kmseong/llama3.2-3b-instruct-safety-FT-lr1e-6

TEXT GENERATIONConcurrency Cost:1Model Size:3.2BQuant:BF16Ctx Length:32kPublished:Apr 11, 2026License:llama3.2Architecture:Transformer Cold

The kmseong/llama3.2-3b-instruct-safety-FT-lr1e-6 model is a 3.2 billion parameter instruction-tuned language model, likely based on the Llama architecture, with a context length of 32768 tokens. It incorporates perlayer application and non-freeze training, focusing on safety alignment through a weight space rotation process. This model is optimized for tasks requiring safety-aligned responses, making it suitable for applications where controlled and ethical AI output is critical.

Loading preview...

Model Overview

The kmseong/llama3.2-3b-instruct-safety-FT-lr1e-6 is a 3.2 billion parameter instruction-tuned language model, featuring a substantial context length of 32768 tokens. This model is distinguished by its training methodology, which includes the application of perlayer techniques and non-freeze training. A core aspect of its development is the focus on safety alignment, achieved through a "Weight space Rotation Process" (Warp).

Key Capabilities

  • Safety Alignment: The model is specifically fine-tuned for safety, utilizing a novel "Weight space Rotation Process" to guide its responses.
  • Instruction Following: Designed to accurately follow instructions, making it suitable for various prompt-based tasks.
  • Extended Context: With a 32768-token context window, it can process and generate longer, more coherent texts while maintaining conversational history or complex document understanding.
  • Efficient Training: Employs perlayer application and non-freeze training, suggesting an approach to optimize performance and adaptability.

Good For

  • Applications requiring robust safety features and ethical AI responses.
  • Use cases where instruction-following is paramount.
  • Tasks benefiting from a large context window, such as summarization of long documents or extended dialogue generation.
  • Research into safety alignment techniques and weight space manipulation in large language models.