huihui-ai/Huihui-Qwopus3.5-9B-v3-abliterated

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Huihui-Qwopus3.5-9B-v3-abliterated is a 9 billion parameter uncensored language model derived from Jackrong/Qwopus3.5-9B-v3, created using an abliteration technique to remove refusal behaviors. This model is specifically designed for research and experimental use in environments where reduced safety filtering is desired, offering a 32K context length. It is intended for exploring model capabilities without standard safety constraints, making it suitable for controlled testing rather than production applications.

Loading preview...

Overview

Huihui-Qwopus3.5-9B-v3-abliterated is a 9 billion parameter language model based on the Jackrong/Qwopus3.5-9B-v3 architecture. Its primary distinction lies in its "abliterated" nature, meaning its safety filtering mechanisms have been significantly reduced or removed. This process, detailed in the remove-refusals-with-transformers project, aims to create an uncensored version of the original model.

Key Characteristics

  • Uncensored Output: Safety filtering has been substantially reduced, allowing for potentially sensitive, controversial, or inappropriate content generation.
  • Research-Oriented: Recommended for research, testing, and controlled environments due to its lack of standard safety guarantees.
  • Experimental: Represents a proof-of-concept for removing refusal behaviors from LLMs without relying on TransformerLens.
  • 9 Billion Parameters: A moderately sized model offering a balance between capability and computational requirements.
  • 32K Context Length: Supports processing longer input sequences.

Usage Considerations

This model comes with significant usage warnings due to its uncensored nature:

  • Risk of Inappropriate Content: Users must exercise extreme caution and rigorously review all generated outputs.
  • Not for Public/Production Use: It is explicitly stated as unsuitable for public settings, underage users, or commercial applications requiring high security.
  • User Responsibility: Users are solely responsible for ensuring compliance with legal and ethical standards for any content generated.

When to Use This Model

This model is best suited for:

  • Academic Research: Investigating the effects of censorship removal on LLM behavior.
  • Controlled Testing: Exploring model responses to prompts that would typically be refused by standard, safety-aligned models.
  • Development of Safety Tools: Potentially useful for developing and testing new safety filters or moderation techniques against a less constrained model.