Raziel Alvarez

Distinguished Engineer, Apple AIML. Previously: Lead @PyTorch and @TensorFlow. Co-founder of ExecuTorch and TensorFlow Lite

[PyTorch @PyTorch]

Want to learn more about PyTorch 2.0? We’re hosting a series of live Q&As for you to Ask the Engineers!

Sessions start on Dec. 13 with @subramen talking about the new developer experience of using PyTorch 2.0. See full schedule here: https://t.co/oZFV4QDOV1

[Raziel Alvarez @razielag]

2023 and 2.x will see more from PyTorch for high-efficiency production systems, including (but not exclusive to) mobile and #tinyml — livestream #PyTorchConference 2022 https://t.co/XIo75CtrOS

[AI at Meta @AIatMeta]

Many of the mobile experiences people know and love across our apps like Facebook & Instagram are made possible by running AI models on-device.

Today, we’re sharing more information about how Meta uses @PyTorch to enable these experiences ⬇️

[Christian Keller @christiankeller]

With @razielag presenting the advances of @PyTorch in on-device AI at the #PyTorchConference, make sure to check out how PyTorch is being used on-device at Meta in this blog!

https://t.co/owfaXME6qG

[PyTorch @PyTorch]

We just introduced PyTorch 2.0 at the #PyTorchConference, introducing torch.compile!

Available in the nightlies today, stable release Early March 2023.

Read the full post: https://t.co/DFO0HojkMt

🧵below!

1/5

[Raziel Alvarez @razielag]

This Friday (Dec 2nd, 9am-11:20am CT) I’ll be speaking at the #PyTorchConference on how PyTorch will support On-Device ML and the principles on which this new framework is being developed. Presentations will be live-streamed on Facebook. https://t.co/fKW63kjixe

[Raziel Alvarez @razielag]

Just getting started…

[Meta Newsroom @MetaNewsroom]

Today, Mark Zuckerberg announced the start of a new era for @PyTorch.

The new PyTorch Foundation, run by the Linux Foundation, will help accelerate progress in AI research by maintaining open collaboration and standardizing resources.

Learn more 👇
https://t.co/VdnIDLciVM

[Ashok Elluswamy @aelluswamy]

Presented some of the recent work from the Tesla Autopilot team at CVPR this year, especially about "Occupancy Networks" - our approach to solve general obstacle detection and using it to enable sophisticated collision avoidance. Full talk here: https://t.co/wpGXlNHaWl (1/12)

[AI at Meta @AIatMeta]

“Make-A-Scene provides a level of control that’s been missing in other SOTA generative AI systems. Text prompting alone is constrained, often like wandering in the dark. Being able to control the composition is a powerful extension for artists & designers” - artist @_ScottEaton_

[AI at Meta @AIatMeta]

(1/2) We recently announced the release of OPT-175B and now, as promised, we’re releasing OPT-66B, the largest unrestricted open-sourced model to date. We're also releasing our logbook used for training all of our baselines: 125M through 66B. https://t.co/pjr8PeqW2A

[Raziel Alvarez @razielag]

@dmarthal @petewarden I think it's a UX call of how much you'd be able to get in terms of value from a purely on-device experience. For assistants it's tricky to split on-device only from the full experience (it'd be confusing to users and arguably not worth it). So focus has been on latency.

[Raziel Alvarez @razielag]

A lot of the effort to move ML on device is to have better privacy guarantees (plus other benefits in terms of user experience).

[Pete Warden @petewarden]

Is Google spying on your conversations?
https://t.co/v8H5VluatV

[Raziel Alvarez @razielag]

@nutsiepully There's always a tradeoff between how much we share and what convenience we get out of the platform where we share it. Overall we want to make sure users are always aware of what they're sharing and what they'll get in return. Btw we never listen unless a user triggers it.

[Pete Warden @petewarden]

Is Google spying on your conversations?
https://t.co/v8H5VluatV

[PyTorch @PyTorch]

Today marks 5 years since the public release of PyTorch! We didn't expect to come this far, but here we're🙂- 2K Contributors, 90K Projects, 3.9M lines of "import torch" on GitHub. More importantly, we're still receiving lots of love and having a great ride. Here's to the future!

[PyTorch @PyTorch]

PyTorch Developer Day brings together leading academics, researchers and developers from the Machine Learning community to learn more about software releases on PyTorch and ways PyTorch is being used in academia and industry. https://t.co/4Mu3K2TkK8

[PyTorch @PyTorch]

5. @geeta4c - Manager, Partner Engineering, AI @MetaAI
6. @razielag + @raedle - Software Engineers @MetaAI
7. @txhf - Software Engineer @MetaAI
8. Christian Keller (Product Manager) + Linbin Yu (Product Manager) - @MetaAI

7/10

[PyTorch @PyTorch]

Get ready for PyTorch Developer Day on December 1-2, 2021! We’ve got an amazing lineup of speakers for you on Day 1.

And don’t forget to register for #PTD2 Day 2 taking place on https://t.co/WvzVX0JKyV: https://t.co/lwHkXwfnRn

Check the thread below to see the speakers ⬇️
1/10

[Raziel Alvarez @razielag]

Is textless NLP a truer form of NLP?

[AI at Meta @AIatMeta]

GSLM, our first-of-its-kind “textless #NLP” model, learns to generate expressive speech using only raw audio as input. https://t.co/zSevs3Xujm
It generates novel content & prosody congruently with a prompt’s expressive style, as in this example from Jane Austen’s Mansfield Park:

[Edward Z. Yang @ezyang]

State of PyTorch core, September 2021 edition: https://t.co/PA7WBJ3JnR