v000000/L3-8B-Stheno-v3.2-abliterated

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Jul 9, 2024Architecture:Transformer0.0K Cold

The v000000/L3-8B-Stheno-v3.2-abliterated model is an 8 billion parameter language model, a linear merge of Sao10K/L3-8B-Stheno-v3.2 and grimjim/Llama-3-Instruct-abliteration-LoRA-8B. It is designed with an 8192 token context length and incorporates 'abliteration' adaptation. This model is primarily intended for tasks benefiting from the combined characteristics of its merged components, offering a unique blend of capabilities.

Loading preview...

Model Overview

v000000/L3-8B-Stheno-v3.2-abliterated, also referred to as "Llama3-Stheno-v3.2-Zeta," is an 8 billion parameter language model. It was created through a linear merge using mergekit, combining the base model Sao10K/L3-8B-Stheno-v3.2 with an 'abliteration' adaptation from grimjim/Llama-3-Instruct-abliteration-LoRA-8B.

Key Characteristics

  • Architecture: A merged model based on the Llama 3 family, incorporating specific adaptations.
  • Parameter Count: 8 billion parameters.
  • Context Length: Supports an 8192 token context window.
  • Abliteration Adaptation: Includes a tested and successful 'abliteration' adaptation, suggesting potential for specific instruction-following or content modification tasks.

Use Cases

This model is suitable for developers exploring the effects of model merging and 'abliteration' techniques. It can be used for:

  • Experimental AI Development: Ideal for testing and understanding the impact of specific LoRA adaptations on a base model.
  • Instruction-Following Tasks: Potentially enhanced by the 'abliteration' component for nuanced instruction adherence.
  • Research: Provides a platform for investigating merged model performance and the utility of abliteration in language models.

Quantized versions (GGUF and GGUF imatrix) are available from mradermacher for efficient deployment.