkit installed and reboot

Run

Run text-generation-webui with a pretrained LLama-30B model on e.g. RTX 3090, RTX 4090

  • get docker-compose.yml (clone repo, copy or else)
  • Download OpenAssistant-SFT-7-Llama-30B-GPTQ into ../models
  • Run docker compose up
  • Model will load into gpu
  • open/refresh http://localhost:7860

Runtime options

Environment variables to set for the specific service

compute capability selection COMPUTE_VERSION

Prepend, e.g. COMPUTE_VERSION=6.1 for Tesla P40

local model directory LOCAL_MODEL_DIR

Prepend, e.g. LOCAL_MODEL_DIR=/my/path/to/models

Get the latest builds / update

docker compose pull

Cleanup

docker compose rm

Contributing

When there is a new version and there is need of builds or you require the latest main build, feel free to open an issue

Problems?

Open an issue on the Issue Tracker

Limitations

We cannot support issues regarding the base software. Please refer to the main project page mentioned in the second line of this card.