CCSSNE/DreamFast-qwen3-8b-heretic

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 23, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

CCSSNE/DreamFast-qwen3-8b-heretic is an 8 billion parameter language model based on Qwen's Qwen3-8B architecture, fine-tuned using the Heretic tool. This model significantly reduces refusal rates from 100/100 to 13/100 while maintaining quality, making it suitable as an uncensored text encoder. It features a 32768 token context length and is optimized for applications requiring less restrictive content generation, particularly as a text encoder for image generation models like Klein 9B.

Loading preview...

Overview

CCSSNE/DreamFast-qwen3-8b-heretic is an 8 billion parameter model derived from Qwen's Qwen3-8B, specifically modified using the Heretic v1.2.0 tool. The primary goal of this "abliteration" process was to reduce model refusals, achieving a significant drop from 100/100 to 13/100 in tested scenarios, with minimal impact on overall model quality (KL Divergence of 0.0838).

Key Capabilities & Features

  • Reduced Refusals: Significantly lowers the model's tendency to refuse prompts, making it more versatile for diverse content generation.
  • Minimal Model Damage: The abliteration process was carefully balanced to ensure low KL divergence, preserving the base model's capabilities.
  • Optimized for Image Generation: Designed to function effectively as an uncensored text encoder, particularly for models like Klein 9B.
  • Multiple Quantization Formats: Available in various formats including HuggingFace, ComfyUI (bf16, fp8, nvfp4), and GGUF (Q8_0, Q6_K, Q5_K_M, Q4_K_M, etc.) for broad compatibility and performance.
  • NVFP4 Support: Includes NVFP4 variants for efficient use with ComfyUI, leveraging native FP4 tensor cores on Blackwell GPUs or software dequantization on older hardware.

Recommended Use Cases

  • Uncensored Text Encoding: Ideal for applications where a less restrictive text encoder is required, such as guiding image generation models.
  • Creative Content Generation: Suitable for scenarios demanding more freedom in text output without frequent refusals.
  • ComfyUI Workflows: Directly compatible with ComfyUI, especially for integration with Klein 9B, offering various precision options for different hardware setups.

Finetunes
1 models