/transformers-stack

A full stack solution for deploying a transformers model from HuggingFace

Primary LanguageJavaScript

Transformers Stack

A full stack solution for deploying a language model from hugging face with a simple UI for prompting the model and tracking the results.

This project uses docker compose for orchestrating two containers, model which contains the model with a simple http interface, and ui which is a next.js application that provides access to the model in a UI.

Setup

update the variable MODEL_KEY in docker-compose.yml to the name of the model you want, for example bigscience/bloom-560m or EleutherAI/gpt-neo-1.3B.

And then run docker compose up --build! thats all you need! The UI will be available on http://localhost:9999 (you can change the port also in docker-compose.yml )

Run on CPU

create a file called docker-compose.override.yml and put the following in it:

services:
  model:
    deploy: {}