2024 GrandmasterLevelChessWithoutSea

Subject Headings: Chess Playing Algorithm, Stockfish, Chess Playing.

Notes

The paper introduces a novel approach for achieving grandmaster-level chess play without traditional search algorithms.
The paper utilizes a transformer model with 270 million parameters trained on 10 million chess games annotated with action-values by Stockfish 16.
The paper achieves a blitz Elo rating of 2895 on Lichess, demonstrating high-level performance against human players.
The paper compares the model's performance with AlphaZero and GPT-3.5-turbo-instruct, showing its superiority without the need for Monte Carlo tree search (MCTS).
The paper emphasizes the importance of model and dataset size, indicating strong chess capabilities emerge with sufficient scale.
The paper conducts extensive ablation studies to validate the significance of scale and specific design choices for the model's success.
The paper contributes insights into the potential of supervised learning in complex tasks like chess, traditionally dominated by search-based methods.

The recent breakthrough successes in machine learning are mainly attributed to scale: namely large-scale attention-based architectures and datasets of unprecedented scale. This paper investigates the impact of training at scale for chess. Unlike traditional chess engines that rely on complex heuristics, explicit search, or a combination of both, we train a 270M parameter transformer model with supervised learning on a dataset of 10 million chess games. We annotate each board in the dataset with action-values provided by the powerful Stockfish 16 engine, leading to roughly 15 billion data points. Our largest model reaches a Lichess blitz Elo of 2895 against humans, and successfully solves a series of challenging chess puzzles, without any domain-specific tweaks or explicit search algorithms. We also show that our model outperforms AlphaZero's policy and value networks (without MCTS) and GPT-3.5-turbo-instruct. A systematic investigation of model and dataset size shows that strong chess performance only arises at sufficient scale. To validate our results, we perform an extensive series of ablations of design choices and hyperparameters.

;

	Author	volume	Date Value	title	type	journal	titleUrl	doi	note	year
2024 GrandmasterLevelChessWithoutSea	Anian Ruoss Sourabh Medapati Jordi Grau-Moya Li Kevin Wenliang Elliot Catt John Reid Tim Genewein Grégoire Delétang			Grandmaster-Level Chess Without Search				10.48550/arXiv.2402.04494		2024