Virgile Boulanger

@mjhewitt1 Did you figure out how to finetune the Transformer LM with your dataset ? @TParcollet It's not clear to me why we would still need the HuggingFace Dataset dependency...

Thanks a lot for your answer. I did have to modify the `dataio_prepare` function in `train.py` as well in order to make it work.

Hello ! Check out this [forked version](https://github.com/cannoneyed/RenderMan) that can load and render entire MIDI files :)

Hello, I think the problem mainly comes from the fact that in Keras, any custom loss should be designed this way: "The function should takes the following two arguments: y_true:...

Hi, Not sure if it's really a memory leak, as the audio batches can have different length during training. In my case I had the GPU mem % increasing then...

One option is to disable Pytorch Lightning's `auto_scale_batch_size`. When set to `False` there is not OOM error during the 1st epoch. I guess that it is scaling the batch size...

Same issue here with hparams : ```python ./openspeech_cli/hydra_train.py \ dataset=librispeech \ tokenizer=libri_subword \ model=contextnet_transducer \ audio=fbank \ lr_scheduler=warmup_reduce_lr_on_plateau \ trainer=gpu \ criterion=transducer \ trainer.sampler=smart \ trainer.batch_size=4 \ trainer.accumulate_grad_batches=8

Hi @OleguerCanal, Did you figure out which encoder - decoder pair was the most successful regarding your experiments ? I am training Contextnet & Conformer encoders together with transducer decoders...

Hello ! Very much looking forward to this feature. Right now is it very complicated / hacky to be able to build and containerize Bentos in a CI pipeline for...

Any update on this ?