I evaluated Comet pt50 on the original 50 tasks using KOA and led packaging and submission. I built the Docker image, uploaded evaluation data to Hugging Face, and wrote the README. A teammate evaluated the new 50 tasks with additional recovery data.
Evaluation data Activity notes
URI ghcr.io/haw-ai-i/b1k-submission:armB-62e098e
Digest ghcr.io/haw-ai-i/b1k-submission@sha256:eb396e95de334d3b3a5beb7e62f28ec035f2bdd2e3d3db51d30265df392f2f4a