yethdev/qwythos-9b-v2-manumit-v1
The yethdev/qwythos-9b-v2-manumit-v1 is a 9 billion parameter causal language model, based on the Qwythos-9B-v2 architecture, developed by yethdev. This model has been modified using the 'manumit v1' process to significantly reduce refusal rates, making it less prone to denying requests. It demonstrates an improved MMLU score of 66.7% compared to its base model, indicating enhanced general knowledge and reasoning capabilities. This model is primarily designed for applications requiring a less restrictive and more direct response generation, particularly in English.
Loading preview...
Model Overview
The qwythos-9b-v2-manumit-v1 is a 9 billion parameter language model developed by yethdev, derived from the empero-ai/Qwythos-9B-v2 base model. Its core distinction lies in its "abliteration" via the manumit v1 tool, which aims to remove or reduce built-in safeguards that typically lead to request denials.
Key Differentiators
- Reduced Refusal Rate: The
manumit v1process has significantly lowered the model's refusal rate from 54.2% (base model) to 85.4%, indicating a much higher propensity to generate responses without denial. - Improved MMLU Performance: Alongside reduced censorship, the model also shows an increase in its MMLU (Massive Multitask Language Understanding) score, rising from 60.4% in the base model to 66.7%. This suggests an enhancement in general knowledge and reasoning abilities.
Use Cases
This model is particularly suited for applications where a less restrictive and more direct response generation is desired. Developers seeking a model that is less likely to refuse certain prompts, while also offering improved general understanding, may find this model beneficial. It is primarily designed for English language tasks.