Search
Have an existing account?
Sign In
Tag: GPU inference
Perplexity reuses LLM kernels to keep embedding serving cheap
Engineers at Perplexity detail a GPU stack where batch and online embedding workloads borrow prefill and decode paths.