Closed lixuechenAlfred closed 5 days ago
Hi! I see the problem, I did a quick rerun on one dataset for now and this seem to matter. Working on a fix and an update right now.
Hi! We've fixed the trompt implementation and rerun it an all datasets. The performance improved, we've also updated the paper with these new results for trompt
Position: Model class in bin/trompt.py Here, you only create one TromptCell for all cycles.