Zheng-Zong/AronaR1-SFT-stage1-v3

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Mar 17, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

AronaR1-SFT-stage1-v3 is a 7.6 billion parameter Qwen2-based instruction-tuned language model developed by Zheng-Zong, fine-tuned from unsloth/Qwen2.5-Math-7B-Instruct. This model leverages Unsloth and Huggingface's TRL library for accelerated training. With a 32768 token context length, it is designed for general language understanding and generation tasks, building upon its mathematical instruction-tuned base.

Loading preview...

Model Overview

The Zheng-Zong/AronaR1-SFT-stage1-v3 is a 7.6 billion parameter language model, fine-tuned from the unsloth/Qwen2.5-Math-7B-Instruct base model. Developed by Zheng-Zong, this model utilizes the Qwen2 architecture and has been trained with the assistance of Unsloth and Huggingface's TRL library, enabling faster training iterations.

Key Characteristics

  • Base Model: Fine-tuned from unsloth/Qwen2.5-Math-7B-Instruct, suggesting a foundation with strong mathematical reasoning capabilities.
  • Training Efficiency: Leverages Unsloth for 2x faster training, indicating an optimized development process.
  • Context Length: Supports a substantial context window of 32768 tokens, allowing for processing longer inputs and generating more coherent, extended outputs.
  • License: Released under the Apache-2.0 license, providing broad usage permissions.

Potential Use Cases

Given its foundation and instruction-tuned nature, AronaR1-SFT-stage1-v3 is suitable for a variety of general-purpose language tasks, potentially excelling in areas requiring:

  • Instruction following
  • Text generation and completion
  • Question answering
  • Tasks benefiting from a large context window