Model Inference Hangs Indefinitely at RMVPE Model Loading

Copy# Title: Model Inference Hangs Indefinitely at RMVPE Model Loading on RTX 4060

Issue Description

When trying to perform voice conversion inference on RTX 4060, the process stops responding at the RMVPE model loading stage. The last log message shows "Loading rmvpe model,assets/rmvpe/rmvpe.pt" and then hangs for over 5 minutes with no further logs or progress indicators. This makes it impossible to determine if the process is still working or has frozen completely.

Environment

• GPU: NVIDIA GeForce RTX 4060
• CUDA Version: 12.6
• Driver Version: 560.94
• Docker container environment

Steps to Reproduce

1. Trained model successfully (model1)
2. Generated index file successfully at /app/logs/model1/added_IVF378_Flat_nprobe_1_model1_v2.index
3. Attempted inference with the following settings:
   • Input file: /app/data/M_Ehyun/EJ_DeEssed.wav
   • Model: model1
   • Pitch extraction method: Initially RMVPE, then tried pm (both hang)

Reproduce inference in the Docker environment with the reported RTX 4060 setup, starting from the log entry "Loading rmvpe model,assets/rmvpe/rmvpe.pt". Compare the RMVPE and pm pitch-extraction runs and inspect whether either produces further logs or an error; done means the hang is explained or inference progresses past model loading.