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