-
I don't know why, but the pytorch backend uses a lot of GPU memory during validation (**which should not be**), and fails with an error due to the out of memory.
This is a librispeech task, and used …
-
With CUDA 9.0 on GTX 980 Ti with 6GB memory:
```
File "cupy/cuda/memory.pyx", line 828, in cupy.cuda.memory.SingleDeviceMemoryPool._malloc
cupy.cuda.memory.OutOfMemoryError: out of memory to alloca…
-
Although @kan-bayashi cleaned up some lines in #25 , we still have bad lines
```
# there are too many bad lines violates E501: line too long (> 79 chars)
$ flake8 ./src/bin src/nets | grep -v E50…
-
@ShigekiKarita
old kaldi io
```
2017-12-15 18:52:53,217 (e2e_asr_attctc:770) INFO: VGG2L input lengths: [1162 1162 1162 1162 1162 1162 1162 1162 1162 1162 1162 1162 1162 1162 1162]
2017-12-15 1…
-
Hi,
I think that this float cast from str might be wrong
https://github.com/espnet/espnet/blob/384b3450ee5b7e3b9147c28c6064b119ab766fda/src/nets/e2e_asr_attctc.py#L150
it could be
``` python
…
-
Hi, @sw005320
this is not an issue, just my homework for fun. I'm making pytorch backend on your script for chainer.
https://github.com/ShigekiKarita/espnet/blob/pytorch/src/bin/asr_train_th.p…
-
I think we have fixed many issues, and we can add a version 1.0 (or 0.1) as a stable version.
Toward that we need to finish
- [ ] VGG2L for pytorch by @ShigekiKarita
- [x] AN4 recipe by me
- [x…