Anonymiserad/SweMedMCQ-LFM2.5-1.2B-Answer-SV

Overview

Anonymiserad/SweMedMCQ-LFM2.5-1.2B-Answer-SV is a 1.2 billion parameter model, fine-tuned from LiquidAI/LFM2.5-1.2B-Instruct, designed for medical multiple-choice question answering. Its unique characteristic is its training on answer-level data in Swedish, where the model learns to output only the correct answer index (e.g., {"answer_index": "B"}). This model is part of a research effort detailed in the paper "Benchmarking and Adapting LLMs for English–Swedish Medical Question Answering."

Key Capabilities

• Specialized Medical QA: Optimized for answering medical multiple-choice questions.
• Swedish Language Focus: Trained exclusively on Swedish medical questions, machine-translated from English.
• Structured Output: Designed to provide answers in a specific JSON format, indicating the correct option letter.
• Small Footprint: At 1.2 billion parameters, it offers a compact solution for its specialized task.

Good For

• Research in Medical NLP: Particularly for studies involving Swedish medical question answering and structured output generation.
• Benchmarking: Useful for evaluating performance on the SweMedMCQ benchmark, especially for answer-only prediction.
• Developing Prototype QA Systems: Can serve as a component for systems requiring direct, letter-based answers to medical MCQs in Swedish.

Limitations

• This is a research model and not intended for clinical use, diagnosis, or treatment decisions.
• Trained and evaluated solely on exam-style multiple-choice questions; its accuracy is lower than larger models, and explanations (if generated) may be incorrect.
• Relies on machine-translated Swedish data, which may introduce translation errors.