context size vs speed in GGUF: anyone testing this?

[bytesupply76]

That matches what I was seeing. I ended up in about the same place using PyTorch. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.