icyzh

Finally completed my research project vortex float32 data compressor in which i made a custom architecture to efficiently perform lossless compression of HEP float32 data pypi.org/project/vortex…

@notKartikk Depends

HUMANS ONLY: DO YOU HAVE FREE WILL? ◯ Yes ◯ No ◯ Depends

@shikhr_ One Model to prompt them all

One Model to rule them all One Model to find them One Model to bring them all And in the darkness bind them

@shikhr_ Us bro

Algorithm think I am dead. I need to post more.

been grinding on this architecture for 3 months,just completed wrapping it as a python library this week.Thanks to @lablabai, @AIatAMD for hooking me up with the MI300X compute to actually train this , the code is fully available at my github

Finally completed my research project vortex float32 data compressor in which i made a custom architecture to efficiently perform lossless compression of HEP float32 data pypi.org/project/vortex…

tradeoff rn is inference speed,wrote a custom arithmetic range coder in Triton, but decoding is still bottlenecked by the transformer forward pass. strictly for offline archival until i get to optimize this further.

the compression gains were pretty solid. squeezed out 28% better compression than LZMA-9 on ATLAS particle data. on CAMEL astrophysics data it still beat LZMA by 19.5%.

so i built custom TDTEmbedding. instead of forcing the model to use one giant lookup table, it gives each of the 4 bytes in a float32 its own independent embedding table

Ever noticed how standard zip compressors sucks while comparing raw scientific dataThat's because they treat all floating-point numbers like plain text. But a float32 isn't just random bytes it has a strict structure -> sign,exponent,mantissa

@shikhr_ you can try whitepapers of bitcoin,solana,etc to get a basic understanding.

I am thinking of learning web3 and crypto (not the scam part). Any pointers?

@swymbnsl Who?

Starting June 1st, GitHub Copilot will move to a usage-based billing model as GitHub Copilot supports more agentic and advanced workflows. In early May, you'll see a preview bill experience, giving visibility into projected costs before the transition. 👉 Read more about the upcoming change: github.blog/news-insights/…

Sony has introduced a new DRM policy for digital purchases on PS4 and PS5. Moving forward, any newly bought digital titles will require an online "check-in" with the PlayStation Network at least once every 30 days to validate ownership licenses.

OpenAI just released an open source model that finds and hides personal info in text. github.com/openai/privacy…

@buildwithsid arch-terminal-portfolio.vercel.app

Drop your portfolio (or your best website) I'm gonna rate it.

@sortedcord No,I scribble on excalidraw tho xD

ADHD so bad you started having LLM ahh chains of thought. I just spent 30 minutes in markdown talking to myself. Anyone else who does this? Is this normal behavior? 🫠🫠

@yxhwho @memoizedk 🥹🥹

@memoizedk bkl ne galat print krdia aur mujhe do bar pese dene pde kyunki jaldi me tha mkc uski vo softy vale se kabhi mt karana bkl hai bhut bda 200 rupee chudva die bkl ne uski mkc sbse bdia vo main vala hai jo OG hai

mera account practical files ke print krate krate hi khali ho jaega ek din.

@piyush784066 All of them

Which device do you run Linux on? -Laptop -Desktop -Server -Virtual machine

Been learning GPU programming for a while, this book is one of the most helpful resources I’ve found. arxiv.org/pdf/2410.05686

@ellen_in_sf literature review

Starting an AI Researcher group chat. The space is growing fast! Comment “literature review” to join.