Hello,
first of all thank you for your results and the code that you shared.
I have a question concerning the appendix B:Datasets combination. In that paragraph you explore how dataset expansion improves the model performance. The comparison is done to a 'baseline' results, obtained from default datasets. The question is about the model configuration that you used in these studies:
- was the model pertained on HowTo100M?
- which experts did you use?
- do you have 'baseline' results for MDMMT with configuration where it was not pertained, but all three experts were used?
Kind regards,
Vlad
Hello,
first of all thank you for your results and the code that you shared.
I have a question concerning the appendix B:Datasets combination. In that paragraph you explore how dataset expansion improves the model performance. The comparison is done to a 'baseline' results, obtained from default datasets. The question is about the model configuration that you used in these studies:
Kind regards,
Vlad