returnn icon indicating copy to clipboard operation
returnn copied to clipboard

The RWTH extensible training framework for universal recurrent neural networks

Results 204 returnn issues
Sort by recently updated
recently updated
newest added

Hi, I have trained a model with unidirectional lstm and local attention. How can I use this model for online decoding? Where should I make changes in the existing code?...

I want to convert the [returnn full attention setup model](https://github.com/rwth-i6/returnn-experiments/tree/master/2018-asr-attention/librispeech/full-setup-attention) to TFLite for on-device deployment. Should I reimplement the model in TF2.0, load weights and convert to TFLite? Or any...

This PR currently only provides a failing test case, not the corresponding fix. I've created an issue to discuss how to solve the problem.

I've run into a bug: If I have a base layer `state` and want to use it to initialize an lstm that's used within the recurrent beam search, the base...

I want to use tensorflow as the backend in training the models. How do I switch from Theano to Tensorflow?

E.g. for LibriSpeechCorpus, OggZipDataset, TimitDataset. It's much faster. Actually, I did some intensive research on this, and tried other alternatives, and this implementation with rnd_zoom_order=0 (i.e. nearest neighbor) turned out...

Related is the flat net construction logic (#992). However, I think the current implementation of the flat net construction logic is too difficult and too messy, using exceptions to fill...

difficulty: hard
TensorFlow

**Edit** Originally, this issue was about a proof-of-concept for a new PyTorch backend in RETURNN. This has somehow evolved into a whole new generic frontend API (original idea here: https://github.com/rwth-i6/returnn/issues/1264),...

PyTorch
returnn-frontend

Examples of post-processing: - Raw audio is stored in the HDFDataset, do feature extraction on-the-fly (but not in the network, but instead as part of the dataset). - Ogg is...

After quite a while of training (597 subepochs) with PyTorch backend, I got: ``` RETURNN starting up, version 1.20231119.003753+git.c230d140, date/time 2023-11-22-02-51-52 (UTC+0000), pid 2470397, cwd /work/asr4/zeyer/setups-data/combined/2021-05-31/work/i6_core/returnn/training/ReturnnTrainingJob.tNymT5UR0k6i/work, Python /work/tools/users/zeyer/py-envs/py3.11-torch2.1/bin/python3.11 ... PyTorch:...