There are so many people wanting to give out GPU hours to researchers, do not feel limited by your compute access in academia or your personal wallet. What people struggle with is mostly how to ask for compute for research purposes. Even if your long term goal requires large

Introducing SkyRL tx 🧸, an open-source project to implement the Tinker API. The SkyRL team is excited about the Tinker API and the opportunities of using a single canonical interface that unifies training and inference. SkyRL tx lets you run a Tinker-like service locally today,

deleted huggingface cache 79 GB memory freed 😰

INFINITASIUM and me worked on Aider-polyglot and Uncertainty Bench environments. Two environments merged to prime-environments🥳

Anthropic -> Banglore Here We Go🎉

the more you notice, the more you find

Can't agree more! profiler is a must. was working on the data parallel naive and bucket versions and the profiler was super helpful to understand and clarify things I didn't understand when I first looked at the codebase.

us-east-1 is down Here is why:

NO CLAUDE I SAID USE UV

how me and INFINITASIUM camp in prime intellects bounty sheet for RL environments while will brown is revamping the spreadsheet.

I've had a lot of learning from sandboxes to exploring codebases after I started contributing 🙏 Amazing community, collaborative and super helpful.

Imagine losing first authorship because you got hit by a blue shell on the last lap 💀

small contribution to docs of sglang :) looking forward to contribute to the actual codebase.

the real agi competition is between vllm and sglang

[ENG SUB] how it feels to use eager pytorch in 2025

how it feels to be 20 and still not employed at a cool place

> INFINITASIUM and me worked on 2 safety benchmark environments. > Wildjailbreak and CoCoNot environment by Ai2 merged into prime-environments 🥳

another one : ) merged

the state of evals tells us how early we are in developing any solid understanding of these systems