LLM Fine-Tuning
Clinical Intelligence
AI Tool Use
Domain-Specific Post-Training Closes the Execution Gap in Clinical Agents
Using only 1,530 training examples, a single QLoRA SFT pass lifts Qwen3-32B from 57.0% to 74.0% on MedAgentBench — surpassing Claude 3.5 Sonnet v2. The gains come from protocol adherence, not new capabilities.
Apr 27, 2026
ChartR Team