Revision History
Adam optimizer · 10 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 claims that PaLM, Claude, Gemini and Mistral were trained with AdamW at beta_2 = 0.95 (PaLM used Adafactor; the others never disclosed their optimizers), fixed the bias-correction step count (about 50 steps for beta_1 = 0.9, not 10), and attributed the Llama 3 405B recipe to the Llama 3 report (plus 13 minor fixes)
Version 10 · Nov 2, 2026, 06:02 AM
- Update: added a video.
Version 9 · Nov 1, 2026, 08:14 AM
- Improve internal links (2026-10-13)
Version 8 · Oct 13, 2026, 10:08 AM
- Added 6 contextual internal links
Version 7 · Jul 23, 2026, 03:04 PM
- Render formulas with LaTeX math notation
Version 6 · Jul 11, 2026, 07:13 PM