The Mint

machine intelligence, learning, technology, silicon valley

[The Mint @themintsv]

@julien_c @huggingface 1 Exabyte = 1000 petabytes = 1 million terabytes

[The Mint @themintsv]

@hcompany_ai Hybrid with Internet access is probably the way to go.

Simple tasks: local
Stuck in a complicated task: Internet or cloud

[The Mint @themintsv]

@RemiFabreRobot Those antennas could be prettier @pollenrobotics

[The Mint @themintsv]

GenLIP: Generative Language-Image Pre-training

https://t.co/Lp2hXldwxW

https://t.co/Yu1TR7QCKd

[Ross Wightman @wightmanr]

I had a fun coding session at the end of last week.. I implemented NaFlexGenLIP. It's an impl of GenLIP using NaFlex style image tokenization and packing instead of the full NaViT style sequence packing. Prelim CC12M sanity training on my local RTX Pro 6000s is showing some signs of life even with such a small model and dataset🥳

I was going to do this as part of a new project but w/ recent OpenCLIP refactoring it was easy enough to bolt on there initially to get something that's ready for scale experiments sooner. I threw in some utils to calculate text sequence length and batch budget params based on dataset caption dist. Also hacked together a generative 'zero-shot' image classification idea based on likelihoods.

[The Mint @themintsv]

PolarQuant: Quantizing KV Caches with Polar Transformation (Google)

https://t.co/vzzoEPKCwA

[The Mint @themintsv]

RaBitQ: Quantizing High-Dimensional Vectors with a Theoretical Error Bound for Approximate Nearest Neighbor Search

https://t.co/KDRn7C9CCw

[turbopuffer @turbopuffer]

tpuf quantizes vectors to improve perf (RaBitQ)

the algo randomly rotates vectors, and we were using matmul at O(d²) space & time, brutal at high dims. 10k = 400MB in RAM!

we rebuilt the rotation using FWHT at O(d) space & O(d log d) time. ~no recall loss, 10k = only 5kB in RAM

[The Mint @themintsv]

Beyond the TurboQuant-RaBitQ Debate: Why Vector Quantization Matters for AI Infrastructure Costs

https://t.co/IS4DTVnjBj

[The Mint @themintsv]

@n0riskn0r3ward I agree.

Based on my years of experience training and shipping embedding models, the quality of training data for embedding models is very important.

[The Mint @themintsv]

@n0riskn0r3ward They are doing it!
I get a whole screen every now and then, trying to convince me to subscribe...

[The Mint @themintsv]

@julien_c @huggingface That sounds insane.

Is it a real copy or lazy copy (you incur the cost during the access, or you do not copy anything, you just access it, in which case cross-region access becomes a bottleneck).

[The Mint @themintsv]

@ndea The paper:

Recursive Program Synthesis

Authors: Aws Albarghouthi, Sumit Gulwani, Zachary Kincaid

University of Toronto, Microsoft Research

https://t.co/su2XJapLiG

[The Mint @themintsv]

@poteto Feels like people have been putting too much effort to develop these (tools--including cursor--, skills, harnesses, etc.).
They will become obsolete soon (they will be absorbed or fixed by the AI companies).

[The Mint @themintsv]

AI Agents Glossary

Commonly used new terms in the world of AI Agents: agent, harness, scaffold, policy, skills, tool use,...

https://t.co/8DOBIfvYiC

[The Mint @themintsv]

@fermatslibrary And now, considered a heavy metal for human body by MDs, and should be avoided & detoxed.

[The Mint @themintsv]

@evelovesolive You should also invest a little bit in your website and graphics; they look like from 20 years ago.

[The Mint @themintsv]

@andrewgwils It is actually a crabapple. Cherry trees are significantly different.

[The Mint @themintsv]

@fchollet Was curious about PyTorch, asked Grok: 85.4 million monthly downloads. Posting, in case someone is interested.