Revision History
Mixed-Precision Training · 7 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 FP16 underflow threshold and the explanation of why FP32 master weights are needed, H100 and B200 TFLOPS figures, an FP8 training-throughput claim not in its source, the Quartet paper's model scale, the claim that DeepSeek trains in BF16 (DeepSeek-V3 used FP8), and the author lists of two cited papers (plus 14 minor fixes)
Version 7 · Nov 2, 2026, 08:24 AM
- Improve internal links (2026-10-13)
Version 6 · Oct 13, 2026, 10:07 AM
- Added 6 contextual internal links
Version 5 · Jul 23, 2026, 03:04 PM