Model Card for Llama-3.1-8B-Instruct-NLRL-TicTacToe-Value

Model Details

Model Description

• Developed by: NLRL Team
• Model type: Language Value Function Model for TicTacToe
• Language(s): English
• License: MIT
• Finetuned from model: LLaMA-3.1-8B-Instruct

This model serves as a language value function in Natural Language Reinforcement Learning (NLRL) framework, specifically trained for the TicTacToe game. It evaluates state-action pairs through natural language description and provides value assessment.

Uses

Direct Use

This model can be used as a TicTacToe position evaluator that explains its evaluation through natural language before providing the final assessment. The model generates both reasoning chains and final value judgments.

Out-of-Scope Use

This model is specifically trained for TicTacToe state-action evaluation and should not be used for other games or value assessment tasks.

Training Details

Training Data

Training data consists of state-action pairs collected through NLRL actor-critic learning process, with language-based Monte Carlo estimates serving as training targets for the value function.

Training Procedure

• Trained using FSDP (Fully Sharded Data Parallel) across 4 H100 GPUs
• Learning rate: 1e-5
• Training epochs per iteration: 2
• Batch size: 8
• Max sequence length: 1024

Evaluation

• Tested on both deterministic and stochastic gameplay trajectories
• Demonstrates consistent evaluation capabilities across different game states
• Works in conjunction with the policy model to guide action selection

Model Architecture

• Base model: LLaMA-3.1-8B-Instruct
• Input: Text description of TicTacToe state-action pair
• Output: Chain-of-thought evaluation followed by value assessment

Citation

@misc{feng2024naturallanguagereinforcementlearning,
      title={Natural Language Reinforcement Learning}, 
      author={Xidong Feng and Ziyu Wan and Haotian Fu and Bo Liu and Mengyue Yang and Girish A. Koushik and Zhiyuan Hu and Ying Wen and Jun Wang},
      year={2024},
      eprint={2411.14251},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2411.14251}, 
}

Model Card Contact

[email protected]