Revision History
T5 (language model) · 14 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 T5's pre-training budget (the released models used about 1 trillion tokens, not 34 billion), the fine-tuning batch size, SQuAD scores (91.26 EM / 96.22 F1; removed fabricated SQuAD 2.0 figures), closed-book QA scores and the origin of salient span masking, C4 filtering thresholds, mT5 and Flan-T5 model sizes, the T5 v1.1 embedding change, a misattributed Flan-T5 comparison, an unsupported XSum claim, an incorrect list of models trained with t5x, and a chronologically impossible claim that Flan-T5 influenced InstructGPT (plus 30 minor fixes)
Version 14 · Nov 2, 2026, 06:23 AM
- Improve internal links (2026-10-13)
Version 12 · Oct 13, 2026, 10:09 AM
- Added 6 contextual internal links
Version 11 · Jul 23, 2026, 03:05 PM
- Render formulas with LaTeX math notation
Version 10 · Jul 11, 2026, 10:31 PM