Downloads · 30 days
21
100% of all-time downloads
gopalanj/chessjev-byte-v1
chessjev-byte-v1 is a text classification model from gopalanj. Use it when you need a label for a piece of text. It is set up for pytorch. The card lists the license as mit.
Chess-specialized one-pass option scorer. Architecture follows the public jevlike interface (context + N text options → N probabilities) with an original ChessJev implementation.
Downloads · 30 days
21
100% of all-time downloads
All-time downloads
21
Public
Repo size
176 KB
Likes
0
Public
Click a slice to open those files.
.pt176 KB · 98%
From the Hugging Face model README
Chess-specialized one-pass option scorer. Architecture follows the public jevlike interface (context + N text options → N probabilities) with an original ChessJev implementation.
This is not TypeSafe Jev and is not OpenJev 4B. It is a small byte-embedding model trained on synthetic chess rows (FEN context, legal UCI options, label = heuristic/tactical teacher). No Elo or calibration claim versus Jev.
model.pt — PyTorch checkpoint (OptionAttentionScorer)config.json — architecture hyperparametersfrom pathlib import Path
from chess_jev.model import load_checkpoint
model = load_checkpoint(Path("model.pt"))
Train:
PYTHONPATH=src python -m chess_jev.train --data data/chessjev --output checkpoints/chessjev-byte-v1/model.pt