tiny-aya-base Blind Spots Evaluation Dataset
Model Tested
CohereLabs/tiny-aya-base
tiny-aya-base is a 1.3B parameter multilingual base language model built on the Aya research initiative by Cohere For AI. It is a base model not instruction-tuned or fine-tuned for any specific task trained on multilingual data spanning 101 languages. This dataset documents systematic failure modes observed when probing the model on technically demanding reasoning tasks.
How… See the full description on the dataset page: https://huggingface.co/datasets/rafaym/tiny-aya-base-blind-spots.
