MACHINE LEARNING

→ View original post on X — @terence_mills

Digital Health Startups Transforming Healthcare System Through Innovation

By

@terence_mills

–

03 August 2023 21h30

How Digital Health Startups Can Help Revise The Healthcare System
#AI #AIio #BigData #ML #NLU #Futureofwork @nigelwalsh @pbouillaud @PatrickGunz_CH @pierrepinna @RebekahRadice @Ronald_vanLoon @sbmeunier
@Shirastweet http://
ow.ly/1IBV30swzJz

→ View original post on X — @demishassabis

AlphaFold Database Celebrates Two Years, 200M Protein Structures

By

@demishassabis

–

03 August 2023 21h12

It's 2 years since we launched the #AlphaFold Protein Structure Database with @emblebi
. Since then we’ve predicted 200M+ protein structures & it's been used by over 1M researchers to advance disease understanding, accelerate drug discovery & much more eg:

→ View original post on X — @miketamir

Brain2Music: Reconstructing Music from Human Brain Activity

By

@miketamir

–

03 August 2023 21h01

Brain2Music: Reconstructing Music from Human Brain Activity https://
bit.ly/43D2Xuc https://
bit.ly/3Ovu1at

Optimized LLM Training Framework Reduces Host Memory Overhead

By

–

03 August 2023 19h29

If you use an optimized #LLM training framework like https://
pbase.ai/3DHqnE5, you can get the host memory overhead back down to a more reasonable 7 * 4 = 28 GiB of host memory even when training on multiple GPUs.

Loading Model Weights into GPU Memory with Ray Object Store

By

–

03 August 2023 19h29

How did we do it? By loading the model weights into memory once before training begins and inserting them as numpy arrays into the #Ray object store, we can then zero-copy read the weights directly from shared memory into each GPU worker process.

Ludwig simplifies distributed computing with Ray backend integration

By

–

03 August 2023 19h29

Doing this yourself would normally be a fair bit of cumbersome code, but in Ludwig you get it for free just by running with Ray as the backend runtime. Try it out for yourself: https://
pbase.ai/3qfet19

Loading Pretrained Checkpoints: Multi-GPU Memory Challenge

By

–

03 August 2023 19h29

Before you even get to multi-GPU training with model parallel frameworks like #Deepspeed, you need to load the pretrained checkpoint into memory. To make matters worse for machines with multiple GPUs, you need to load the checkpoint into host memory once for each GPU in your job!