AI computing is a memory hog. An Nvidia-backed startup has an answer
- May 7
- 1 min read
WALL STREET JOURNAL — The startup runs a software tool that it says allows artificial-intelligence models to process queries more efficiently and at a lower cost, in part by better utilizing short-term memory to reduce the total amount of computing power needed to do calculations. Its open-source software engine, called SGLang, operates as a middle layer between the models and the hardware.
SGLang was originally part of LMSYS Org, Large Model Systems Organization, a nonprofit that also incubated Chatbot Arena and was eventually spun out into RadixArk.
Read the full story | WALL STREET JOURNAL
