self-play-reward-model

Self Play Reward Model

Scores self-play traces and policy outputs for downstream training loops.

Model Details

• Artifact type: full fine-tuned model
• Base model: allenai/scibert_scivocab_uncased
• Backbone type: encoder
• Model ID: S2
• Tier: T7_selfplay
• Role in stack: specialized Repository Library component

This model is part of the Repository Library stack, a research system for indexing, retrieving, aligning, and reasoning over scientific papers, structured paper content, repositories, and cross-domain links between them.

Model Sources

• Hugging Face repo: https://huggingface.co/PeytonT/self-play-reward-model
• Hugging Face collection: https://huggingface.co/collections/PeytonT/research-library-6a49c589ef4d763f7539b50d
• GitHub repository: https://github.com/peytontolbert/research_library
• Experiment config: https://github.com/peytontolbert/research_library/blob/main/models/experiments/s2_self_play_reward_model.json
• Models directory: https://github.com/peytontolbert/research_library/tree/main/models

Intended Use

• Primary use: Scores self-play traces and policy outputs for downstream training loops.
• Downstream use: retrieval, ranking, planning, paper understanding, or cross-domain reasoning inside the broader Repository Library system, depending on the model family.
• Out of scope: production safety claims, benchmark claims beyond the tracked experiment config, or deployment without task-specific validation.

Training Data

The training inputs for this package were assembled from the following Repository Library data sources:

• Source generated_selfplay_traces: generated self-play traces collected for reward-model supervision.

Training Procedure

• Sources: generated_selfplay_traces
• Input fields: task, trajectory, outputs, tests
• Target fields: reward
• Train/val/test split: [0.9, 0.1, 0.0]
• Max samples: 4000
• Batch size: 8
• Precision: bf16
• Objective: cross_entropy
• Learning rate: 5e-05
• Max source tokens: 512
• Max target tokens: 128
• Fine-tune strategy: full_finetune
• Max steps: 1000

Compute

• Hardware: 4x RTX_3090 (24 GB)
• Distributed strategy: ddp
• Estimated GPU hours in config: 0

Evaluation

• Declared metrics: accuracy, macro_f1
• Status: this card reflects the current tracked experiment configuration and packaged weights in the Repository Library model stack.

Usage

from transformers import AutoModelForSequenceClassification, AutoTokenizer

repo_id = "PeytonT/self-play-reward-model"

tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForSequenceClassification.from_pretrained(repo_id)

Limitations

• These cards are generated from tracked experiment metadata and packaged artifacts, not from a separate benchmark report or external audit.
• Several training sources are pipeline outputs from the Repository Library codebase rather than standalone public datasets.
• These models are components of a larger research system and should be validated in their target workflow before deployment.

Project Context

• GitHub repository: https://github.com/peytontolbert/research_library
• Model collection: https://huggingface.co/collections/PeytonT/research-library-6a49c589ef4d763f7539b50d
• Publisher: PeytonT