wvnvwn/qwen-2.5-7B-SSFT-gsm8k-lr3e-5

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Apr 28, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

This is a 7.6 billion parameter language model, wvnvwn/qwen-2.5-7B-SSFT-gsm8k-lr3e-5, based on the Llama-3.2-3B-Instruct architecture. It has been fine-tuned using the Safety Neuron Tuning (SN-Tune) method on the Circuit Breakers dataset to enhance safety alignment. This selective fine-tuning approach focuses on modifying only specific 'safety neurons' while preserving general capabilities. Its primary strength lies in improved safety performance with minimal impact on its original instruction-following abilities.

Loading preview...

Model Overview

This model, wvnvwn/qwen-2.5-7B-SSFT-gsm8k-lr3e-5, is a 7.6 billion parameter language model derived from the meta-llama/Llama-3.2-3B-Instruct base. Its key differentiator is the application of Safety Neuron Tuning (SN-Tune), a specialized fine-tuning method designed to enhance safety alignment.

Key Capabilities & Features

  • Enhanced Safety Alignment: Fine-tuned specifically on the Circuit Breakers dataset to improve safety responses.
  • Parameter-Efficient Fine-tuning: SN-Tune selectively modifies only a small subset of 'safety neurons', freezing all other parameters. This ensures efficient training and minimizes the risk of degrading general model capabilities.
  • Base Model Preservation: Aims to retain the core instruction-following and general language understanding abilities of the original Llama-3.2-3B-Instruct model.

When to Use This Model

This model is particularly well-suited for applications where:

  • Safety is a paramount concern: Ideal for use cases requiring robust safety guardrails and reduced generation of harmful content.
  • Maintaining base model performance is crucial: When you need the capabilities of Llama-3.2-3B-Instruct but with an added layer of safety without extensive retraining.
  • Resource-efficient safety improvements are desired: The SN-Tune method offers a targeted approach to safety enhancement without the computational overhead of full model fine-tuning.