Ephraimmm/medgemma-soap-finetuned1

VISIONConcurrency Cost:1Model Size:4.3BQuant:BF16Ctx Length:32kPublished:Apr 16, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Ephraimmm/medgemma-soap-finetuned1 is a 4.3 billion parameter language model, finetuned by Ephraimmm from google/medgemma-1.5-4b-it. This model was trained using Unsloth and Huggingface's TRL library, employing 4-bit QLoRA quantization. It is optimized for specific applications, demonstrating a final evaluation loss of 1.0012.

Loading preview...

Ephraimmm/medgemma-soap-finetuned1 Overview

This model, developed by Ephraimmm, is a 4.3 billion parameter language model finetuned from the google/medgemma-1.5-4b-it base model. It leverages the gemma3 architecture and was trained using a combination of Unsloth and Huggingface's TRL library, which facilitated a 2x faster training process.

Key Training Details

  • Base Model: google/medgemma-1.5-4b-it
  • Frameworks: Unsloth + TRL
  • Quantization: 4-bit QLoRA
  • LoRA Configuration: Rank 16, Alpha 32
  • Training Epochs: 3
  • Learning Rate: 2e-4 with Cosine scheduler
  • Final Evaluation Loss: 1.0012

Potential Use Cases

Given its finetuning from a medgemma base and specific training parameters, this model is likely suitable for:

  • Applications requiring a compact yet capable language model.
  • Scenarios where efficient inference with 4-bit QLoRA quantization is beneficial.
  • Tasks that align with the original medgemma model's strengths, enhanced by the finetuning process.