llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved

TEXT GENERATIONConcurrency Cost:3Model Size:35.1BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 24, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved is a 35.1 billion parameter causal language model, a decensored version of Qwen/Qwen3.5-35B-A3B. Developed by llmfan46 using Heretic v1.3.0 with Magnitude-Preserving Orthogonal Ablation, it significantly reduces refusals by 85% (14/100 vs 92/100) while maintaining original model quality with a KL divergence of 0.0487. This model is optimized for applications requiring less content restriction and supports a native context length of 262,144 tokens, extensible up to 1,010,000 tokens via YaRN scaling.

Loading preview...

Overview

This model, llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved, is a 35.1 billion parameter decensored variant of the Qwen/Qwen3.5-35B-A3B base model. It was created by llmfan46 using the Heretic v1.3.0 tool and a Magnitude-Preserving Orthogonal Ablation (MPOA) method, specifically targeting attn.o_proj, attn.out_proj, and mlp.down_proj components.

Key Differentiators

  • Reduced Refusals: Achieves an 85% reduction in content refusals (14/100 vs. 92/100 for the original model), making it suitable for less restricted content generation.
  • Quality Preservation: Maintains high model quality with a low KL divergence of 0.0487 compared to the original, indicating minimal degradation in coherence and reasoning.
  • Multimodal Capabilities: Inherits Qwen3.5's unified vision-language foundation, supporting image and video inputs, and excelling in tasks like VQA, document understanding, and spatial intelligence.
  • Extended Context: Natively supports a 262,144-token context length, extensible up to 1,010,000 tokens using YaRN scaling techniques.
  • Agentic Features: Strong tool-calling capabilities, recommended for use with Qwen-Agent and Qwen Code for building advanced AI applications.

Ideal Use Cases

  • Uncensored Content Generation: Perfect for applications where strict content filtering is undesirable or needs to be bypassed.
  • Multimodal AI: Suitable for tasks involving complex visual and textual understanding, such as image captioning, video analysis, and document processing.
  • Long-Context Applications: Effective for processing and generating very long texts, code, or detailed narratives.
  • Agent Development: Well-suited for building AI agents that require robust tool-use and reasoning abilities.

Finetunes
5 models