Gemma-4-It fine-tuned on PubMedQA using SFT & RLVR
-
Updated
Apr 27, 2026 - Python
Gemma-4-It fine-tuned on PubMedQA using SFT & RLVR
Open-source clinical-evidence RAG with grounded citations and calibrated abstention, evaluated on MIRAGE. Measured trustworthiness: risk-coverage, calibration, citation faithfulness. Research and evaluation only, not for clinical use.
Fine-tune language models to cite sources and quote evidence using RAFT (Retrieval-Augmented Fine-Tuning) on biomedical QA datasets (PubMedQA, BioASQ).
Archived fork — canonical version moved to vardhjain/graphrag-pubmedqa-ablation.
Independent multi-agent framework for verifying AI-generated medical diagnoses using specialized verifier agents and persistent memory.
Fine-tuned BiomedBERT on PubMedQA for 3-class biomedical question answering (yes/no/maybe) · Streamlit demo · HuggingFace Hub
Retrieval-augmented QA over biomedical literature: a reusable pipeline (ingest, retrieve, generate, evaluate) with BM25 and dense FAISS retrieval, grounded LLM answers, REST and GraphQL ingestion, and baseline-benchmarked evaluation on PubMedQA.
GraphRAG vs PlainRAG ablation on PubMedQA. Parent expansion wins +22.5pp (McNemar p<0.0001). Leakage-free, 4-arm design.
Fine-tune TinyLlama, Phi-2, and Mistral on PubMedQA using LoRA/QLoRA — side-by-side base vs fine-tuned comparison with Streamlit UI
Add a description, image, and links to the pubmedqa topic page so that developers can more easily learn about it.
To associate your repository with the pubmedqa topic, visit your repo's landing page and select "manage topics."