ResearchTasksToken-level ReconstructionFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedHigh-entropy random token sequences Full-OOD87M parameter Transformer Autoencoder57Reconstruction Accuracy (TRA)1Feb 26, 2026Salesforce/wikitext-103 Semi-OOD (val)87M parameter Transformer Autoencoder47.76Reconstruction Accuracy (TRA)1Feb 26, 2026codeparrot-clean In-Domain (val)87M parameter Transformer Autoencoder99.47Reconstruction Accuracy (TRA)1Feb 26, 2026
High-entropy random token sequences Full-OOD87M parameter Transformer Autoencoder57Reconstruction Accuracy (TRA)1Feb 26, 2026
Salesforce/wikitext-103 Semi-OOD (val)87M parameter Transformer Autoencoder47.76Reconstruction Accuracy (TRA)1Feb 26, 2026
codeparrot-clean In-Domain (val)87M parameter Transformer Autoencoder99.47Reconstruction Accuracy (TRA)1Feb 26, 2026