|
quant choices for 70B local
|
10-19-2026, 03:34 AM
Good point. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.
2x3090 • llama.cpp • too much VRAM is still not enough
10-19-2026, 03:40 AM
Can you fill in a bit more context? Is this local-only or are you routing anything externally?
2x3090 • llama.cpp • too much VRAM is still not enough
10-19-2026, 03:53 AM
Did a small comparison this afternoon. On Threadripper box with TensorRT-LLM, I saw basically the same behavior. The main thing that moved was latency, not quality.
10-19-2026, 04:24 AM
What exact setup are you running? What model, what runtime, and roughly what context size?
10-19-2026, 05:16 AM
What exact setup are you running? Do you see the same thing with a smaller batch or shorter prompt?
2x3090 • llama.cpp • too much VRAM is still not enough
|
|
« Next Oldest | Next Newest »
|
