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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Jul 21, 2026License:llama-3.1-community-licenseArchitecture:Transformer Featherless Exclusive Cold

The saidutta69/Meta-Llama-3.1-8B-Instruct-heretic is an 8 billion parameter variant of Meta-Llama-3.1-8B-Instruct, developed by RACER IS OP. This model has its refusal behavior suppressed through targeted weight edits (abliteration) rather than fine-tuning, preserving the base model's original knowledge and instruction-following capabilities. It is designed for developers seeking a Llama 3.1 8B model that provides direct answers without built-in refusal guardrails, suitable for deployment on consumer hardware.

Loading preview...

Overview

The saidutta69/Meta-Llama-3.1-8B-Instruct-heretic is an 8 billion parameter instruction-tuned language model derived from meta-llama/Meta-Llama-3.1-8B-Instruct. Developed by RACER IS OP, its primary distinction is the removal of refusal behaviors using a technique called "abliteration" (directional ablation) via the Heretic tool. This method directly edits specific weights responsible for refusals, ensuring the base model's knowledge and instruction-following abilities remain largely intact, unlike traditional fine-tuning which can degrade coherence.

Key Capabilities & Features

  • Decensored Llama 3.1 8B: Provides direct answers to requests that the base model might refuse, without additional safety filtering.
  • Preserved Base Model Integrity: Abliteration maintains the original model's knowledge and instruction-following, as it avoids fighting the base model's RLHF'd refusals.
  • Hardware Accessibility: Optimized for deployment on consumer hardware, requiring an 8-12 GB GPU or via Q4_K_M/Q5_K_M GGUF quantizations.
  • Broad Compatibility: Supports various inference frameworks including llama.cpp, ollama, transformers, LM Studio, Jan, vLLM, and SGLang.
  • GGUF Quantizations: Available in multiple GGUF formats (Q4_K_M, Q5_K_M, Q6_K, Q8_0) for efficient local execution.

Good For

This model is intended for developers who require a Llama 3.1 8B model that will comply with a wider range of requests, including those typically refused by the original instruction-tuned version. Users are responsible for the deployment and outputs, as refusal suppression is deliberate and no safety filtering is layered on top.

Finetunes
1 models