Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
Not assessedPlannedFindingclaim-downstream-zero-shot-780m

Mamba-2-780M trained on 300B tokens of the Pile achieves zero-shot downstream task performance of 61.7% on LAMBADA, 54.9% on HellaSwag, 72.0% on PIQA, 61.0% on Arc-Easy, 28.5% on Arc-Challenge, 60.2% on WinoGrande, 36.2% on OpenbookQA, and 53.5% average accuracy across tasks.

Source: source-paper:Table 1, page 29 (and Table 10, page 52)

Reported and observed measurements

lambada_acc

rm-780m-lambada-acc

Reported 61.7 percent

Observed — percent

hellaswag_acc

rm-780m-hellaswag-acc

Reported 54.9 percent

Observed — percent

piqa_acc

rm-780m-piqa-acc

Reported 72 percent

Observed — percent

arc_easy_acc

rm-780m-arc-e-acc

Reported 61 percent

Observed — percent

arc_challenge_acc

rm-780m-arc-c-acc

Reported 28.5 percent

Observed — percent

winogrande_acc

rm-780m-winogrande-acc

Reported 60.2 percent

Observed — percent

openbookqa_acc

rm-780m-openbookqa-acc

Reported 36.2 percent

Observed — percent

average_acc

rm-780m-avg-acc

Reported 53.5 percent

Observed — percent

Assessments (0)

No immutable Assessment has been published for this Claim yet.

Runs (0)

No execution has been linked to this Claim yet.