Ryze: Evidence-Enriched Data Synthesis from Biomedical Papers

2026-05-30

AI Summary

This work addresses the limited reliability of general-purpose vision-language models in biomedicine, stemming from their inability to effectively integrate multimodal evidence scattered across figures, tables, captions, and main text. The authors propose Ryze, a system that, for the first time, enables fully automatic construction of question-answering training data while preserving the complete evidential structure. Ryze combines layout-aware analysis of figures and tables with OCR error correction and large language model–driven data cleaning, and introduces a progress-gated progressive post-training strategy that synergistically integrates supervised fine-tuning and reinforcement learning. Built upon Qwen3-VL-8B, the resulting BioVLM-8B model was trained at a cost under \$200 and achieves a weighted accuracy of 48.0% on LAB-Bench—surpassing the baseline by 12.6 percentage points and outperforming GPT-5.2 by 3.8 points.

Abstract

General-purpose VLMs remain unreliable for biomedical research because valid answers in scientific papers depend on evidence split across figures, tables, charts, captions, and referring text. Existing post-training pipelines are bottlenecked by costly expert annotation and by synthetic data that drops this evidence structure. We present Ryze, a fully automated system that converts raw biomedical papers into an evidence-enriched training set and a domain-specialized VLM. Ryze synthesizes QA pairs with complete supporting evidence (visual element, caption, extracted structure, and referring paragraphs), reduces layout and OCR errors via chart/table-aware extraction and LLM-based cleansing, and applies a progress-gated post-training strategy combining supervised fine-tuning with reinforcement learning. Starting from Qwen3-VL-8B, Ryze produces BioVLM-8B at under USD 200, achieving 48.0% weighted accuracy on LAB-Bench, outperforming the base model by +12.6 percentage points (pp) and surpassing GPT-5.2 by +3.8 pp. We release Ryze as open source together with the trained BioVLM-8B model.

Problem

• biomedical research
• visual language models
• evidence structure
• data synthesis
• scientific papers

Innovation

• evidence-enriched data synthesis
• biomedical VLM
• chart/table-aware extraction
• progress-gated post-training
• automated QA generation