Implement attentive pooling for finetuning #168
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
There are several other changes: - Upgrade beaker-py to 2.x (queues are not supported in older version). - Update helios.Dockerfile to use pytorch 2.7.0 to reduce build time. - Update one_off_projects/convert_satlas_webmercator_to_rslearn/lib/__init__.py for new rslearn VectorFormat.encode_vector API. - Remove manage_scratch_dir_on_data_disk option since it's not needed anymore (since the Docker volumes are now on the big /data disk across all Beaker nodes).
I also updated the launcher code to accept specifying a list of configs. This enables reducing duplication between some of the config files, although makes it more complicated to start the experiments since you need to specify a list of configs to get the right combination (this is documented in the README files within each task dir in data/helios though).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See the corresponding PR in
rslearn. Implements a simple attention mechanism on top of the shared helios encoder. Should be merged after #165.Breaking change: the helios encoder now returns a full M x B x C x H x W x T tensor, where M is the fused modality and bandset dimension, B is batch, C is channel, H/W are spatial, and T is temporal. The pooling layers in
rslearnhandle the dimension reduction via averaging or attention. Otherwise, backwards compatible with older checkpoints, since a pooling layer is automatically inserted if one isn't found.