byroneverson/Yi-1.5-34B-Chat-abliterated
byroneverson/Yi-1.5-34B-Chat-abliterated is a 34 billion parameter language model derived from the Yi-1.5-34B-Chat architecture. This model has undergone an 'abliteration' process, which modifies its original training to potentially alter its behavior or capabilities. It is designed for users interested in exploring the effects of such modifications on a large language model's performance and output. The primary use case involves experimentation with modified LLM architectures.
Loading preview...
Model Overview
byroneverson/Yi-1.5-34B-Chat-abliterated is a 34 billion parameter language model based on the Yi-1.5-34B-Chat architecture. This version has been specifically 'abliterated,' a process that involves targeted modifications to the model's original training or structure. The creator, byroneverson, invites users to request further abliterations of other models, indicating an experimental and community-driven approach to model development.
Key Characteristics
- Parameter Count: 34 billion parameters, offering substantial capacity for complex language tasks.
- Context Length: Supports a context window of 32768 tokens, enabling processing of lengthy inputs and generating coherent, extended outputs.
- Abliteration Process: The model's unique characteristic is its 'abliterated' state, suggesting a departure from its original Yi-1.5-34B-Chat behavior. Details of this process are available in the provided jupyter notebook.
Use Cases
This model is particularly suited for:
- Experimentation: Ideal for researchers and developers interested in understanding the impact of specific modifications on large language models.
- Customization: Users can explore how 'abliteration' alters model responses and adapt it for niche applications where standard models might not suffice.
- Comparative Analysis: Useful for comparing the performance and characteristics of an abliterated model against its original counterpart.