Arpan

founder me is better than employed me

quantization: explain like i'm five

Read STAIR

“You’re closer than you think” is a delusional person’s best friend.

chunked prefill & prefix caching: explain like i'm five

Things Men Go Over Budget On

single-gpu optimization: explain like i'm five

optimize one gpu: explain like i'm five

🚢⛴️🛳️⚓️

Looking for someone with expertise in Substack growth strategies and publication setup. If you’ve helped grow a Substack publication, or know someone who has, let’s connect!

I think AGI may be written in Python, but it will run on Rust.

Pls learn system & db design

pagedattention: explain like i'm five

continuous batching: explain like i'm five

moving to vllm: explain like i'm five

serving engines: explain like i'm five

break one gpu: explain like i'm five

ttft vs itl: explain like i'm five

measuring inference: explain like i'm five

kv cache: explain like i'm five

one token: explain like i'm five