On this day: Jan 18
2 records from this date, 2021 to 2024, gathered year by year — the most recent year first.
2024
2 years ago · 1 recordDistServeUCSD, PKU (Zhong et al.)
Splits prefill and decode into separate GPU pools so that latency metrics can be tuned independently
2021
5 years ago · 1 recordMicrosoft and UC Merced present ZeRO-OffloadJie Ren et al. (Microsoft, UC Merced)
The start of the lineage that lowered the hardware barrier to entry for training large models