RuntimeRacer

I tried this using docker-compose both on: - Win 10 - Ubuntu Server 22.04 Also changed CORS to accept everything. Same problem. Error Says HTTP 426 Upgrade required :/ UPDATE:...

All of that didn't work for me. I'm always getting 426 Upgrade Required by the HTTP Requests. As if for some reason it is routing the HTTP calls always to...

https://github.com/0417keito/VALL-E-X-Trainer-by-CustomData

I think the Inference times on an RTX 3090 / 4090 are quite acceptable actually. Not perfect, but acceptable. What Hardware are you using? > Implementing `kv cache` can speedup...

Can confirm this also happens with `max-duration` 60 after exactly 164.500 steps again. Probably Epoch switch, which however breaks the process. Not sure what it tries to save there, but...

Might also be related to the Issue I mentioned here in combination with Japanese / Chinese symbols, since my dataset contains these: https://github.com/lifeiteng/vall-e/issues/94#issuecomment-1530323074 I'll try another iteration on the model...

From my testing yesterday it was able to transfer dialect from input sample to output even if it's a different language. What Vall-E X does is actually getting rid of...

> Do you mean we can train it on multiple languages already? I have a PR open for the Commonvoice Dataset and it is currently training on 24 different languages...

Further debugging the issue revealed that the crash seems to happen frequently if there are Cyrillic letters in the batch; processing these takes considerably more time and VRAM. It's less...

@nivibilla I did follow the exact process of dataset preparation for my CommonVoice training; I assume the phoneme conversion has already happened; also it apparently IS able to process these...