mirror of
https://github.com/RVC-Boss/GPT-SoVITS.git
synced 2025-12-17 01:48:25 +08:00
d3b8f7e09e
This commit migrates the project from using NVIDIA CUDA to Intel XPU for GPU acceleration, based on the PyTorch 2.9 release. Key changes include: - Replaced `torch.cuda` with `torch.xpu` for device checks, memory management, and distributed training. - Updated device strings from "cuda" to "xpu" across the codebase. - Switched the distributed training backend from "nccl" to "ccl" for Intel GPUs. - Disabled custom CUDA kernels in the `BigVGAN` module by setting `use_cuda_kernel=False`. - Updated `requirements.txt` to include `torch==2.9` and `intel-extension-for-pytorch`. - Modified CI/CD pipelines and build scripts to remove CUDA dependencies and build for an XPU target.