ahmedalkhair64/qwen2.5-14b-medical-sft

TEXT GENERATIONConcurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 30, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The ahmedalkhair64/qwen2.5-14b-medical-sft is a 14.8 billion parameter language model, fine-tuned from Qwen2.5-14B-Instruct using LoRA. It is specifically optimized for complex medical reasoning and multi-step diagnostic scenarios across various clinical specialties. This model excels at generating evidence-based responses for medical case studies, leveraging its training on the English subset of the medical-o1-reasoning-SFT dataset.

Loading preview...

Model Overview

This model, ahmedalkhair64/qwen2.5-14b-medical-sft, is a 14.8 billion parameter language model fine-tuned from the Qwen2.5-14B-Instruct base model. It was developed using LoRA (Low-Rank Adaptation) with a rank of 64, specifically targeting enhanced medical reasoning capabilities. The training utilized the English subset of the FreedomIntelligence/medical-o1-reasoning-SFT dataset, which comprises medical case studies with detailed reasoning chains.

Key Capabilities

  • Enhanced Medical Reasoning: Designed for complex diagnostic scenarios and multi-step clinical reasoning.
  • Specialty-Specific Knowledge: Covers a broad range of medical specialties including cardiology, neurology, pulmonology, endocrinology, gastroenterology, infectious disease, nephrology, and emergency medicine.
  • Evidence-Based Responses: Generates responses based on detailed reasoning chains from medical case studies.

Usage Considerations

  • Recommended Inference: Best used with llama.cpp in GGUF format.
  • Temperature Sensitivity: Crucially, a temperature of 0.3 is recommended for accurate medical reasoning; higher temperatures (0.7+) can lead to hallucinations and misdiagnoses.
  • Language Limitation: Trained exclusively on English medical data; performance in other languages is not guaranteed.
  • Ollama Incompatibility: This model is not compatible with Ollama due to differing quantization formats.
  • Context Length: Optimized for a context length of 4096 tokens, though some complex cases might benefit from longer contexts.