Open dchaley opened 2 months ago
Each prediction job fetches the container. If we're going to fetch it anyhow … why not pre-fetch it by including it in the container?
One reason: a larger container would slow down each task that doesn't use it. Question: does image streaming help with that?
Question: is it Really That Bad if we have 1 container with the GPU stuff … and another without?
Each prediction job fetches the container. If we're going to fetch it anyhow … why not pre-fetch it by including it in the container?
One reason: a larger container would slow down each task that doesn't use it. Question: does image streaming help with that?
Question: is it Really That Bad if we have 1 container with the GPU stuff … and another without?