generalization performance

[lebron123zc]

Hello. We have collected 80 datasets using the psi‑0 shaking manipulation method and successfully completed fine‑tuning and deployment. I would like to improve generalization performance. Is it feasible to continue fine‑tuning for similar tasks based on the first fine‑tuned model? Are there any other approaches? Thank you!

[songlin]

Glad to hear you've gone through the whole pipeline. I would prefer diversity over quantity.
By the way, we are going to release a new checkpoint for psi0 next week; stay tuned.

[lebron123zc]

Glad to hear you've gone through the whole pipeline. I would prefer diversity over quantity. By the way, we are going to release a new checkpoint for psi0 next week; stay tuned.

thanks