Shannon Shen
Shannon Shen
Thanks for your questions! I'll get back to them later this week!
> Q1: I noticed that your code appears to work only when the two models use the same tokenizer. Does the current version support scenarios where the tokenizers differ, such...
> Furthermore, the optimal_deferral_threshold identified by find_optimal_deferral_threshold does not represent the final optimized value η . Instead, the true value of η is determined by the optimal result of a...
Ah which models you are running with? I remember that I did manually adjust some vocab_size params earlier (which should have been fixed). Thanks!
Thanks! I think that's also a good idea for a different way of implementing this. From my perspective event triggers are mostly for pipelining actions and might be less suitable...