mees / calvin

CALVIN - A benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Tasks
http://calvin.cs.uni-freiburg.de
MIT License
409 stars 60 forks source link

What is the exact language model weights/name used to train the MCIL baseline? #78

Closed the-nishant closed 6 months ago

the-nishant commented 6 months ago

I want to try different language annotations other than the validation annotations but do not want to retrain the MCIL baseline model. Would it be possible to know the full name of the MiniLM used to train the MCIL baseline?

Also, is there a way to evaluate custom language annotations interactively as I see that currently the language embeddings are precomputed during evaluation based on the validation annotations and hence I cannot change them interactively.