-
Notifications
You must be signed in to change notification settings - Fork 51
Description
Hi CodaBench team,
We’re currently running a challenge on the platform (NTIRE 2026 Efficient LLIE, https://www.codabench.org/competitions/13382) and have observed a recurring issue during the final test phase:
-
Submissions with result ZIP files around 500–600 MB typically complete within ~15–20 minutes, produce valid scores, and show full logs.
-
However, submissions with ZIP sizes exceeding ~1 GB almost always either:
Remain stuck in "Running" status indefinitely, or
Eventually return a score of 0.0 with no evaluation logs—suggesting failure during early stages (e.g., unpacking or I/O), before the scoring script even starts.
Our scoring script itself is lightweight and finishes quickly when given valid inputs. The large ZIPs contain only PNG images (no code or extra files), but due to the nature of the task (image enhancement on a large test set), file size can grow significantly depending on bit depth and channel format.
We’re unsure whether this is due to:
-
A hard limit on ZIP size or extracted directory size,
-
Timeout during extraction/unpacking,
-
Disk quota limits in the worker environment,
-
Or another resource constraint.
Could you clarify if there are known limits on submission size or extracted content volume? And if so, what are the recommended thresholds to avoid silent failures?
Thank you very much for your support—and for maintaining this valuable platform!