Stavros Kassinos (@kassinoss) 's Twitter Profile
Stavros Kassinos

@kassinoss

Prof @ UofCyprus | @Stanford: Fear the🌲| @UTAustin: Hook them🤘🏻| Computational science with impact | Amazed by Nature’s innate wisdom/beauty | My views only

ID: 1680178574

linkhttp://www.ucy-compsci.org calendar_today18-08-2013 08:10:57

1,1K Tweet

1,1K Takipçi

354 Takip Edilen

Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

Annular fluid flow w/ plug inlet looks deceptively simple, but steep gradients at the inlet are a PINN’s worst nightmare! Check out our preview results—proving that 'simple' cases are fertile ground for uncovering new tricks. Solved w/ 🚀 MLX on an M2U 🚀 & managing in the end

Awni Hannun (@awnihannun) 's Twitter Profile Photo

We’re hiring someone to work with us on MLX at Apple. Join the most cracked team to advance the frontier of ML and systems. DMs open.

We’re hiring someone to work with us on MLX at Apple. Join the most cracked team to advance the frontier of ML and systems. DMs open.
Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

🦎MLX-Kourkoutas🦎: good progress 🚀 Now works seamlessly with mlx.optimizers and mx.compile. Additional benchmarks against Adam look good (same # epochs, similar final loss). 🚧🔧🏗️working on introducing regularizations where needed Awni Hannun Angelos Katharopoulos

🦎MLX-Kourkoutas🦎: good progress 🚀 Now works seamlessly with mlx.optimizers and mx.compile. Additional benchmarks against Adam look good (same # epochs, similar final loss).  🚧🔧🏗️working on introducing regularizations where needed <a href="/awnihannun/">Awni Hannun</a> <a href="/angeloskath/">Angelos Katharopoulos</a>
Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

Wow 🤩: MLX-Kourkoutas beating the grand daddy of them all optimizers is not a small feat 🚀. Solving the 2D Heat Conduction Equation using PINNs and with MLX-Kourkoutas or Adam. Awni Hannun Angelos Katharopoulos

Wow 🤩: MLX-Kourkoutas beating the grand daddy of them all optimizers is not a small feat 🚀. Solving the 2D Heat Conduction Equation using PINNs and with MLX-Kourkoutas or Adam. <a href="/awnihannun/">Awni Hannun</a> <a href="/angeloskath/">Angelos Katharopoulos</a>
Awni Hannun (@awnihannun) 's Twitter Profile Photo

Top 5 reasons to work on MLX: - Small exceptional team - Fully open source with an amazing community - Push the limits of local AI at the best company in the world do it - Get to work on the whole ML stack - 🚀

Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

🦎🦎 It’s cooking … a small preview of my thinking about it. The full write-up once ablation testing is complete and this holds to the promise. Awni Hannun

🦎🦎 It’s cooking … a small preview of my thinking about it. The full write-up once ablation testing is complete and this holds to the promise. <a href="/awnihannun/">Awni Hannun</a>
Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

I’m sooo 🦎🦎happy 🦎🦎w/ Kourkoutas-β. Completed 1st full test on training a heavy Transformer model as PDE surrogate in data-driven mode. Kourkoutas-β blows vanilla Adam out of the desert, especially for smaller training datasets & quantized models where the grads tend to be

I’m sooo 🦎🦎happy 🦎🦎w/ Kourkoutas-β. Completed 1st full test on training a heavy Transformer model as PDE surrogate in data-driven mode. Kourkoutas-β blows vanilla Adam out of the desert, especially for smaller training datasets &amp; quantized models where the grads tend to be
Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

Kourkoutas is a desert lizard. The secret behind Kourkoutas-β success is the “sunspike”: high 🌞 —>🦎 runs; low 🌞—> 🦎 sluggish. This is how “sunspike” 🌞 evolves during training a quantized Transformer to learn the evolution of the heat conduction PDE. It’s a dynamic per

Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

“Beyond Language: Applying MLX Transformers to Engineering Physics” is now accepted and freely accessible at: sciencedirect.com/science/articl… The most interesting part actually was added during revisions. The Transformer was trained on a normalized temperature range & a limited number

“Beyond Language: Applying MLX Transformers to Engineering Physics” is now accepted and freely accessible at:  sciencedirect.com/science/articl…

The most interesting part actually was added during revisions. The Transformer was trained on a normalized temperature range &amp; a limited number
Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

As useful and fast as o3 is, and I do like it, I’m finding that it is still no match to o1-pro. o3 is more prone to occasionally make up things or offer loose interpretations compared to o1-pro — something that o1-pro almost never does. In fact, o1-pro is so consistently

Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

The new Engineering School complex at the University of Cyprus University of Cyprus is finally a reality. Stunning architecture in every step you take. Our beautiful new home will bring all the Engineering Departments under a common roof and will strengthen collaboration across

The new Engineering School complex at the University of Cyprus <a href="/UCYOfficial/">University of Cyprus</a> is finally a reality. Stunning architecture in every step you take. Our beautiful new home will bring all the Engineering Departments under a common roof and will strengthen collaboration across
Stavros Kassinos (@kassinoss) 's Twitter Profile Photo

🥳🚀 1st installment of Mac Studios M4 Max for the “HPC+AI Synergies” teaching lab has arrived. 🚀🚀 Benefits: 👉🏻Ultra low energy footprint 👉🏻 Quiet student-friendly environment for hands on experimentation w/ parallel connectivity topologies 👉🏻Students get trained on machines

🥳🚀 1st installment of Mac Studios M4 Max for the “HPC+AI Synergies” teaching lab has arrived. 🚀🚀
Benefits:
👉🏻Ultra low energy footprint 
👉🏻 Quiet student-friendly environment for hands on experimentation w/ parallel connectivity topologies
👉🏻Students get trained on machines