12 ms·
They do show their model as winning every category in Long Range Arena (LRA) benchmark. Hopefully they have not excluded losing categories or better models.
by TheDudeMan 2y ago
They do show their model as winning every category in Long Range Arena (LRA) benchmark. Hopefully they have not excluded losing categories or better models.
- yorwba 2y agoWinning against their own baseline, not against the current best-performing model. Which apparently is S5 currently https://paperswithcode.com/sota/long-range-modeling-on-lra https://paperswithcode.com/sota/long-range-modeling-on-lra with 87.46 overall vs. 58.31 here.