grpo-chess-api / README.md
noamdwc's picture
chore: bake model into Docker image
41a2a9d
metadata
title: grpo-chess-api
emoji: ♟️
colorFrom: indigo
colorTo: red
sdk: docker
pinned: false

Hugging Face Space (Inference API)

This Space serves the chess model with a /move endpoint. The model weights are baked into the Docker image for instant startup — no runtime downloads needed.

Model Files

The model/ directory must contain:

  • model.safetensors — model weights (~24 MB, tracked via Git LFS)
  • config.json — model configuration

These are copied into the Docker image at build time.

Environment Variables

  • LOCAL_MODEL_DIR (default: /app/model) — path to model directory inside the container
  • WEIGHTS_FILE (default: model.safetensors)
  • CONFIG_FILE (default: config.json)

Local Docker Run

docker build -t grpo-chess-api .
docker run --rm -p 7860:7860 grpo-chess-api

API

POST /move

Request:

{ "fen": "...", "temperature": 1.0, "greedy": false }

Response:

{ "uci": "e2e4", "san": "e4", "fen": "..." }