Open leviathanch opened 8 months ago
Interesting.
RuntimeError: Buffer is not valid for unallocated defvice
This is actual problem. Something fails in cpu fallback. I need to check why it fails.
In any case I'm working on layer_norm
now to enable VITS in general - they currently don't work.
Also I must say lots of stuff is still missing and I hadn't tested anything related to NLP (mostly due to lack of experience in that area)
That's the code I've been running, in case it helps:
import torch
torch.ops.load_library("/usr/local/lib/libpt_ocl.so")
torch.ops.load_library("/usr/local/lib/libdlprim_core.so")
torch.utils.rename_privateuse1_backend('ocl')
from TTS.tts.configs.vits_config import VitsConfig
from TTS.tts.models.vits import Vits
from TTS.api import TTS
print(TTS().list_models())
tts = TTS("tts_models/bn/custom/vits-female").to("ocl:0")
wav = tts.tts(text="Hello world!", speaker_wav="annitta.wav")
tts.tts_to_file(text="Hello world!", speaker_wav="annitta.wav", file_path="output.wav")
When I try to run VITS I get this: