Neptune was printing extra lines between tqdm progress bar updates, which has been suppressed so the progress bar displays as expected.
Where different runs are finishing at the same time, data_key is not unique. This causes two JSON files to be stored in one folder with the same data_key name and messes up plotting downstream. This has been fixed so that each JSON file is stored in a unique folder.
Why?
Prettier prints and easier to read.
JSON files are downloaded as marl-eval's concatenating and plotting tools require.
How?
During the download phase, set Neptune to only log messages at an error level or higher (i.e., we don't want the extra info).
Append a counter to each file name before creating the directory for the JSON file.
What?
tqdm
progress bar updates, which has been suppressed so the progress bar displays as expected.data_key
is not unique. This causes two JSON files to be stored in one folder with the samedata_key
name and messes up plotting downstream. This has been fixed so that each JSON file is stored in a unique folder.Why?
marl-eval
's concatenating and plotting tools require.How?