Detailed results¶

We provide below the many-to-one results for the baseline models: every layer, metric, language, and finetuning duration, on the test split. Select a language, or the average over the dev or test languages. The units and scores are available in the artifacts dataset.

Across layers¶

Best layer, by finetuning duration¶

The best layer of each model and finetuning duration minimizes the continuous ABX on dev languages.