PyTorch adds new dev tools

ai.facebook.com

85 points·pesenti··7 comments

Despite being on the FB-naysayers' side, I think that scientific and engineering work that FB outputs is amazing and greatly adds to our (tech) community. I use several of their frameworks on a daily basis and look forward into dabbling with the new stuff.

PyTorch is the best thing which came out of FB (I love it). However, their company’s core products are IMHO a cancer on today’s society

However, despite that engineering work, FB's UI seems broken at times, at least on Android. The app crashes a lot, it often seems to be in a confused state, and it contains basic UI design errors (such as when using the Back button, you almost never end up at exactly the same screen and scroll-position as where you started from).

Absolutely! PyTorch and React are two of my favorite tools ever.

Pytorch is amazing and Facebook's open source contributions to AI are great.

However, tooling for deep learning in general is not ready for industry grade technology. Many bugs could be prevented by dependent types, but compilers are not there. Also, debugging models feels like alchemy and random changes until it works. In addition, in production systems, rigorous testing is not a standard. The closest thing I have heard of is Tesla's data engine and AI system, they do have unit tests and a shadow mode. Of course big companies will have similar technologies for critical systems, but it's not as standardized as testing in software engineering.

This stuff looks super cool. I’m a fan of pytorch and can’t help but add a shameless plug for the ML experiment management tools I built - https://www.wandb.com/blog/monitor-your-pytorch-models-with-...

The `nn.BatchNorm` CPU inference speedup is a Big Deal™ to us.

Thanks for the contributions everyone, and glad to have bet on PyTorch a few years back.

Reply on news.ycombinator.com