Meissa-SFT: 25,018 Medical Agentic and Multi-Modal Training Samples

Meissa-SFT consists of 25,018 supervised fine-tuning samples for medical multi-modal LLMs, published by CYX1998 in March 2026. The collection focuses on radiology, pathology, and clinical reasoning to support agentic capabilities such as tool-calling and multi-step simulation.

Use Cases

• Fine-tuning medical LLMs for visual question answering using radiology and pathology images
• Developing agentic clinical-simulation models that utilize tool-calling features
• Training models on multi-modal diagnostic reasoning sequences

Strengths

• 25,018 curated medical samples
• Multi-modal support for radiology and pathology images
• Includes agentic features like tool-calling and clinical simulation labels

Limitations

• Restricted to English language
• Relatively small sample size of 25,018 records for multi-modal training
• Domain concentration on radiology and pathology may limit general clinical utility

Provenance

Source

CYX1998 (Hugging Face), associated with Arxiv 260309018

Freshness

Last updated March 2026.

Geography

United States