returnn
returnn copied to clipboard
rwth-i6
The RWTH extensible training framework for universal recurrent neural networks
Via `test_nn_conformer` from returnn-common, it generates a net dict which causes `Exception: maximum recursion depth exceeded`. We have `flat_net_construction` for that but this fails with some other error (probably due...
When running RETURNN, I get ``` /.../returnn/returnn/tf/network.py:1621: SyntaxWarning: "is" with a literal. Did you mean "=="? if loss is 0: /.../returnn/returnn/tf/util/basic.py:2024: SyntaxWarning: "is not" with a literal. Did you mean...
I have a case where the inputs to the `CombineLayer` are of shape `(B, F, T)` and `(1, F)`. In this case, it does currently not work to find the...
When running the `compile_native_op.sh` you will always get the BLAS lib from numpy linked into the binary. - When not adding any flag, it will load the BLAS lib from...
For the contrastive loss implementation (#918), we flatten the masked encoder frames via `FlattenBatchLayer` and end up with `B&Packed{'input_masked_frames:masked:time'}` batch dim. For all those frames, we want to create a...
Consider I have some input of shape `[T]`, where `T` has dyn_sizes with some other axes, e.g. `[B]`. I want to use `ReduceLayer` on this axis, which needs to apply...
This is kind of an follow up to #666. I have the problem currently, that `SpatialDim("a", 1)` matches `SpatialDim("b", 1)` in many cases, e.g. `get_common_data`, `copy_compatible_to` and `find_matching_dim_map` used by...
In `_SubnetworkRecCell._construct_template`, we currently fail directly on these exception types: ``` fail_directly_exception_types = (DataNotFound, LayerNotFound, BehaviorVersion.RequirementNotSatisfied) ``` When debugging, it in general is much nicer to just fail on the...