Performance
ULL trading firms go to a lot of trouble to get their servers and switches within the same buildings as the exchanges they trade with to reduce latency. Some firms don’t even use layer 1 switches to be competitive.
My kernels go 2x faster than MKL for matrices that fit in L2 cache, which makes them a work in progress, since the speedup works best for prompts having fewer than 1,000 tokens.
https://news.ycombinator.com/item?id=37205053 > For people (like me) who are wondering why a kernel needs to boot in under 28ms: It's for virtual machines that get launched on-demand in services like AWS Lambda. https://www.daemonology.net/blog/2022-10-18-FreeBSD-Firecrac...
A Java version of simdjson - a JSON parser using SIMD instructions, based on the paper Parsing Gigabytes of JSON per Second by Geoff Langdale and Daniel Lemire.