wvnvwn/qwen-2.5-7B-Instruct-SafeInstr-lr5e-5-lr5e-5-0.1

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 29, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

This is a 7.6 billion parameter instruction-tuned language model, based on Llama-3.2-3B-Instruct, developed by wvnvwn. It has been fine-tuned using the Safety Neuron Tuning (SN-Tune) method on a Circuit Breakers dataset to enhance safety alignment. This approach selectively fine-tunes only safety-critical neurons, preserving general capabilities while improving safety. It is primarily designed for applications requiring enhanced safety alignment in conversational AI.

Loading preview...

Overview

This model, wvnvwn/qwen-2.5-7B-Instruct-SafeInstr-lr5e-5-lr5e-5-0.1, is a 7.6 billion parameter instruction-tuned variant of the meta-llama/Llama-3.2-3B-Instruct base model. It has undergone a specialized fine-tuning process called SN-Tune (Safety Neuron Tuning) to significantly improve its safety alignment.

Key Capabilities & Features

  • Enhanced Safety Alignment: Achieved through SN-Tune, which focuses on fine-tuning specific "safety neurons" within the model.
  • Parameter-Efficient Fine-tuning: The SN-Tune method freezes most parameters and only fine-tunes a small, critical set of neurons, making the process highly efficient.
  • Preservation of General Capabilities: By selectively tuning, the model aims to enhance safety without negatively impacting its broader language understanding and generation abilities.
  • Base Model: Built upon the robust Llama-3.2-3B-Instruct architecture.

When to Use This Model

This model is particularly well-suited for use cases where:

  • Safety is paramount: Applications requiring a high degree of safety alignment in AI responses.
  • Minimizing harmful outputs: Scenarios where reducing the generation of unsafe or undesirable content is a primary concern.
  • Efficient safety integration: Developers looking for a model with built-in safety enhancements without extensive retraining.