- Fix the runtime repository conflict between the uncensored app and the original app. - Move the Hugging Face and ModelScope runtimes to dedicated repositories to prevent releases from overwriting each other's metadata. - Pin runtime revisions and verify archive size and SHA256 to ensure the correct image is downloaded.

1. Fixed deployment failure caused by insufficient memory during deployment process 2. Supports disk/video memory management interface, making it convenient for users to pause/continue/delete operations 3. The CUDA kernel increases cache packages, greatly reducing deployment time 4. The problem of output garbled code has been solved. Pay attention to the configuration file settings, temperature 0.3, top_p 0.95, top_k 20 5. Unifies the model ID, models id, and Pi Agent id, and all must be set to qwen3.8-flash-next-uncensored