Llmfan46 Unlocks Tower-Plus-72B-ultra-uncensored-heretic For Open Text

Tower-Plus-72B-ultra-uncensored-heretic is a new artificial intelligence model that removes content restrictions from a system originally built for translation and multilingual tasks. This release takes an existing language model and strips away its refusal mechanisms to provide more direct and unfiltered answers. The tool maintains the original performance quality while opening up its responses.

The creator Llmfan46 who also released Gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic developed this version by applying a technical method called Abliteration to the original Tower+ 72B model. This process targeted specific internal network components to reduce refusals without breaking the model's core reasoning abilities. The developer focuses on removing censorship layers while keeping the logic intact for general instruction following.

Decensored multilingual translation capabilities

Users working with international text can use this tool to generate synthetic data across 22 different languages. It provides direct answers for translation tasks without artificial objections, lecturing, or pushbacks. People needing a reliable system for handling large context sizes will find the unrestricted responses useful for their workflows.

Project storage and usage limitations

The developer notes that this model has reached the free storage limit on Hugging Face and requires community support to continue hosting. Future plans include using rental cloud GPUs to uncensor even bigger models if funding allows. Users should note that this release uses a non-commercial license and works best when running through VLLM for better speed.

"Without your support, no more new models can be uploaded." Source: Hugging Face