Excited to announce the general availability of L40 GPUs on Lightning Studios! ✅ As low as $1.20 for 1 GPU ✅ 48 GB VRAM 🤯 ✅ 1, 4 and 8 GPU instances Honestly becoming my favorite GPU now over the A10G! Start a Studio today to see it in action yourself:

I'm in Mexico City this week and we're bringing together the latino AI community (This Thursday 12 sept!) Share with your friends in CDMX! lu.ma/LAI-takesover-…

Everyone is up in arms over missed benchmarks or demos falling short of the hype... well yes... it’s because AI research is happening on social media now, skipping peer reviews and lab scrutiny. phd students leak data pipelines all the time and falsely "solve" things lol

Asking AI startups about their revenue:

If you watched the thunder session today, there was a bug that I gave up on debugging live. Turns out that it was just a bit too early: After adding also the output shape to the print I added last, it was obvious where the bug was and then also what was wrong... Classic

Humbled at how well LitServe has been received ☺️. LitServe augments FastAPI to dramatically simplify serving for AI models with batching, streaming, GPU auto-detection and more! Try it now github.com/Lightning-AI/L…

One example that blows my mind 🤯 with LitServe is how easy RAG systems become when you can merge the LLM, embedding model, DB and more in the same server.... Check it out: github.com/Lightning-AI/l…

Deploy a private multi-modal API with the new Pixtral model in <30 lines of code 🤯🤯 lightning.ai/lightning-ai/s…

Batching is a simple way to speed up model serving. Add batching to your FastAPI server easily with LitServe 🚀🚀🚀 Batching groups requests and runs them through the server at once instead of sequentially. github.com/Lightning-AI/l…

Right now Sebastian Raschka of Lightning AI ⚡️ keynotes on navigating the architectural timeline of LLMs #PyTorchConf

I really had a ton of fun at this year’s ⁦PyTorch⁩ Conference Thanks to the organizers + everyone who showed up and made it so engaging Here’s a picture for the records: Sebastian Raschka⁩ and ⁦ptrblck⁩ meeting irl for the first time, both mutually starstruck haha 😊

I’ll be in Madison this week! come say hi. Anyone I should meet? 😊

Llama 3.2 came out 6 hours ago and you can already deploy a private, serverless API on Lightning 🤯 🤯 lightning.ai/lightning-ai/s…

Who's already tried Llama 3.2 🤯🤯? If you haven't, run your own private API here for free: lightning.ai/lightning-ai/s… Congrats to Meta friends for a great Llama release! AI at Meta Soumith Chintala Ahmad Al-Dahle Joseph Spisak

Thanks for having me over! The OSS center at UW-Madison looks amazing and Madison itself is so quaint and beautiful 🤯

💻 Really enjoyed talking about open-source and its impact on AI last night at the Data Science Institute in UW–Madison! Panel session and lightning demos with William Falcon ⚡️ (Lightning AI ⚡️) and Dmitry Petrov (@iterativeai), hosted by Kyle Cranmer and Nhi Lê (Alpha Intelligence

Serving a private Llama 3.2 RAG API is now easier than ever! You can deploy a RAG application behind a private API using LitServe in a framework-agnostic and fully customisable setup. We’ve used vLLM for efficient LLM serving and a self-hosted, high-performance

Gave a talk yesterda at UW–Madison where Kyle Cranmer and team are building a great OSS center to get more students building opensource ⚡️ ⚡️ Bonus ⚡️, got to hang with 👩‍💻 Paige Bailey from the Gemini team, Dmitry from DVC and Alpha Intelligence Captial Fund I

I've been saying vLLM and LitServe are complementary tools! Check out this RAG example for Llama 3.2B with vLLM and LitServe for serving! Github: github.com/Lightning-AI/l…