1 – 1 of 1 jobs
Inference has become a networking problem. With disaggregated serving, where prefill runs on one set of GPUs, decode on another, and KV cache streams between them, performance depends on how fast you can move tensors between GPUs,
Recommended jobs for you
