Hi Team,
Thanks for sharing your work with the research community. The idea of making agents communicate entirely in latent spaces is interesting!
I have a quick, small question. How do you compute the token counts for the experiments? I noticed the logs do not print the token count like they do for accuracy and speed, which makes me think you might have a script you use to do that. If you do, would you mind sharing it or expound on the logic you used to compute it?
Thank you very much. Looking forward to hearing from you.
Best,
Mackay
Hi Team,
Thanks for sharing your work with the research community. The idea of making agents communicate entirely in latent spaces is interesting!
I have a quick, small question. How do you compute the token counts for the experiments? I noticed the logs do not print the token count like they do for accuracy and speed, which makes me think you might have a script you use to do that. If you do, would you mind sharing it or expound on the logic you used to compute it?
Thank you very much. Looking forward to hearing from you.
Best,
Mackay