returnn
returnn copied to clipboard
rwth-i6
The RWTH extensible training framework for universal recurrent neural networks
I think we never officially stated the minimum PyTorch version we require for RETURNN (or did we? where?). I remember that 1.13 or so was mentioned (or assumed) in some...
[Haste: a fast, simple, and open RNN library](https://github.com/lmnt-com/haste). Could be both for TensorFlow and PyTorch. Put it under `extern/`, do it in a similar way like `blocksparse`, `WarpRna`, `HawkAaronWarpTransducer` or...
File `IpcFabricConfigClient.cpp`. ``` $ gdb --args python3 -m pytest tests/test_rf_array.py GNU gdb (Ubuntu 12.1-0ubuntu1~22.04) 12.1 ... Reading symbols from python3... (gdb) r Starting program: /work/tools/users/zeyer/linuxbrew/bin/python3 -m pytest tests/test_rf_array.py [Thread debugging...
Via #487, we are now supporting Python >=3.7. Do we want to bump this and require Python >=3.8 or later? See also the Python feature list in #487. Similarly, Python...
Follow-up from #1315, where @JackTemaki noticed the `NextGenHDFDataset` goes through the entire data on startup to perform integrity checks -- and indeed RETURNN startup w/ that dataset is quite slow....
And a test case which tests that together with masked-computation. See the test case for a demonstration of what should be possible with this. This is work-in-progress. I'm not sure...
Now that we have the generic Tensor and TensorDict, we can remove the old ambiguous and limited num_outputs and num_inputs from the dataset and replace it by extern_data using TensorDict...
This becomes relevant for efficient decoupled weight decay implementation. If it is not decoupled, it's inefficient anyway.
The `relative_positional_encoding` implementation in RC uses: ```python with nn.control_flow_ctx(None): ... ``` This is relevant for graph-based backends, once we have control flow logic like `nn.Cond` and `nn.Loop`. I wonder how...
Currently when `learning_rate_control_error_measure` (the error key) does not match exactly, it will use some heuristics. See `LearningRateControl.get_error_key`. This is also because the error key can change depending on whether there...