On-device ML Engineer | 🤖Passionate about reverse-engineering neural nets | 🚀Optimizing large models for the edge 💻📱fguzman82.github.io ColombiaJoined May 2011
This one paper might kill the LLM agent hype.
NVIDIA just published a blueprint for agentic AI powered by Small Language Models.
And it makes a scary amount of sense.
Here’s the full breakdown:
🧵 @karpathy dropped the most compelling lecture at the AI startup school :) Here's my notes summarized with Opus.
Main thesis: We're not in the "year of agents" — we're in the DECADE of agents. Here's the historical arc that explains why...
==== The Software Evolution Story…
Luminal can discover flash attention entirely automatically.
We've been working towards this north star in our search compiler. Check out the prototype demo below ↓
Vercel AI Gateway (alpha):
• Built on the @aisdk 5 alpha
• Switch between ~100 AI models without API keys
• Handles auth, usage tracking, and more
vercel.com/blog/ai-gateway
A major mistake I made in my undergrad is that I focused way too much on mathematical lens of computing - computability, decidability, asymptotic complexity etc. And too little on physical lens - energy/heat of state change, data locality, parallelism, computer architecture. The…
BOOOOM: Today I'm dropping TINY AGENTS
the 50 lines of code Agent in Javascript 🔥
I spent the last few weeks working on this, so I hope you will like it.
I've been diving into MCP (Model Context Protocol) to understand what the hype was all about.
It is fairly simple, but…
Interesting new paper from NVIDIA on "FFN Fusion"
arxiv.org/pdf/2503.18908…
They achieved 1.71× faster inference for Llama-3.1-405B fused to 235B params while reduced KV cache x2
Sometimes you're just exploring. What if your next favorite spot is right around the corner, and you don't know it. Get the perfect nudge at the perfect time.
Taking a photo can do more than capture your best memories. It can remind you to share them with your favorite people. It’s the only camera you’ll ever want to use.
🚀 Day 1 of #OpenSourceWeek: FlashMLA
Honored to share FlashMLA - our efficient MLA decoding kernel for Hopper GPUs, optimized for variable-length sequences and now in production.
✅ BF16 support
✅ Paged KV cache (block size 64)
⚡ 3000 GB/s memory-bound & 580 TFLOPS…
Okay, SigLIP 2 weights for OpenCLIP and timm (image encoder only) are on the @huggingface hub. Merged to main, release probably this weekend. I tested the IN-1k zero-shot and these are the OpenCLIP numbers:
B/32 256 top1: 73.9 top5: 93.4
B/16 224 top1: 78.4 top5: 95.7
B/16 256…
SigLIP 2 is the most powerful image-text encoder
you can use it to do
> image-to-image search
> text-to-image-search
> image-to-text search
> image classification with open-ended classes
> train vision language models
we will show you how to do all this week 🤝
SigLIP 2 is out! 👀 Another day, another model:
🥳Strong vision-language encoder
🔥Outperforms across sizes
📏Dynamic resolutions
🤏86M to 1B
huggingface.co/blog/siglip2
15K Followers 9K FollowingIndependent App Intents/Apple Intelligence expert & consultant. Team at Workflow before @Apple’s Shortcuts. Making apps at @suprcmptr. Get my shortcuts 👇
128 Followers 380 FollowingCo-founder/CEO @RunLocalAI (YC S24). Making it easier to ship better on-device AI. Also into house/techno from early 90's/00's,
5K Followers 8K Followinggeek, entrepreneur, 'I strictly color outside the lines!', opinions r my own indeed. @ayirpelle , universal handle at this time
409 Followers 3K FollowingEntrepreneur, iOS and AI developer. Photography amateur. Learning to see the light and shadow. Making pictures of beautiful and meaningful captured photons.
7 Followers 94 Followingim 30 and i cant believe how corrupt and misused the goverment its like we forgot that we are all still human not gods trying to leave merrier island fl
5K Followers 895 FollowingPartner @SteptoeLLP. Emerging technology and national security law. AI | Chips | FinTech | Crypto. Not legal advice. Opinions are my own.
54K Followers 979 FollowingTeaches math to engineers: https://t.co/TJ5i3Pg678
Professor @UW researching #MachineLearning for #Dynamics and #Control, especially for #FluidDynamics.
128 Followers 380 FollowingCo-founder/CEO @RunLocalAI (YC S24). Making it easier to ship better on-device AI. Also into house/techno from early 90's/00's,
5K Followers 8K Followinggeek, entrepreneur, 'I strictly color outside the lines!', opinions r my own indeed. @ayirpelle , universal handle at this time
2K Followers 56 FollowingAxolotl is the premier open source LLM fine tuning framework. find us on discord https://t.co/wlcE2wlJa9 or email us at [email protected]