Spaces:
Running on Zero
Running on Zero
|
Download README.md from datafreak/laya-chess: direct link, hf CLI and curl.
- Browser
- Download file 1.08 kB
-
https://huggingface.co/spaces/datafreak/laya-chess/resolve/main/README.md
- Command line
-
hf download hf://spaces/datafreak/laya-chess/README.md
-
curl -L -o README.md https://huggingface.co/spaces/datafreak/laya-chess/resolve/main/README.md
1.08 kB
A newer version of the Gradio SDK is available: 6.31.0
metadata
title: LayaChess
emoji: ♟️
colorFrom: gray
colorTo: red
sdk: gradio
sdk_version: 6.29.1
app_file: app.py
pinned: false
short_description: Play chess against Laya, a fine-tuned decision model
LayaChess
Play chess against Laya, a 421M-parameter System 1 decision model, fine-tuned on 2 million Stockfish-rated moves and wrapped in a Monte Carlo tree search.
- Write-up: https://devroopsaha744.github.io/portfolio/blog/laya-chess/
- Code: https://github.com/devroopsaha744/LayaChess
- Demo video: https://www.youtube.com/watch?v=bPpAlWArs7E
- Model: https://huggingface.co/datafreak/laya-chess
- Run it locally: see the setup steps
Laya thinks on ZeroGPU, a shared free GPU, so the first move can take a while and you may wait in a queue. Laya can think up to 30 seconds a move; that time comes out of each visitor's daily ZeroGPU quota. Stockfish's preferred move is shown next to Laya's for comparison only; it never picks Laya's moves.