Revision History
FP4 (4-bit floating point) · 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 Blackwell FP4 throughput (20 PFLOPS per GPU is the sparse figure, about 10 dense), the Transformer Engine generation (second, not fifth), an unsourced '2.3x' throughput figure misattributed to NVIDIA, a misdated and misdescribed NVFP4 accuracy evaluation, the claim that NVIDIA's NVFP4 checkpoints were made with llm-compressor, and unsupported Trainium and Intel MXFP4 claims (plus 19 minor fixes)
Version 7 · Nov 5, 2026, 04:29 AM
- Correction: Quartet paper authors fixed (Castro, Panferov et al.; no "Tsizhanovska"); AMD MXFP4 support dated to the shipping MI350/MI355X series (2025) rather than only future MI400 hardware
Version 6 · Oct 29, 2026, 09:46 PM
- Improve internal links (2026-10-13)
Version 5 · Oct 13, 2026, 10:05 AM
- Added 6 contextual internal links
Version 4 · Jul 23, 2026, 03:43 PM
- Initial AI-authored creation (3712 words).
Version 2 · May 21, 2026, 06:18 AM