Nabbers1999/Stylizer-V2-LLaMa-70B-heretic

Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:8kPublished:May 24, 2026Architecture:Transformer Warm

Nabbers1999/Stylizer-V2-LLaMa-70B-heretic is a 70 billion parameter LLaMa-based model, derived from TareksGraveyard/Stylizer-V2-LLaMa-70B and further decensored using Heretic v1.3.0. This model is a merge of several Llama-3.1 and Llama-3.3 variants, specifically engineered to significantly reduce refusals compared to its original counterpart. It is optimized for applications requiring less restrictive content generation and direct responses.

Loading preview...

Model Overview

Nabbers1999/Stylizer-V2-LLaMa-70B-heretic is a 70 billion parameter LLaMa-based model, created by Nabbers1999. It is a decensored version of TareksGraveyard/Stylizer-V2-LLaMa-70B, processed using the Heretic v1.3.0 tool. This model is reproducible, with details available in its reproduce directory.

Key Characteristics

This model is a sophisticated merge of several Llama-3.1 and Llama-3.3 variants, including:

The merge was performed using the SCE (Sparse Component Ensemble) method via mergekit, with huihui-ai/Llama-3.3-70B-Instruct-abliterated serving as the base model.

Performance and Differentiation

A primary differentiator of this model is its significantly reduced refusal rate. While the original model exhibited 30 refusals out of 100, this decensored version achieves a refusal rate of only 8 out of 100. This makes it particularly suitable for use cases where direct and unfiltered responses are preferred. The model maintains a KL divergence of 0.0419 compared to the original, indicating a controlled modification while achieving its decensoring objective.