OpenYourMind/OpenYourMind-Qwen3.6-35B-A3B-kuato-DPO-abliterated-uncensored
OpenYourMind-Qwen3.6-35B-A3B-kuato-DPO-abliterated-uncensored is a 35.1 billion parameter Qwen3.6 MoE vision-language model developed by OpenYourMind. This model has undergone refusal ablation and DPO retraining with a private dataset to eliminate hedging behavior, achieving 0 refusals on HarmBench in testing. It retains the original Qwen3.6 multimodal functionality, making it suitable for applications requiring uncensored responses and visual understanding.
Loading preview...
Overview
OpenYourMind-Qwen3.6-35B-A3B-kuato-DPO-abliterated-uncensored is a 35.1 billion parameter Qwen3.6 MoE vision-language model developed by OpenYourMind. It is a modified version of Qwen/Qwen3.6-35B-A3B that has been specifically engineered to remove refusal and hedging behaviors.
Key Capabilities
- Refusal Ablation: The model underwent a process of refusal ablation, followed by DPO (Direct Preference Optimization) retraining using a private dataset of fewer than 1000 samples.
- Uncensored Responses: In testing, this model demonstrated 0 refusals on HarmBench and exhibited 0 hedging behavior compared to other ablated models.
- Multimodal Functionality: The original Qwen3.6 vision layers were readded after ablation and DPO, ensuring the model retains its full multimodal capabilities, including image and video processing.
- FP16 Safetensors: The model is provided in FP16 safetensors format, including the repaired visual tower.
Good for
- Applications requiring uncensored and direct responses without hedging.
- Tasks that benefit from multimodal understanding, combining text and visual inputs.
- Developers seeking a powerful 35.1B parameter model with robust refusal-free performance.