saidutta69/DeepSeek-R1-Distill-Llama-8B-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/DeepSeek-R1-Distill-Llama-8B-heretic is an 8 billion parameter language model, a decensored variant of deepseek-ai/DeepSeek-R1-Distill-Llama-8B. This model suppresses refusal behavior through targeted weight edits using Heretic v1.4.0, rather than fine-tuning, preserving the base model's knowledge and instruction-following capabilities. It is designed for developers seeking a capable 8B reasoning model that provides direct answers without refusal, suitable for deployment on consumer hardware with 8-12 GB GPUs.

Loading preview...

DeepSeek-R1-Distill-Llama-8B-heretic Overview

This model is an 8 billion parameter variant of the DeepSeek-R1-Distill-Llama-8B, specifically modified to suppress refusal behaviors. Developed by saidutta69 using the Heretic v1.4.0 tool, it employs a technique called "abliteration" which involves targeted weight edits to the attention output and MLP down-projections. This method is distinct from traditional fine-tuning, as it aims to remove refusal guardrails while largely preserving the base model's original knowledge and instruction-following abilities.

Key Capabilities

  • Decensored Responses: Provides direct answers to queries that the base model might refuse, without additional safety filtering.
  • Preserved Core Abilities: Maintains the strong reasoning and instruction-following capabilities of the DeepSeek-R1-Distill-Llama-8B base model.
  • Efficient Deployment: Optimized for running on consumer-grade GPUs, with GGUF quantizations (Q4_K_M, Q5_K_M, Q6_K, Q8_0) available to fit various hardware configurations (e.g., 8GB, 12GB, 16GB, 24GB GPUs).
  • Broad Compatibility: Supports various inference frameworks including llama.cpp, transformers, Ollama, LM Studio, Jan, vLLM, and SGLang.

Good For

  • Developers: Ideal for those who require a powerful 8B reasoning model that delivers direct, uncensored responses.
  • Resource-Constrained Environments: Suitable for deployment on consumer hardware, including gaming PCs, due to its efficient GGUF quantizations.
  • Research and Experimentation: Useful for exploring model behavior without refusal mechanisms, understanding the impact of abliteration, or building applications where directness is prioritized.