AIQB
TutorialsOrdinary

Why the KV cache, not the weights, decides how many users an LLM server can hold

Source: DEV Community·

Summary

A 7B model in fp16 takes 13.5 GB of GPU memory for its weights. On an 80 GB card that leaves about 66...
TierOrdinary
Published
Indexed by AIQB
SourceDEV Community
AIQB record IDintel-97460d75c9de7ae5e37f0f10