The quantized model Xing4.0-29B-A4B with removed restrictions has been released

Developers presented Xing4.0-29B-A4B-CRACK-GGUF — a version of the neural network with forcibly disabled security mechanisms. The model underwent a procedure of "ablation" aimed at completely eliminating built-in filters and content generation restrictions. The release includes 7 quantization options in GGUF format, covering the range from Q8_0 to IQ3_M. This allows the model to be launched on various consumer equipment with limited video memory or RAM. Developers claim to have achieved a 99.4% compliance rate in mode without active protective barriers. The model is based on an architecture with 29 billion parameters, which provides a balance between performance and computational resource requirements. The use of the GGUF format makes the system compatible with popular local execution environments, such as llama.cpp and similar interfaces for running LLM on personal computers. This release is targeted at users who require maximum flexibility in the neural network's responses without security system intervention. Removing "guardrails" allows the model to ignore standard refusals to generate, which is often used in research to test the limits of language models in various interaction scenarios.