lexuanbach/relationalrag-tables-model-sources

RelationalRAG-Tables Model Sources

This repository is a public model/source manifest for the RelationalRAG-Tables artifact. The artifact does not contain local trained checkpoint files; the learned verifier is trained by the artifact scripts, and the real-data layer uses public upstream models.

Upstream Models Used

Component	Identifier / Link	Role
Sentence embedding	BAAI/bge-small-en-v1.5	Value/entity embedding features
Cross-encoder	cross-encoder/ms-marco-MiniLM-L-6-v2	Strong reranker / entity-matching baseline
SQL coder, 7B	Qwen/Qwen2.5-Coder-7B-Instruct	Text-to-SQL execution runs; artifact uses a local 4-bit MLX deployment
SQL coder, 14B	Qwen/Qwen2.5-Coder-14B-Instruct	Scale check; artifact uses a local 4-bit MLX deployment

Artifact Data

The public fixture/result payloads are available at:

https://huggingface.co/datasets/lexuanbach/relationalrag-tables-artifact-data

Note On Checkpoints

No additional trained neural weights are uploaded here because the artifact does not ship persistent verifier checkpoints. The small MLP/ensemble verifier is retrained from deterministic fixtures by the reproduction scripts.