Skip to content

encoder/decoder and multi-step prediction implementation#2

Open
krohling wants to merge 24 commits intomainfrom
dev
Open

encoder/decoder and multi-step prediction implementation#2
krohling wants to merge 24 commits intomainfrom
dev

Conversation

@krohling
Copy link
Copy Markdown
Owner

@krohling krohling commented Sep 4, 2025

No description provided.

@krohling krohling changed the title copy over recent updates to hdf5_dataset.py encoder/decoder and multi-step prediction implementation Sep 8, 2025
Kevin Rohling added 15 commits September 9, 2025 16:59
…ed hdf5 data loader and dataset to enable loading all dataset files
…ature shapes, update enc/dec defaults in train_idm.sh
…-step prediction during evaluation to support n-step rollouts
…g, added encoder std to logging, adjusted lr and vq parameters for individual datasets
…t, only save checkpoints when model loss improves
…ng pre-trained checkpoints, added support for using a randomized subset of the validation dataset during evaluation
…ng evaluation use randomly sample la's to infer image predictions from the wm model. metrics from random la's are logged as well as diff metrics (pred - random_pred), fixed a bug that was only sampling 5 images evaluation ignoring configuration value
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant