d2hshsj/Qwen2.5-7B-Instruct-abliterated-v2
The d2hshsj/Qwen2.5-7B-Instruct-abliterated-v2 is a 7.6 billion parameter instruction-tuned causal language model, derived from Qwen/Qwen2.5-7B-Instruct. This version has been 'abliterated' to be uncensored, offering an improved iteration over its predecessor. It maintains a 32K context length and is primarily designed for applications requiring an instruction-following model with reduced content moderation, as indicated by its improved IF_Eval score.
Loading preview...
Overview of Qwen2.5-7B-Instruct-abliterated-v2
This model, developed by d2hshsj, is an uncensored variant of the Qwen/Qwen2.5-7B-Instruct, a 7.6 billion parameter instruction-tuned language model. It leverages an 'abliteration' technique to modify its content moderation, making it suitable for use cases where less restrictive output is desired. This version is an improvement over the previous Qwen2.5-7B-Instruct-abliterated.
Key Capabilities
- Uncensored Output: Modified to provide less restricted responses compared to its base model, making it suitable for diverse applications.
- Instruction Following: Retains strong instruction-following capabilities from the Qwen2.5-7B-Instruct base.
- Improved IF_Eval Score: Achieves a higher IF_Eval score (77.82) compared to the original Qwen2.5-7B-Instruct (76.44) and its previous abliterated version (76.49), indicating enhanced performance in specific areas.
- Standard Qwen Features: Inherits the robust architecture and 32K context length of the Qwen2.5-7B-Instruct family.
Good For
- Research into Uncensored Models: Ideal for researchers exploring the behavior and capabilities of models with reduced content filters.
- Applications Requiring Less Moderation: Suitable for use cases where the base model's inherent content moderation might be too restrictive.
- Developers Familiar with Qwen: Easy integration for those already working with Qwen models due to its base architecture and
transformerslibrary compatibility.