Toivo0/gemma-3-finetune

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Jul 21, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Toivo0/gemma-3-finetune is a 1.1 billion parameter Gemma-3-1B-IT model fine-tuned by Toivo0 for Turkish statistical question-answering. This variant is specifically trained to provide direct, concise answers without intermediate reasoning steps. It excels in responding to queries across seven modules of statistics in Turkish, making it suitable for educational or quick reference applications in this domain.

Loading preview...

Overview

This model, Toivo0/gemma-3-finetune, is a specialized variant of Google's Gemma-3-1B-IT, fine-tuned by Toivo0 to answer questions on statistical topics in Turkish. Unlike its counterpart, which shows reasoning steps, this model is designed to provide direct and concise final explanations without displaying intermediate thought processes.

Key Capabilities

  • Turkish Statistics Q&A: Proficient in answering questions across 7 modules of statistics, including Descriptive Statistics, Probability, Hypothesis Testing, and Regression.
  • Direct Answers: Specifically trained to omit internal reasoning (<think>...</think>) and deliver immediate, final answers.
  • Language: Fully focused on the Turkish language.
  • Base Model: Built upon unsloth/gemma-3-1b-it-unsloth-bnb-4bit.

Training and Data

The model was fine-tuned using a custom dataset of 400 Turkish question-answer pairs covering various statistical concepts. The training utilized Unsloth and HuggingFace TRL libraries, focusing solely on the final explanation part of the dataset.

Performance Insights

While fine-tuned for a specific domain, a comparative Turkish MMLU benchmark showed a slight -0.87% decrease in overall performance compared to the base model. This is an expected trade-off for domain-specific fine-tuning on a smaller 1B parameter model. However, it showed improvements in specific general knowledge categories like "Tourism and Travel Services" (+12.00%) and "Driving License Exam" (+7.00%), while declining in others like "Private Protection and Security" (-11.00%).

Use Cases

  • Educational Tools: Ideal for applications requiring quick, direct answers to Turkish statistics questions.
  • Reference Systems: Can serve as a rapid information retrieval system for statistical concepts in Turkish.

Finetunes
2 models