Enhance model card with repository context and sources

README.md

@@ -19,44 +19,60 @@ Reranks retrieved candidates with a cross-encoder scoring pass.
 
 - Artifact type: full fine-tuned model
 - Base model: `sentence-transformers/all-MiniLM-L6-v2`
+- Backbone type: `encoder`
 - Model ID: `L2`
 - Tier: `repository_library_search_stack`
-- Local mirror: `/arxiv/models/repository_library/cross-encoder-reranker`
-- Source checkpoint: `models/checkpoints/L2`
+- Role in stack: search-stack component for retrieval, reranking, or routing
 
-This repository is part of the `repository_library` model stack and is mirrored from `/data/repository_library/models/checkpoints` for publication under the `PeytonT` namespace.
+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/cross-encoder-reranker`
+- 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/l2_cross_encoder_reranker.json`
+- Models directory: `https://github.com/peytontolbert/research_library/tree/main/models`
 
 ## Intended Use
 
 - Primary use: Reranks retrieved candidates with a cross-encoder scoring pass.
-- Secondary use: retrieval, ranking, planning, or scientific paper tooling inside the broader Repository Library system, depending on the model family.
-- Out of scope: production safety claims, benchmark claims beyond the bundled experiment config, or use outside the model's narrow training objective without task-specific validation.
+- 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
 
-This package was trained from the following declared datasets or corpus sources:
+The training inputs for this package were assembled from the following Repository Library data sources:
 
-- `source:github_repos`
+- Source `github_repos`: repository graph and code chunk data exported from the Repository Library repo pipeline.
 
 ## Training Procedure
 
 - Sources: `github_repos`
 - Input fields: `query, candidate_row`
 - Target fields: `relevance_label`
+- Train/val/test split: `[0.9, 0.1, 0.0]`
 - Max samples: `4000`
+- Batch size: `8`
 - Precision: `bf16`
 - Objective: `cross_entropy`
-- Batch size: `8`
 - Learning rate: `5e-05`
 - Max source tokens: `256`
 - Max target tokens: `256`
+- Fine-tune strategy: `full_finetune`
 - Max steps: `1000`
 - Notes: Search-stack role added to match models.md coverage; dataset builder may need role-specific supervised labels before promotion.
 
+## Compute
+
+- Hardware: not specified
+- Distributed strategy: `unknown`
+- Estimated GPU hours in config: `unknown`
+
 ## Evaluation
 
 - Declared metrics: `accuracy`
-- Status: local experiment artifact mirrored for release; external benchmark reporting has not been standardized across the full model family yet.
+- Status: this card reflects the current tracked experiment configuration and packaged weights in the Repository Library model stack.
 
 ## Usage
 
@@ -71,14 +87,12 @@ model = AutoModelForSequenceClassification.from_pretrained(repo_id)
 
 ## Limitations
 
-- These model cards reflect the packaged experiment configs and mirrored checkpoint contents, not an independently audited benchmark sheet.
-- Some training datasets are local corpora or exported shards, so reproducibility may require access to the surrounding Repository Library data pipeline.
-- Models in this stack are narrow components of a larger paper-and-repository system and should be validated on downstream tasks before deployment.
+- 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
 
-Repository Library is a research system for indexing, retrieving, aligning, and reasoning over scientific papers, structured paper content, repositories, and cross-domain links between them.
-
-## Contact
-
-Published under `PeytonT` from the local `repository_library` build.
+- GitHub repository: `https://github.com/peytontolbert/research_library`
+- Model collection: `https://huggingface.co/collections/PeytonT/research-library-6a49c589ef4d763f7539b50d`
+- Publisher: `PeytonT`