saidutta69/DeepSeek-R1-Distill-Qwen-7B-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 21, 2026License:deepseek-research-licenseArchitecture:Transformer Featherless Exclusive Cold

The saidutta69/DeepSeek-R1-Distill-Qwen-7B-heretic is a 7.6 billion parameter language model, a decensored variant of deepseek-ai/DeepSeek-R1-Distill-Qwen-7B. It suppresses refusal behavior through targeted weight edits using the Heretic v1.4.0 abliteration method, rather than fine-tuning, preserving the base model's knowledge and instruction-following. This model is designed for developers seeking a capable 7B reasoning model that provides direct answers without refusal guardrails, suitable for deployment on 8-12 GB GPUs.

Loading preview...

Overview

The saidutta69/DeepSeek-R1-Distill-Qwen-7B-heretic is a 7.6 billion parameter model derived from deepseek-ai/DeepSeek-R1-Distill-Qwen-7B. Its primary distinction lies in its "decensored" nature, achieved by suppressing refusal behaviors through a technique called abliteration using Heretic v1.4.0.

Key Capabilities & Differentiators

  • Refusal Suppression: Unlike traditional fine-tuning, this model uses targeted weight edits to the attention output and MLP down-projections to remove refusal guardrails. This method aims to keep the base model's core knowledge and instruction-following capabilities largely intact.
  • Base Model Preservation: The abliteration technique avoids the coherence degradation often seen when fine-tuning a "helpful" persona over RLHF'd refusals, ensuring the original model's reasoning abilities are maintained.
  • Direct Answering: Designed to provide direct answers to queries that the base model might refuse, making it suitable for use cases requiring unfiltered responses.
  • Hardware Compatibility: Optimized for deployment on consumer hardware, specifically recommending an 8-12 GB GPU or via Q4_K_M/Q5_K_M GGUF quantizations.

Intended Use

This model is intended for developers who require a capable 7B reasoning model that offers direct, unfiltered responses. It is important to note that the deliberate suppression of refusal behavior means the model will comply with requests the base model would refuse, including potentially inappropriate ones. Users are responsible for its deployment and usage.

Technical Details

  • Context Length: Supports a context length of 32768 tokens.
  • Quantizations: Available in various GGUF quantizations, including Q4_K_M, Q5_K_M, Q6_K, and Q8_0, for efficient deployment.
  • License: Inherits the deepseek-research-license from its base model.

Finetunes
1 models