Liontix/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Safetensor
Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Aug 29, 2025Architecture:Transformer0.0K Warm

Liontix/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Safetensor is a 4 billion parameter language model based on the Qwen3 architecture, fine-tuned by Liontix. It was distilled from a combination of Claude Sonnet 4 (non-reasoning) and Claude Sonnet 3.7 (reasoning) datasets, making it specialized for tasks requiring reasoning capabilities. This model leverages a 40960 token context length and is designed for applications benefiting from Claude-style instruction following and distilled reasoning patterns.

Loading preview...

Model Overview

Liontix/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Safetensor is a 4 billion parameter language model built upon the unsloth/Qwen3-4B-unsloth-bnb-4bit base model. This model is specifically designed to distill reasoning capabilities from advanced large language models.

Key Characteristics

  • Distilled Reasoning: The model was fine-tuned using a unique combination of datasets:
    • Liontix/claude-sonnet-4-100x (derived from Claude Sonnet 4, focusing on non-reasoning aspects)
    • reedmayhew/claude-3.7-sonnet-reasoning (derived from Claude Sonnet 3.7, specifically targeting reasoning capabilities)
  • Claude-Style Prompting: It utilizes Claude-style prompt formatting with <|im_start|> and <|im_end|> markers, including role tags, for instruction following.
  • Base Architecture: Leverages the Qwen3 architecture, known for its efficiency and performance.

Use Cases

This model is particularly well-suited for applications where:

  • Reasoning tasks are critical, benefiting from the distillation of Claude Sonnet 3.7's reasoning patterns.
  • Claude-style instruction following is preferred or required for consistent interaction.
  • A 4 billion parameter model with a substantial context length (40960 tokens) offers a balance between performance and computational efficiency.