richardyoung/Llama-3.1-8B-Instruct-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Jun 24, 2026License:llama3.1Architecture:Transformer Featherless Exclusive Cold

richardyoung/Llama-3.1-8B-Instruct-heretic is an 8 billion parameter instruction-tuned causal language model, a decensored version of Meta's Llama-3.1-8B-Instruct. This model, created using Heretic v1.4.0, significantly reduces refusals compared to the original, making it suitable for use cases requiring less content moderation. It maintains the Llama 3.1 architecture's multilingual text capabilities and 8192 token context length, optimized for dialogue and general natural language generation tasks.

Loading preview...

Model Overview

This model, richardyoung/Llama-3.1-8B-Instruct-heretic, is an 8 billion parameter instruction-tuned variant of Meta's Llama-3.1-8B-Instruct. It has been 'decensored' using Heretic v1.4.0, resulting in a substantial reduction in refusal rates (3/100 compared to 96/100 for the original model).

Key Capabilities & Differentiators

  • Decensored Output: Significantly fewer refusals, offering greater flexibility for applications where strict content moderation is not desired or is handled externally.
  • Llama 3.1 Foundation: Inherits the robust capabilities of the Llama 3.1 architecture, including an 8192 token context length and multilingual text input/output.
  • Multilingual Support: Optimized for dialogue in English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
  • Tool Use: Supports advanced tool use and function calling, with detailed guidance available for integration.
  • Performance: Maintains strong performance across general, reasoning, code, and math benchmarks, with notable improvements in areas like HumanEval (72.6% pass@1) and MATH (51.9% final_em) compared to Llama 3 8B Instruct.

Intended Use Cases

This model is designed for commercial and research use, particularly for assistant-like chat and natural language generation tasks where a less restrictive output policy is preferred. Developers can leverage its capabilities for synthetic data generation and distillation, and it is suitable for integration into agentic systems with external safeguards.

Finetunes
1 models