Pradeep Dasigi
@pdasigi.bsky.social
#NLP research @ai2.bsky.social; OLMo post-training
https://pdasigi.github.io/
https://pdasigi.github.io/
For each "core skill" we care about, we chose a separate set of "development" and "unseen" evaluations. We tracked the performance of models only on the former during development and evaluated only the final checkpoints on the unseen ones.
November 23, 2024 at 11:53 PM
For each "core skill" we care about, we chose a separate set of "development" and "unseen" evaluations. We tracked the performance of models only on the former during development and evaluated only the final checkpoints on the unseen ones.