How many epochs did you train and what was the final loss_gpt_total?
That's a pytorch error, if you google it you will find some suspected causes and possible workarounds.
Depends on the batch size/step count, but in your case yes. Looking at the final line of your training log:
23-06-24 17:35:30.808 - INFO: Training Metrics: {"loss_text_ce": 4.19026517868042,…
Hmm, I can't find anything wrong with the audio samples but 500 steps is probably too few for the number of samples you have. You might need to do around 3-4x more to get something that matches…
Taking a wild guess that the problem is trying to process UTF encoded bopomofo using the default tokenizer. You might need to look at the example in models/tokenizers/japanese.json and write…
Also, i have a question, the documentation says ffmpeg is needed for training, however, it never says where to put ffmpeg after download it.
I believe you need to install it via pip so it can…
"[WinError 2] The system cannot find the file specified" is the same error as before, so it looks like the problem is not related to the cloning software but more likely something to do with your…
Please upload your training data for the model (excluding the finetune folder).
What did you set for Resume State Path when you tried to resume the training?
To install the normal whisper what i did was what the github repo said: pip install git+https://github.com/openai/whisper.git . That worked, in the screenshot you can see it says it loaded…
Hmm, it's a valid .wav file... it should be able to convert it from there. Can you try running whisperx on (or just whisper if that's what you have installed) and see if it throws an error?
It's the second-to-last message that indicates the root of the problem. It can't find the file, so it can't transcribe anything, so the whisper.json never gets made.
Something may have gone wrong with the transcription, please post your console log.
The error is that large model isn't supported
According to the whisperx documentation large is aliased to large-v2, ex:
Can you tell me how to generate the cond_latents files?
Click the (Re)Compute Voice Latents button.
Activate the venv and then run pip install -r requirements.txt
If that doesn't fix it run the setup script again.
Unless your time is worthless the amount of development (not just coding but testing and debugging) this would require would likely cost more than buying a second GPU.