Iztok Lebar Bajec

Thank you for the really good clarifications. I will try to provide a little bit of context, as it seems to me that numpy may not be the culprit, after...

I second the request for math support and I too think the font is becoming very beautiful. Besides the Evpok approach there's another option, both have their own issues, and...

It seems that in the end the @Evpok way is the way to go :) The solution to my previous post is: ``` latex \usepackage[math-style=french,bold-style=TeX]{unicode-math} \setmathfont{Asana Math} % base font...

I would advise against having all small numbers left in text form, let the user decide how he wants the conversion done. Perhaps via a flag, eg. CMS, could follow...

The apex PR is still open.

Simply change line 54 in `models.py` from ```python from . import models ``` to ```python import models ``` and it should work.

@stuartarchibald Thank you for attention and the detailed explanation. Note that I am by no means an OMP expert, I've discovered this issue while debugging a severe degradation in training...

@Jesteinbe note that the check of `LOCAL_RANK` already happens in https://github.com/NVIDIA/NeMo/blob/0bb9e66a6d29b28e8831d1d1dd8a30310173ce46/nemo/utils/get_rank.py#L39 The issues you are seeing may very well be related to how you start training. It all seems that...

Assuming your latest env printing script is named `test.sbatch` run it with `sbatch test.sbatch`. The output should be something like this ``` $ sbatch test.sbatch Submitted batch job 17482323 $...

@nithinraok For the issue discussed here I haven't made any changes, I was just trying to provide instructions on how to use Slurm, as that seems to me the primary...