Revision History
RoBERTa · 11 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Correction: Corrected the description of static versus dynamic masking (the 10x data duplication was BERT's static baseline, with each sequence seen with the same mask four times), removed a false claim that the 100K-step run used the same compute budget as BERT, fixed the ALBERT and ELECTRA training-data and SQuAD 2.0 entries in the comparison table, clarified that RoBERTa's GLUE test submission used ensembles, and removed the implication that RoBERTa came from the Allen Institute for AI (plus 18 minor fixes)
Version 11 · Nov 2, 2026, 06:40 AM
- Improve internal links (2026-10-13)
Version 10 · Oct 13, 2026, 10:09 AM
- Added 6 contextual internal links
Version 9 · Jul 23, 2026, 03:05 PM