-
**Describe your question**
When we load model by using
Quartznet=EncDecCTCModel.load_from_checkpoint(ptl checkpoint path')
And after that we train this model by using
Trainer.fit(Quartznet)
I…
-
**Issue**
According to the documentation (https://www.quartz-scheduler.net/documentation/quartz-3.x/tutorial/configuration-resource-usage-and-scheduler-factory.html#logging), the Quartz logging can b…
-
**Describe the bug**
I'm trying to export Quartznet model to onnx by using asr_model.export() function. However, the generated onnx model doesn't have the input length as an input and it only has 'au…
Slyne updated
2 years ago
-
I think we should have a better default way to manage jobs that are "in processing".
If we poll the BackgroundJobs table every 5 seconds (default), then any jobs which have been started but not fin…
-
Hi,
Thank you for your great repo. I have 2 questions:
1- Is it better to use G2P for English ASR (or just use the raw text for it)? if yes, what model (G2P) do you suggest for this work?
2- Is it…
-
Would you please provide some benchmark on the performance of NeMo ASR on CPU and GPU?
I am particularly interested to know if I can increase the throughput of ASR by using a GPU instead of a CPU? …
-
Hello,
Thank you for great toolkit, tutorials and models.
I have some questions:
1. I want to use pretrained Citrinet in Online_ASR_Microphone_Demo instead of QuartzNet. I changed normalization t…
-
I try perform steps from https://colab.research.google.com/github/NVIDIA/NeMo/blob/stable/tutorials/asr/ASR_with_Transducers.ipynb witn my own dataset.
After 10% of first epoch i see in tensorboar…
-
## 🐛 Bug
Since the CLI subcommands were added in 1.5 my config files break after running because a new top level key "fit" is added. I run the command `python train_scripts/trainer.py fit --config …
-
I noticed that for ASR training batch size setup, I noticed that there is a trick to set the accumulate_grad_batches value to cater for large batches, the performance is still quite different between…