Rumiii/Qwen3-8B-MedReasonPath

Overview

Rumiii/Qwen3-8B-MedReasonPath is an 8.2 billion parameter model built upon the Qwen/Qwen3-8B base. It has been specifically fine-tuned to improve step-by-step medical reasoning, particularly within Qwen3's native <think>...</think> mode. A key focus during its development was to enhance clinical reasoning quality while preserving the base model's inherent thinking mode, non-thinking mode, and agentic tool-calling functionalities.

Key Capabilities

• Enhanced Medical Reasoning: Demonstrates improved step-by-step reasoning for medical and clinical questions.
• Retained Agentic Behavior: Maintains the base Qwen3-8B's ability for tool-calling, allowing for integration with external functions.
• Flexible Reasoning Modes: Supports both detailed, step-by-step reasoning via enable_thinking=True and direct answers with enable_thinking=False.
• Standalone Model: Provided as a full fp16 checkpoint, requiring no additional libraries like Unsloth for inference, only transformers.

Training Details

The model was fine-tuned using QLoRA (4-bit, rank 16, alpha 32) on a single T4 GPU. The training incorporated two key datasets:

• MedReason (5,000 examples): Focused on knowledge-graph grounded medical reasoning to boost clinical reasoning quality.
• xLAM Function-Calling (681 examples): Included to help preserve and reinforce general tool-calling capabilities.

Good For

• Applications requiring detailed, explainable medical reasoning.
• Scenarios where a language model needs to interact with external tools in a medical context.
• Research and educational projects exploring advanced medical AI capabilities. (Note: Not validated for clinical use.)