Alelcv27/Qwen3-4B-INST-Math-v2

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:May 13, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

Alelcv27/Qwen3-4B-INST-Math-v2 is a 4 billion parameter Qwen3 instruction-tuned language model developed by Alelcv27, fine-tuned from unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit. This model was specifically optimized for mathematical tasks, leveraging Unsloth and Huggingface's TRL library for faster training. It offers a 32768 token context length, making it suitable for processing longer mathematical problems and related instructions.

Loading preview...

Alelcv27/Qwen3-4B-INST-Math-v2 Overview

Alelcv27/Qwen3-4B-INST-Math-v2 is a 4 billion parameter instruction-tuned model based on the Qwen3 architecture, developed by Alelcv27. It was fine-tuned from unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit with a specific focus on enhancing its capabilities in mathematical reasoning and problem-solving. The training process utilized Unsloth and Huggingface's TRL library, which enabled a 2x faster fine-tuning compared to standard methods.

Key Capabilities

  • Mathematical Proficiency: Specifically optimized for handling mathematical instructions and tasks.
  • Efficient Training: Benefits from Unsloth's accelerated training techniques, making it a product of efficient fine-tuning.
  • Qwen3 Architecture: Leverages the robust base of the Qwen3 model family.
  • Extended Context: Supports a 32768 token context length, allowing for more complex and multi-step mathematical problems.

Good For

  • Applications requiring strong mathematical understanding and problem-solving.
  • Tasks involving numerical reasoning, equations, and logical mathematical sequences.
  • Developers looking for a compact yet capable model for math-centric AI applications.