ONNX / GGML inference
#2
by 234r89r23u89023rui90 - opened
Hi,
congrats on this spectacular release. The model, and especially the voice‑cloning capabilities, seem insanely good. Do you plan to release or contribute inference code that can run on edge devices without GPUs e.g., ONNX models or a lightweight GGML implementation?
Best regards.
This comment has been hidden (marked as Off-Topic)
The current PyTorch version is rather slow on CPU. I’m not sure whether an ONNX or GGML implementation can resolve this issue, but we will look into it. Contributions for a CPU-optimized, faster version are very welcome.
zhu-han changed discussion status to closed