Model Overview

This model, 0xA50C1A1/Mistral-Nemo-Instruct-2407-Heretic-v2, is a 12 billion parameter instruction-tuned language model based on the Mistral-Nemo architecture. It is a modified version of unsloth/Mistral-Nemo-Instruct-2407, specifically processed using the Heretic v1.2.0 tool.

Key Differentiator

The primary distinction of this model is its decensored nature. Through the Heretic process, its refusal rate has been drastically reduced from 88/100 in the original model to 4/100. This makes it suitable for use cases where the original model's content restrictions might be too prohibitive.

Technical Details

The modification process involved adjusting specific abiteration parameters related to attn.o_proj and mlp.down_proj weights. The model maintains a 32768 token context length, characteristic of the Mistral-Nemo family.

Use Cases

This model is particularly well-suited for applications that require a language model with fewer content restrictions and a lower propensity for refusals, while still leveraging the performance of the Mistral-Nemo-Instruct-2407 base.