bah quand tu peux acheter 4 roues pour un mac pro a 1500€, ça va
COMPUTING
-
GPT-2: Perfect Introduction to LLMs and Distributed Training
By
–
Exactly as intended! GPT-2 is a beautiful "hello world" to LLMs but also distributed training etc.
-
Cerebras Achieves 180x Acceleration in Molecular Dynamics Simulations
By
–
Cerebras is a 2024 Gordon Bell Finalist! In collaboration with Sandia National Laboratories, Lawrence Livermore National Laboratory, and Los Alamos National Laboratory, we have achieved an astounding 180x acceleration in time-to-solution for molecular dynamics simulations
-
Chiplets and UCIe Power Future Automotive AI Computing
By
–
#Chiplets & @UCIexpress are powering the future of automotive computing. #AI #compute now rivals data centers. Chiplets are the new automotive norm. #UCIe enables seamless integration, paving the way for mass self-driving. Read more in our blog: https://
untether.ai/the-need-for-c
hiplets-and-ucie-in-automotive/
… -
Mobile API Complexity and Efficient Model Execution Challenges
By
–
That would be a much more complicated and less flexible API to sort out, but I wouldn’t be surprised if they wind up going that way, years later than they could have just offered camera access. Efficient model execution on mobile is a real challenge, regardless of whether it is
-
Turing’s Paper Discussion on Our Opinions Are Correct
By
–
@alexhanna and I had a blast digging into Turing's paper on the Our Opinions Are Correct podcat with Charlie Jane and Annalee:
-
Future of Fintech: Autonomous Ecosystems and Digital Currencies
By
–
Discover the future of #fintech! Explore transformative #trends like autonomous #financial ecosystems, universal #digital #currencies, and the #virtual #economy. https://
youtube.com/watch?v=aechc_
h4jvQ
… #Fintech #FutureFinance #DigitalCurrency #TechTrends #QuantumComputing -
Training Large Sparse Modular Models Across Distributed Data Centers
By
–
Some really nice work by @Ar_Douillard and many coauthors on how to efficiently train large, sparse, modular models across many data centers that are geographically distributed. As @fouriergalois pointed out in the replies, this is a step in a longer journey:
-
Groq Recognized as Critical Technology for National Security and Defense
By
–
It’s an honor to have Groq recognized as critical technology in support of our Nation’s Security and Defense! @SVDG_official
, IQT (In-Q-Tel), JPMorganChase and many others for this recognition. See the full report here -
Estimating compute improvements from 2019 to 2024
By
–
This information was never released but I'd expect it was a lot more. In terms of multipliers let's say 3X from data, 2X from hardware utilization, in 2019 this was probably a V100 cluster (~100 fp16 TFLOPS), down from H100 (~1,000), so that's ~10X. Very roughly let's say ~100X