modrill/math-think-s1-v2-qwen3-4b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 13, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The modrill/math-think-s1-v2-qwen3-4b is a 4 billion parameter Qwen3-based language model developed by modrill. This model is specifically fine-tuned for mathematical reasoning and 'think' style supervised fine-tuning (SFT) tasks, utilizing a 24576 token cutoff during its training. It is optimized for processing and generating content related to mathematical problem-solving and logical thought processes.

Loading preview...

Overview

The modrill/math-think-s1-v2-qwen3-4b is a 4 billion parameter language model built on the Qwen3 architecture. Developed by modrill, this model has undergone a specialized 'think' style supervised fine-tuning (SFT) process, designated as 'Light-R1 stage1'.

Key Characteristics

  • Architecture: Qwen3 base model.
  • Parameter Count: 4 billion parameters.
  • Fine-tuning Focus: Specialized in 'think' style SFT, indicating an optimization for reasoning and problem-solving tasks, particularly in mathematics.
  • Training Details: Fine-tuned for 2 epochs with a learning rate of 1e-5, utilizing packing during training.
  • Context Length: Trained with a cutoff of 24576 tokens, suggesting a robust capability for handling moderately long contexts relevant to complex reasoning.

Ideal Use Cases

  • Mathematical Reasoning: Excellent for applications requiring step-by-step mathematical problem-solving and logical deduction.
  • Educational Tools: Can be integrated into tools for teaching or assisting with math-related queries.
  • Research in AI Reasoning: Suitable for researchers exploring the capabilities of LLMs in complex thought processes and mathematical understanding.