Stas Bekman

@samyam and I had a brief discussion about it, Samyam suggested that actually this check should continue through the whole training, because the rescaling might need to happen at a...

When you hit an OVERFLOW under fp16 the optimizer skips this step, so yes it's harmless. Usually you get a few of those at the start while the program tries...

Actually I then run into a bunch of other problems related to `build_ext` as we have discussed elsewhere - dependencies `ninja` and `tqdm` won't build, even though binary wheels are...

thank you for the report, @ebowman - I have just fixed that. Please update your clone and it should work. As you're probably wanting to build the pdf I just...

It's a good suggestion, Jonas and I agree that this would be nice. The structure is still emerging and changes every month or so, e.g. I'm not sure if each...

I finally had the time to do a massive re-org in https://github.com/stas00/ml-engineering/pull/25 Hopefully it's much easier to read now. And of course I will continue improving the layout.

That's an excellent idea, Zhangzhi. If it can be automated for sure I'd be happy to receive such a PR. Have you checked whether the github market perhaps already has...

the grip recipe won't work well since this is a multi-file situation. the SO answer @amorehead linked to mentions https://wkhtmltopdf.org/downloads.html which is probably a much better tool for building pdfs...

The PDF is almost ready, please give me a few more weeks. The building workflow is ready, but I need to finish the stylesheets and restructuring the chapters. If you're...

the pdf is finally done: https://github.com/stas00/ml-engineering#pdf-version