saidutta69/Meta-Llama-3.1-8B-Instruct-heretic

TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Jul 21, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The saidutta69/Meta-Llama-3.1-8B-Instruct-heretic is an 8 billion parameter instruction-tuned causal language model, a decensored variant of Meta-Llama-3.1-8B-Instruct. Developed by saidutta69 using the Heretic v1.4.0 method, it suppresses refusal behavior via targeted weight edits rather than fine-tuning, preserving the base model's knowledge and instruction-following. This model is designed for developers seeking a Llama 3.1 8B variant that provides direct answers without built-in refusal guardrails, suitable for deployment on consumer hardware with 8-12 GB GPUs.

Loading preview...

Overview

saidutta69/Meta-Llama-3.1-8B-Instruct-heretic is an 8 billion parameter instruction-tuned model derived from meta-llama/Meta-Llama-3.1-8B-Instruct. Its primary distinction is the suppression of refusal behavior through a technique called "abliteration" (directional ablation) using the Heretic v1.4.0 tool. This method involves targeted weight edits to the attention output and MLP down-projections, which removes refusal guardrails while largely preserving the base model's original knowledge and instruction-following capabilities.

Key Capabilities & Features

  • Decensored Llama 3.1 8B: Provides direct answers to requests that the base Meta-Llama-3.1-8B-Instruct model would typically refuse.
  • Preserved Base Model Quality: Unlike fine-tuning, abliteration aims to maintain the coherence and capabilities of the original Llama 3.1 8B model.
  • Hardware Accessibility: Optimized GGUF quantizations (Q4_K_M, Q5_K_M, Q6_K, Q8_0) are provided, enabling efficient deployment on consumer GPUs with 8-24 GB VRAM or CPU-only setups.
  • Broad Compatibility: Supports various inference frameworks including llama.cpp, transformers, Ollama, LM Studio, Jan, vLLM, and SGLang.

Good For

  • Developers who require a Llama 3.1 8B model that will comply with a wider range of requests without built-in safety filtering or lecturing.
  • Use cases where the base model's refusal behavior is undesirable, and direct, unfiltered responses are preferred.
  • Experimentation with models that have had their refusal mechanisms surgically removed, understanding the implications of such modifications.

Note: This model deliberately suppresses safety guardrails. Users are responsible for its deployment and outputs.

Finetunes
1 models