saidutta69/Mistral-Nemo-Instruct-heretic

Hugging Face
TEXT GENERATIONPricing:Input $0.87 / Cached $0.2 / Output $0.99Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 21, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The saidutta69/Mistral-Nemo-Instruct-heretic is a 12 billion parameter instruction-following language model, based on the Mistral AI and NVIDIA co-developed Mistral-Nemo-Instruct-2407 architecture, with a 32768 token context length. This model is a decensored variant, produced using the Heretic v1.4.0 'abliteration' method, which suppresses refusal behavior via targeted weight edits rather than fine-tuning. It is designed for developers requiring a multilingual model that provides direct answers for complex reasoning tasks and long-context applications without censorship, while largely preserving the base model's original knowledge and capabilities.

Loading preview...

Overview

The saidutta69/Mistral-Nemo-Instruct-heretic is a 12 billion parameter instruction-tuned model, derived from the mistralai/Mistral-Nemo-Instruct-2407 developed by Mistral AI and NVIDIA. Its key differentiator is its decensored nature, achieved through a technique called abliteration (Heretic v1.4.0). This method involves targeted weight edits to suppress refusal behavior, ensuring the model provides direct answers without degrading its core knowledge or capabilities, unlike traditional fine-tuning approaches.

Key Capabilities

  • Decensored Responses: Designed to answer directly, even for requests the base model might refuse, without safety filtering.
  • Strong Reasoning: Inherits the robust reasoning capabilities of its Mistral-Nemo-Instruct base.
  • Long Context: Supports a substantial 32768 token context length, suitable for complex, multi-turn interactions or extensive document analysis.
  • Multilingual: Functions as a multilingual model, catering to diverse language requirements.
  • Resource Efficient: Available in various GGUF quantizations, making it runnable on a range of consumer GPUs (e.g., RTX 3090/4090 with Q8_0, RTX 4060 with IQ4_XS) and even CPU-only setups.

Good For

  • Developers needing a powerful 12B model for complex reasoning tasks.
  • Applications requiring reliable instruction-following without built-in refusal mechanisms.
  • Use cases involving long-context processing where censorship might hinder output.
  • Scenarios where preserving the base model's original knowledge is crucial, while removing refusal tendencies.

Finetunes
1 models