wanglab/bioreason-pro-sft

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Mar 20, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

wanglab/bioreason-pro-sft is a supervised fine-tuned (SFT) checkpoint of BioReason-Pro, a multimodal reasoning LLM developed by wanglab. Built on a Qwen3-4B backbone, this model integrates ESM3 protein embeddings, a GO graph encoder, and biological context like InterPro domains and STRING interactions. It is specifically designed for protein function prediction, generating structured reasoning traces and functional annotations.

Loading preview...

BioReason-Pro SFT Overview

BioReason-Pro SFT is a supervised fine-tuned (SFT) checkpoint of the BioReason-Pro model, developed by wanglab. This model is engineered for multimodal biological reasoning, specifically targeting protein function prediction.

Key Capabilities

  • Multimodal Integration: Combines ESM3 protein embeddings, a GO (Gene Ontology) graph encoder, and additional biological context such as InterPro domains and STRING interactions.
  • Structured Reasoning: Utilizes a Qwen3-4B backbone to generate detailed, structured reasoning traces alongside functional annotations.
  • Protein Function Prediction: Optimized to accurately predict and explain protein functions by leveraging diverse biological data sources.

Good For

  • Researchers and developers working on protein bioinformatics and functional genomics.
  • Applications requiring interpretable protein function predictions with supporting reasoning.
  • Tasks involving the integration of various biological data types for deeper insights into protein mechanisms.

For full setup instructions and local inference guides, refer to the BioReason-Pro GitHub repository. The model was trained using the wanglab/bioreason-pro-sft-reasoning-data dataset.