-
**Describe the Bug**
The compilation failed. I installed it in my torch18 conda virtual environment, compiled to the following line, waited a long time, and then stopped. No error will be reported. T…
-
I am not able to run training using tf.distribute.Strategy
However, it works fine without distribution.
Below is the code block for training loop
```
from __future__ import absolute_import
from…
-
Tried to train Transformer Transducer model on Librispeech "train-clean-100" dataset on 16GB GPU. But getting "CUDA out of memory" error . Also tried by splitting various layers on 3 GPUs, 16GB each, …
-
cuda: 11.3, pytorch: 1.10.1, python: 3.8.5
I trained a conformer transducer model using aishell recipe, but got this as follows, could anyone give me a hand?
2022-01-08 19:48:09,317 (asr_train:563)…
-
After 30 epochs of model training, inf loss appeared in a certain batch of training on my own dataset.
My train script is modified from egs/librispecch/pruned_transducer_stateless2.
The problem see…
-
Hi,
I've seen that you use Numba for the implementation forward-backward algorithm. I was wondering what is the benefit of using Numba instead of trying to rely on PyTorch functionality to have it i…
-
There are still a number of open issues which should be done for the first "official" release (or at least addressed, or maybe deliberately postponed) (#32).
Even before that, I think the current f…
-
Hi, Fangjun @csukuangfj , can you add the benchmark on https://github.com/1ytic/warp-rnnt ? `warp-rnnt` says it's faster than `warp-transducer` (non-optimized ver.)
And I have a question about the …
-
I used `kaldifeat` to extract some features and stored them using the default storage type, which is `LilcomChunkyWriter`, but it seemed to be throwing some errors at the time of data loading:
```
…
-
Hi! In PyTorch 1.10 the [RNNTLoss was added](https://pytorch.org/blog/pytorch-1.10-new-library-releases/#beta-rnn-transducer-loss). Will you maintain the speechbrain [transducer_loss](https://github.c…