Downstream Evaluation Suite
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Downstream Evaluation Suite Zero-shot (FW-Edu, Wiki., LAMBADA, PIQA, HellaSwag, WinoGrande, ARC, SIQA, SciQ) (val)
10.52FW-Edu Perplexity
9
Downstream Evaluation Suite (ARC-C, Hellaswag, PIQA, SciQ, Winograde, SocialIQA, RACE) zero-shot (test)
48.7ARC-C Accuracy
9
Downstream Evaluation Suite (Arc-e, PIQA, Hellaswag, OpenBookQA, Winogrande, MMLU, BoolQ)
53.83Arc-e
4