DruAman/Huihui-Qwen3-14B-abliterated-v2

TEXT GENERATIONPricing:Input $0.48 / Output $0.96Concurrent Unit Cost:1Model Size:14BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 12, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

DruAman/Huihui-Qwen3-14B-abliterated-v2 is a 14 billion parameter Qwen3-based causal language model developed by huihui-ai. This model is an uncensored version, created using an abliteration method to remove refusals, offering improved performance over previous iterations. It is specifically designed for research and experimental use where reduced safety filtering is desired, providing direct and unfiltered outputs.

Loading preview...

Model Overview

Huihui-Qwen3-14B-abliterated-v2 is a 14 billion parameter language model based on the Qwen3 architecture, developed by huihui-ai. This version is specifically engineered to be uncensored, utilizing an "abliteration" method to remove refusal behaviors from the base Qwen3-14B model. It represents an improved iteration over its predecessor, focusing on providing direct and unfiltered responses.

Key Capabilities

  • Uncensored Output: Designed to generate responses without the typical safety filtering found in standard LLMs, allowing for a broader range of content generation.
  • Improved Abliteration Method: Incorporates a new and faster abliteration technique, yielding better results in removing refusals compared to previous versions.
  • Qwen3 Base: Leverages the underlying capabilities of the Qwen3-14B model for language understanding and generation.
  • Ollama Integration: Directly available for use with Ollama, including a toggle for "thinking" mode.

Usage Warnings and Considerations

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

  • Risk of Sensitive Content: Outputs may include sensitive, controversial, or inappropriate content.
  • Not for All Audiences: Not suitable for public-facing applications or underage users due to limited content filtering.
  • Legal and Ethical Responsibility: Users are solely responsible for ensuring compliance with local laws and ethical standards.
  • Research and Experimental Use: Primarily recommended for research, testing, or controlled environments, not for production or commercial applications.
  • Monitoring Required: Users are advised to monitor outputs in real-time and conduct manual reviews to prevent the dissemination of inappropriate content.

Good for

  • Research into Unfiltered LLM Behavior: Ideal for studying how LLMs respond without safety constraints.
  • Experimental Content Generation: Suitable for generating diverse content where typical LLM guardrails are undesirable.
  • Controlled Testing Environments: Useful in scenarios where strict control over output review is maintained.

Finetunes
1 models