New here, sort of. I'm Tanmay - I build things and occasionally post about them. Right now: a mini LLM inference engine from scratch. Before this: fine-tuned policies that ran on a humanoid (that's him, on his chair). Posting the wins, the bugs and the confusion. Here for the CUDA people, the inference nerds and anyone who learns by breaking stuff. Bangalore soon.
Kinda pissed at myself lately. So time for a challenge... Gonna finish this bad boy start to end within this weekend no matter what.
Will report back!!
Thanks for the amazing share. This was a genuinely good watch. Y'all should also watch this if you are into inference and looking for a good talk about it. It's basically a map of the field you're learning.
@official_cumi Bro, $500 a week and you couldn't even drop a link to the company? You want my CV, my GitHub and my previous projects, but I can't get a website to see who I'm sending them to. Easy pass, I'll wait for the version of this post with an actual company in it.
We fantasize the stars but forget that darkness makes them visible.
We dream about a life without sufferings but forget things become meaningful only once we have experienced their absence.
What makes AI start thinking better? Enter Reinforcement Learning.
Exploring how RL helps LLMs improve their reasoning, one experiment at a time.
This is just the beginning. Deeper dives, mathematics, more implementations are coming soon.
Stay curious!🚀 https://t.co/YOcyiWtP85