harimuliawan999/Huihui-Qwen3-14B-abliterated-v2
Huihui-Qwen3-14B-abliterated-v2 is an uncensored 14 billion parameter Qwen3-based causal language model developed by harimuliawan999. This model is created using an abliteration method to remove refusals, offering an improved version over its predecessor. It is designed for research and experimental use where reduced safety filtering is desired, providing a proof-of-concept for refusal removal without TransformerLens.
Loading preview...
Overview
Huihui-Qwen3-14B-abliterated-v2 is a 14 billion parameter language model based on the Qwen3 architecture, developed by harimuliawan999. Its primary distinction is being an uncensored version achieved through an "abliteration" process, which aims to remove refusal behaviors from the model. This iteration is an improvement over the previous huihui-ai/Qwen3-14B-abliterated model, featuring a new and faster abliteration method that yields better results and addresses issues like garbled codes by changing the candidate layer.
Key Capabilities
- Uncensored Content Generation: Designed to produce outputs without the typical safety filtering found in standard LLMs.
- Proof-of-Concept for Refusal Removal: Demonstrates a method for eliminating model refusals without relying on TransformerLens.
- Improved Abliteration: Utilizes an enhanced and faster abliteration technique compared to its predecessor.
- Ollama Support: Directly available for use via Ollama, with a toggle for "thinking" mode.
Good For
- Research and Experimentation: Ideal for exploring the effects of reduced safety filtering and refusal removal techniques.
- Controlled Environments: Suitable for use cases where strict content moderation is not required or is handled externally.
- Exploring Model Behavior: Useful for understanding how models respond when typical refusal mechanisms are bypassed.
Usage Warnings
Users should be aware that this model carries a risk of sensitive or controversial outputs due to significantly reduced safety filtering. It is not suitable for all audiences or production environments requiring high security. Users are solely responsible for ensuring their usage complies with legal and ethical standards, and monitoring and review of outputs are strongly recommended.