Finetuning on concatenated datasets #165
Draft
+15,416
−221
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).
favyen2
reviewed
Jul 22, 2025
| + "_" | ||
| + model_path_parts[-1].replace(".ckpt", "") | ||
| ) | ||
| eval_task = "__".join(config_paths[0].split(os.path.sep)[-2:]).strip( |
There was a problem hiding this comment.
Is it possible to reuse the rslp_project and rslp_experiment here instead of extracting a model name and task name from the provided files?
rslp/helios/launch_finetune.py
Outdated
| from rslp.log_utils import get_logger | ||
|
|
||
| DEFAULT_RSLP_PREFIX = "project_data/" |
There was a problem hiding this comment.
I don't think this script should have defaults for RSLP_PREFIX, since the assumption everywhere else in this codebase is that the user is responsible for setting the environment variable. It can be placed in .env file.
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.
Mostly just applying changes from rslearn (allenai/rslearn#207)
New configs in
data/helios/v3_multitaskanddata/helios/v3_perf_benchmarkare for multi-dataset training. These are the majority of the edits, plus a bunch of scripts for early evals I did (kind of messy, I don't mind deleting them since they're not really useful for anyone else). Onlymake_multidataset_config.pyshould really be used with any frequency, to create multi-dataset training configs.Also, made some improvements to
launch_finetune. It now supports aprofilerflag, ado_evalflag (run on validation set and save metrics, useful for eval sweeps), and alocalflag (useful for debugging finetuning in the current Beaker session). Seedata/helios/v3_multitask/README.mdfor docs on how to run a multi-dataset job.BREAKING: When building docker images, please place
rslearnandheliosin./docker_build/rslearnand./docker_build/helios, instead of at the repository root. This is to avoid linter issues, where the linter thinks thatrslearnandheliosare local packages instead of standardpipinstalls.