09-14-2026, 02:22 PM
subscribed, curious where this lands.
measure twice, benchmark three times
|
student models for local use
|
|
09-14-2026, 04:07 PM
I tested this tonight. On mini PC with SGLang, I saw basically the same behavior. I expected the model choice to matter more than it did.
09-14-2026, 08:12 PM
I hit almost the same issue last night.
09-17-2026, 08:34 PM
What exact setup are you running? Are you testing this on dual 3090 or something else?
09-17-2026, 08:50 PM
Can you fill in a bit more context? What model, what runtime, and roughly what context size?
09-17-2026, 09:48 PM
A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Mistral and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
|
|
« Next Oldest | Next Newest »
|