← Back to context Comment by kleiba2 3 days ago What actually is "scaling post-training"? 6 comments kleiba2 Reply FergusArgyll 3 days ago More RLVR. Give it verifiable problems, if it doesn't find a solution move on, if it does, use that as a reward signal. Gecko4072 3 days ago Can’t this be extended quite far? Use a cerebras-served model, use verification techniques to generate and solve millions of problems and then use that as training? kevincox 3 days ago This isn't latency bound, it is trivially parallelize. So you want to run it on the most efficient compute you have, not the fastest. 2 replies → gvkhna 3 days ago That’s the whole point, just cost and compute limitations in your way (mostly).
FergusArgyll 3 days ago More RLVR. Give it verifiable problems, if it doesn't find a solution move on, if it does, use that as a reward signal. Gecko4072 3 days ago Can’t this be extended quite far? Use a cerebras-served model, use verification techniques to generate and solve millions of problems and then use that as training? kevincox 3 days ago This isn't latency bound, it is trivially parallelize. So you want to run it on the most efficient compute you have, not the fastest. 2 replies → gvkhna 3 days ago That’s the whole point, just cost and compute limitations in your way (mostly).
Gecko4072 3 days ago Can’t this be extended quite far? Use a cerebras-served model, use verification techniques to generate and solve millions of problems and then use that as training? kevincox 3 days ago This isn't latency bound, it is trivially parallelize. So you want to run it on the most efficient compute you have, not the fastest. 2 replies → gvkhna 3 days ago That’s the whole point, just cost and compute limitations in your way (mostly).
kevincox 3 days ago This isn't latency bound, it is trivially parallelize. So you want to run it on the most efficient compute you have, not the fastest. 2 replies →
More RLVR. Give it verifiable problems, if it doesn't find a solution move on, if it does, use that as a reward signal.
Can’t this be extended quite far? Use a cerebras-served model, use verification techniques to generate and solve millions of problems and then use that as training?
This isn't latency bound, it is trivially parallelize. So you want to run it on the most efficient compute you have, not the fastest.
2 replies →
That’s the whole point, just cost and compute limitations in your way (mostly).