🔥 Learn React in 10 tweets (with hooks) 👇.

And finally, when you’re ready to serve your LLM, check out vLLM:. - PyTorch or JAX.- GPU, TPU, or CPU.- caching support.- Multi-LoRA

Marin is a Stanford created, _actually_ open source LLM! 🔥. They do everything in the open: training recipes, data pipelines, mistakes and experiments are all on github. If you want to train your own LLM fully from scratch, definitely check this out!

Then if you want to train your own LLM based on a STOA open source model (Llama, Gemma, etc), check out MaxText!. It has all the latest pre and post training best practices built in. Custom LLM in a box!

Tunix (Tune-in-JAX) gives you a state of the art way to tune your own models. Out of the box LoRA, Q-LoRA, and RL algorithms for tuning LLMs to your specific needs!. This means you can fine tune your own LLM without all the hassle of coding all this up yourself 🎉

NNX is one library for creating neural networks in JAX, and now it’s going to finally use pytrees!. This uses the JAX experimental mutable arrays, which makes JAX feel more imperative

Matt Johnson (coauthor of JAX): “JAX gives you AGI!”…. - Awesome scalability.- Great expressiveness.- Intriguing portability

Roy Frostig: JAX lets you tradeoff between flexibility and control, which has resulted in a large ecosystem. Each JAX library represents a different tradeoff and use case

@random_forests planned the event (thanks! 🎉) and kicks us off

Learning about JAX and OpenXLA today at DevLab! . I’ll share interesting tidbits throughout the event 👇.

Excited to say I've switched roles at Google: I'm now DevRel for AI Frameworks! (JAX and Keras).

RT @kylekeesling: Do I have any Ruby friends out there that might be looking for a new opportunity?. I’ve been a team of one since the begi….

Light clouds made it a bit blurry, but still very neat!

RT @mitchhanberg: Help me reach my goal of 100 sponsors! Currently 43% of the way there!.

My latest project! He's definitely been keeping me busy.

Milky way 📸

Basement rock wall, stage 1: complete

On the latest episode, we chatted with @TimLeland about doubling MRR with his link shortener and how he's creating a new API for detecting malicious links: with @cgenco.

I haven't posted about it for a bit, but is still going strong 💪; @cgenco just got the first revenue for *the video clipper .com*! . 🤑 .

This was a great team - hard work right to the end!.

Job update: I got a job!. I'll be joining @Google as a Developer Relations Engineer for the Google Photos team. It's a big change after being independent for so long - but I'm looking forward to it! 🥳.