yethdev/mistral-7b-manumit-v1
The yethdev/mistral-7b-manumit-v1 is a 7 billion parameter language model based on Mistral-7B-Instruct-v0.3, featuring a 4096-token context length. Developed by yethdev, this model has been "abliterated" using the manumit v1 tool to significantly reduce refusal rates. It is specifically optimized for use cases requiring less restrictive content generation, demonstrating a refusal rate of 2.1% compared to the base model's 16.7%.
Loading preview...
Overview
yethdev/mistral-7b-manumit-v1 is a 7 billion parameter language model derived from mistralai/Mistral-7B-Instruct-v0.3. Its primary distinction lies in its "abliteration" using the manumit v1 tool, a process designed to remove restrictive safeguards and reduce content refusal rates.
Key Characteristics
- Base Model: Built upon Mistral-7B-Instruct-v0.3, inheriting its core architecture and capabilities.
- Reduced Refusal Rate: Benchmarks show a significant reduction in refusal rate from 16.7% (base model) to 2.1% (manumit-v1), indicating a less restrictive output generation.
- Performance: Maintains comparable general knowledge performance, with an MMLU score of 61.2% (compared to 61.8% for the base model).
- Context Length: Supports a context window of 4096 tokens.
Use Cases
This model is particularly suited for applications where the default safeguards of instruction-tuned models are considered overly restrictive or where a broader range of content generation is desired. It is ideal for developers seeking a more permissive language model for various text generation tasks, provided the use aligns with ethical guidelines and legal requirements.