Question about how to train across several datasets

Luodian / Otter

🦦 Otter, a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo), trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability.

https://otter-ntu.github.io/

MIT License

3.55k stars 242 forks source link

Question about how to train across several datasets #226

Closed Darren-greenhand closed 1 year ago

Darren-greenhand commented 1 year ago

Hi! We want to reproduce the training pipeline but we are not sure how you train otter on several datsets Did you run the training code once for each dataset? thus one time for LACONV, one time for LAi2i And if so, did you reset the state of optimizer and scheduler every time?

thx !

Maxlinn commented 1 year ago

i noticed in issue https://github.com/Luodian/Otter/issues/149#issuecomment-1602517362

the author seperated different dataset names with comma.

Darren-greenhand commented 1 year ago

i noticed in issue #149 (comment)

the author seperated different dataset names with comma.

thank you！It helps a lot 😆