@abhagsain@netcup No it isn't that good, it vps, not root server like hetzner, even the dedicated is just a vps, and bandwidth will be rate limited when past 24 hour traffic is above 2 tb
@ash_twtz It is not that they are hiding, but model itself can't infer what model they are, without system prompt telling them what they are, which the web harness doesn't do
@consyin@thsottiaux Apearantly he said pst not pdt, even tho he does it on pdt, pst is 1 hour later apearantly, that gives him 34 more minutes to his deadline
@FixlationAI@chetaslua Anthropic models say it's qwen when you call by api, and ask in Mandarin, everyone uses everyone's synthetic data π₯, Besides you can't Distil to perform better than teacher, k3 was first model to have incredible 3d spatial and visual understanding
@FixlationAI@chetaslua If 3rd party can distill mode to make anything cheaper and faster, might I say better in some ways, why can't first party do it bro π₯π₯π₯, everyone uses everyone's synthetic data