Skip to content
Other

chess-pgn-2400plus-seed1

character-level chess transformer

2 min readupdated

A compact 51-million-parameter language model by Arisp that predicts chess moves in PGN text format (the standard notation for recording games). Built on a character-level transformer architecture, it was trained on 10 million Lichess games where both players were rated 2400+. This checkpoint serves as a control run for the base chess-pgn-2400plus model, using a different random seed (seed 1) and a single cosine learning rate schedule over 250,000 steps. It is useful for researchers evaluating run-to-run variance and developers who need a lightweight move generator without heavy chess engines.

Key features

  • Character-level generation: processes raw PGN notation within a 1024-character context window, inferring board state solely from move history.
  • Compact footprint: only 51M parameters in safetensors format, making inference easy to run even on a CPU.
  • Improved validation loss: achieved 0.3566 bits per character on a 5,000-game held-out split, compared to 0.3622 for the first version.
  • Control checkpoint: helps separate real experimental gains from natural variance caused by changing the seed and learning rate schedule.

What you can do

  • Build a chess bot or training tool that mimics the play style of strong club players and 2400+ masters, including realistic human-like mistakes.
  • Pair it with the original chess-pgn-2400plus as a baseline when testing new hyperparameters and fine-tuning methods for PGN models.
  • Generate opening and middlegame continuations in Python scripts using the built-in generate_moves function and the python-chess library.

Example

import json, chess, sys
from huggingface_hub import snapshot_download
from safetensors.torch import load_file

path = snapshot_download("Arisp/chess-pgn-2400plus-seed1")
sys.path.insert(0, path)
from pgn_model import PGNTransformer, generate_moves

model = PGNTransformer(**json.load(open(f"{path}/config.json"))["architecture"])
model.load_state_dict(load_file(f"{path}/model.safetensors"))
model.eval()

board = chess.Board()
board.push_san("e4")
move = generate_moves(model, [board], temperature=0.0, dev="cpu")[0]
print(board.san(move))
Read moreDetails
Validation loss: 0.3566 bits per character (0.2472 nats) across 5,000 games.
Training: 250,000 steps (checkpoint saved at step 242,500), AdamW, bf16 on a single NVIDIA B200.
Limitations: not a full chess engine, occasionally outputs illegal moves, and loses context after move 80–100 due to the 1024-character window.
Discuss Telegram 0 23

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.2 GB · ready files0.7 GB
  • BF16weights 0.1 GB0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.6 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Arisp/chess-pgn-2400plus-seed1
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Arisp/chess-pgn-2400plus-seed1 --local-dir ~/models/chess-pgn-2400plus-seed1
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of chess-pgn-2400plus-seed1 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Arisp/chess-pgn-2400plus-seed1
Author
Arisp
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

230

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments