abstract-code-relevance-pairs

Abstract Code Relevance Pairs

Scores abstract and code pairs using the paired-training variant of the abstract-code-relevance model.

Model Details

• Artifact type: LoRA adapter
• Base model: meta-llama/Llama-3.2-1B
• Backbone type: decoder
• Model ID: A1
• Tier: T2_abstract
• Role in stack: abstract-layer component for fast paper interpretation

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/abstract-code-relevance-pairs
• 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/a1_abstract_code_relevance.json
• Models directory: https://github.com/peytontolbert/research_library/tree/main/models

Intended Use

• Primary use: Scores abstract and code pairs using the paired-training variant of the abstract-code-relevance model.
• 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 arxiv_metadata: arXiv metadata records spanning titles, abstracts, authors, and category labels.
• Source github_repos: repository graph and code chunk data exported from the Repository Library repo pipeline.

Training Procedure

• Sources: arxiv_metadata, github_repos
• Input fields: abstract
• Target fields: repo_relevance
• Train/val/test split: [0.8, 0.1, 0.1]
• Max samples: 0
• Batch size: 4
• Precision: bf16
• Objective: contrastive
• Learning rate: 0.0001
• Max source tokens: 2048
• Max target tokens: 512
• Fine-tune strategy: peft_lora
• Max steps: 1000

Compute

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

Evaluation

• Declared metrics: recall_at_10, ndcg_at_10
• Status: this card reflects the current tracked experiment configuration and packaged weights in the Repository Library model stack.

Usage

from transformers import AutoModel, AutoTokenizer
from peft import PeftModel

repo_id = "PeytonT/abstract-code-relevance-pairs"
base_id = "meta-llama/Llama-3.2-1B"

tokenizer = AutoTokenizer.from_pretrained(repo_id)
base = AutoModel.from_pretrained(base_id)
model = PeftModel.from_pretrained(base, 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