GaborMadarasz/gemma_3_270m_HuHotPotQA_16bit

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.3BQuant:BF16Context Size:32kPublished:Aug 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

GaborMadarasz/gemma_3_270m_HuHotPotQA_16bit is a 0.3 billion parameter Gemma 3 model developed by GaborMadarasz. This model was fine-tuned using Unsloth and Huggingface's TRL library, achieving faster training times. It is specifically optimized for question answering tasks, demonstrating an exact match score of 61.9469 and a token F1 score of 66.5318 on its evaluation dataset. Its compact size and specialized fine-tuning make it suitable for efficient deployment in QA applications.

Loading preview...

Model Overview

GaborMadarasz/gemma_3_270m_HuHotPotQA_16bit is a compact 0.3 billion parameter Gemma 3 model, developed by GaborMadarasz. It was fine-tuned from unsloth/gemma-3-270m-it using the Unsloth library in conjunction with Huggingface's TRL library, which enabled a 2x faster training process.

Key Capabilities

  • Efficient Fine-tuning: Leverages Unsloth for accelerated training, making it resource-efficient.
  • Question Answering Focus: Specifically fine-tuned for question answering tasks.
  • Performance Metrics: Achieved the following evaluation scores:
    • Exact Match: 61.9469
    • Token F1: 66.5318
    • Rouge-L: 66.0725

Good For

  • Resource-constrained QA: Its small parameter count and efficient training make it suitable for environments with limited computational resources.
  • Specific QA Applications: Ideal for use cases requiring a focused question answering model with demonstrated performance on its evaluation set.
  • Fast Deployment: The optimized training process suggests potential for quicker iteration and deployment in projects.