zannvznn/qwen3-0.6b-math-l45-qlora-merged-fp16-v2

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:0.8BQuant:BF16Ctx Length:32kPublished:May 17, 2026Architecture:Transformer Warm

The zannvznn/qwen3-0.6b-math-l45-qlora-merged-fp16-v2 model is a 0.8 billion parameter Qwen3-0.6B-Base variant, fine-tuned using QLoRA for solving mathematical problems at levels 4-5. It is specifically optimized for precise mathematical reasoning and outputting solutions in LaTeX format. This model is designed for applications requiring accurate step-by-step mathematical problem-solving capabilities.

Loading preview...

Model Overview

This model, zannvznn/qwen3-0.6b-math-l45-qlora-merged-fp16-v2, is a specialized version of the Qwen/Qwen3-0.6B-Base model, enhanced through QLoRA fine-tuning. It focuses on mathematical problem-solving, specifically targeting problems at difficulty levels 4 and 5.

Key Capabilities

  • Mathematical Problem Solving: Trained to solve complex math problems with step-by-step reasoning.
  • LaTeX Output: Generates mathematical expressions and solutions using valid LaTeX notation, including commands like \frac, \sqrt, and superscripts.
  • Precise Output Contract: Adheres to a strict output format, ensuring the final answer is clearly marked with Final Answer: \boxed{...}.
  • Efficient Training: Utilized a QLoRA SFT adapter, trained on 3994 rows from a math dataset, with a max sequence length of 2048 and 3 epochs.

Use Cases

This model is particularly well-suited for applications requiring:

  • Automated generation of detailed mathematical solutions.
  • Educational tools that provide step-by-step explanations for advanced math problems.
  • Systems needing to parse and generate mathematical content in LaTeX format.

It is important to note that this is a corrected SFT v2 run, with evaluation intentionally kept separate from the training process.