Anonymiserad/SweMedMCQ-LFM2.5-1.2B-Trace-SV-RW
Anonymiserad/SweMedMCQ-LFM2.5-1.2B-Trace-SV-RW is a 1.2 billion parameter language model, fine-tuned from LiquidAI/LFM2.5-1.2B-Instruct, specifically for English and Swedish medical multiple-choice questions. It utilizes reasoning-trace data with rewritten Swedish reasoning, making it particularly adept at generating medical explanations in Swedish. This model is designed for medical question answering, offering improved accuracy over its base model in both English and Swedish medical contexts, and is part of a research effort to benchmark and adapt LLMs for medical QA.
Loading preview...
Overview
Anonymiserad/SweMedMCQ-LFM2.5-1.2B-Trace-SV-RW is a 1.2 billion parameter language model, fine-tuned from the LiquidAI/LFM2.5-1.2B-Instruct base model. It is specifically adapted for medical multiple-choice question answering in both English and Swedish. A key differentiator is its training on reasoning-trace data where Swedish reasoning explanations were rewritten, enabling it to generate structured medical explanations in Swedish.
Key Capabilities
- Medical Question Answering: Specialized in answering multiple-choice medical questions.
- Multilingual Support: Proficient in both English and Swedish medical contexts.
- Reasoning Trace Generation: Capable of producing detailed reasoning blocks (
<think>...</think>) before providing a JSON-formatted answer, with rewritten Swedish reasoning. - Improved Accuracy: Demonstrates enhanced accuracy on the SweMedMCQ test set compared to its base model, achieving 44.3% in Swedish and 44.0% in English for reasoning-based decoding.
Good For
- Research in Medical AI: Ideal for researchers studying LLM performance in medical question answering, especially in a multilingual context.
- Developing Medical Education Tools: Can be used as a component in systems designed to generate explanations for medical questions.
- Benchmarking: Serves as a student model within the SweMedMCQ benchmark for evaluating LLMs on English-Swedish medical QA.
Limitations
This is a research model and not intended for clinical use, diagnosis, or treatment. Its accuracy, while improved, is still below that of larger models, and explanations may contain inaccuracies. The Swedish training and test data are machine-translated, which may introduce errors.