28-dataset aggregate

28 GUE datasets across seven task categories

Protocol fingerprint

PF-v1-85A1BBC0AB45D24B12B765DD

The result chunk loads only after both the metric and protocol-metric fingerprint are selected.

Split: official train/validation/test splits
Aggregation: arithmetic mean of task-specific MCC/F1 scores across 28 datasets
External data: pretrained models allowed; no extra task labels beyond official training splits
Ensemble / best-of-N: single model · 1
Critical unknowns: inference_budget

Interactive results

Filter and compare inside one fingerprint

Choose a metric and fingerprint to load results.

canonical dataset ids: gue_h3;gue_h3k14ac;gue_h3k36me3;gue_h3k4me1;gue_h3k4me2;gue_h3k4me3;gue_h3k79me3;gue_h3k9ac;gue_h4;gue_h4ac;gue_covid;gue_mouse0;gue_mouse1;gue_mouse2;gue_mouse3;gue_mouse4;gue_prom_300_all;gue_prom_300_notata;gue_prom_300_tata;gue_prom_core_all;gue_prom_core_notata;gue_prom_core_tata;gue_splice_reconstructed;gue_tf0;gue_tf1;gue_tf2;gue_tf3;gue_tf4
data version: GUE 28-dataset paper release
label visibility: train labels visible; test labels hidden from training
input modality: nucleotide sequence
preprocessing: official benchmark preprocessing
model selection rule: validation loss checked every 200 steps; best validation-loss checkpoint
inference budget: NR
protocol completeness: exact