Peter Pisljar

this works ok with my data/config on 2.0.6

in database.py the batch size is set to total batch size (rather than batch size per gpu). this makes _collate_fn return empty batch array. by fixing this i get batches...

trying to set `find_unused_parameters=True` on DistributedDataParallel does NOT solve the problem

File "/home/tts/efficientspeech/model.py", line 156, in forward │ return self.phoneme2mel(x, train=True) if self.training else self.predict_step(x) │ File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl │ return forward_call(*args, **kwargs) │ File "/home/tts/efficientspeech/layers/networks.py", line 425,...

seems the problem is that the font does not support all the characters. you could try replacing the font with one that supports CN

I am not familiar with transit API, probably it would require some modifications to the code.

yes, this is on my todo list for the future, just not sure when i'll get to it.

in paper its stated they used batch size 6000 on 16 v100 32gb gpus .... that makes batch size 375 per gpu (on 32gb) is batch size referred to in...