andresnowak/Qwen3-0.6B-MNLP_mcqa_model_text

Hugging Face
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 27, 2025Architecture:Transformer Featherless Exclusive Warm

The andresnowak/Qwen3-0.6B-MNLP_mcqa_model_text is a 0.8 billion parameter Qwen3-based language model fine-tuned by andresnowak. It is specifically optimized for multiple-choice question answering (MCQA) tasks, trained using a Seq2Seq method with TRL. This model excels at providing direct answers to MCQA problems across various domains, including STEM and general knowledge. It features a context length of 32768 tokens, making it suitable for detailed question analysis.

Loading preview...

Model Overview

The andresnowak/Qwen3-0.6B-MNLP_mcqa_model_text is a specialized 0.8 billion parameter language model built upon the Qwen3-0.6B-Base architecture. Developed by andresnowak, this model has undergone supervised fine-tuning (SFT) using the TRL library, specifically targeting Multiple Choice Question Answering (MCQA) tasks. Its training methodology involves a Seq2Seq approach, where the model is trained to generate answers in a "Letter. Answer" format.

Key Capabilities

  • Multiple Choice Question Answering (MCQA): The model is explicitly fine-tuned for MCQA, making it proficient in selecting correct answers from a given set of options.
  • Broad Domain Coverage: Training datasets include a diverse range of MCQA benchmarks such as Math QA, ScienceQA, MMLU, AI2 ARC (Challenge and Easy), MedMCQA, OpenBookQA, and SciQ.
  • Language Modelling: The training process included language modeling with loss calculated on both the prompt and completion, enhancing its generative capabilities within the MCQA context.
  • Context Length: Supports a substantial context length of 32768 tokens, allowing for comprehensive question and option processing.

Performance Highlights

Evaluated across several MCQA benchmarks, the model demonstrates varying levels of accuracy:

  • ARC Challenge: 60.6% Accuracy
  • ARC Easy: 78.6% Accuracy
  • MMLU: 48.6% Accuracy
  • Overall Accuracy: 43.9%

Good for

  • Automated MCQA Systems: Ideal for applications requiring automated answering of multiple-choice questions.
  • Educational Tools: Can be integrated into platforms for quizzes, assessments, or study aids.
  • Research in MCQA: Provides a fine-tuned base for further research and development in multiple-choice question answering models, particularly for its size and specialized training.