Thanks to everyone who came out for last night's NYC meetup! We had a blast demoing AI Hub and how to deploy DeepSeek-R1 securely on our cloud or your own VPC—at a fraction of the cost. If you want secure, scalable AI solutions that allow you to move fast and control your data

Great work, I’ve been wondering whether something like this would have worked for a while 🔥 Also great to see the code evolved out of LitGPT github.com/Lightning-AI/l…, I’m finding this is a pretty common pattern. Working off a simple codebase is key to pushing the boundaries 🚀

Was fun chatting about all things AI with Dave Briggs and Cheddar from the NYSE 🏛

Looking good 😄

Here I show you how to finetune and deploy DeepSeek R1 (8B) for < $1.00 in 8 minutes using the AI Hub from Lightning AI ⚡️ ⚡️⚡️

Super simple finetuning DeepSeek R1 8B on a reasoning dataset + getting a private endpoint out of it. Just a few easy steps on the Lightning AI ⚡️ AI Hub. This is only the appetizer : )

How can we use small LLMs to shift more AI workloads onto our laptops and phones? In our paper and open-source code, we pair on-device LLMs (ollama) with frontier LLMs in the cloud (@openai, @together), to solve token-intensive workloads on your 💻 at 17.5% of the cloud cost

Simone Angarano Great job with MLA (Multi-head Latent Attention, DeepSeekv2) github.com/Lightning-AI/l… DeepSeek. I just found it, and I hope that it gets merged in official litgpt asap by Lightning AI ⚡️ team, so we can try it out. 👏

Lost my bet for a couple of months it seems :)

Sometimes I have no idea what I’m doing. It’s almost 2 AM, and here I am, enjoying thunder sessions. 😅⚡ #lightningthunder #thundersessions

🤦‍♂️

Optimize your model training with smarter memory management! Check out our latest blog post to learn how PyTorch’s new activation checkpointing techniques—featuring Selective Activation Checkpointing and the Memory Budget API—help you balance speed and memory during model

Sometimes, all you need is good tracing—like finding your lost TV remote! 📍🔍 Learning tracing with LitData! ⚡🚀🚀🚀 #Optimizations #LitData #Debugging

Today PyTorch Lightning surpassed 200M downloads 😳 And for the first time ever, people use PyTorch Lightning just as much as Keras! We’re not backed by Google - but instead a world-class community of the best open-source engineers in the world. Lightning makes training ANY

This is a killer episode, I thoroughly enjoyed listening to it this morning. Big kudos to Ivan Zhao. As an aside: being the nerd I am :-), I loved the references to Smalltalk on the Alto system and I can see how this permeates Notion philosophically

It'll be interesting to see how advertising will evolve as vision-enabled agents start crawling the web in real time to carry out their tasks. That's a can of worms right there, waiting for someone to open it.

🚨 TorchServe is no longer actively maintained! Looking for a high-performance alternative with a similar API? 🚀 LitServe is your best option — faster, more efficient, and easy to switch to. 🔧 Benchmarks in thread 🧵 #MLOps #PyTorch #AI #DeepLearning

Thank you Claude 🧀

[3/17] minions daily ship 🚢 ⚡ minions now live on Lightning AI ⚡️! — access via Dan Biderman’s studio app—big s/o to Luca Antiga ⚡️ + William Falcon ⚡️ for the support and hacker energy this am 🎉 also—500 ⭐️'s on github! y'all are the best 💛

🚀 Join Luca Antiga ⚡️ & Thomas Viehmann for their #GTC25 session: Make My PyTorch Model Fast, and Show Me How You Did It. Learn how Thunder - the PyTorch-to-Python compiler - optimizes models with ease and without changing the PyTorch code! ⚡️ 11:00 AM - 11:40 AM PDT ⚡️ San Jose