Revision History
Inference optimization · 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: qualify workload-dependent performance claims and replace unsupported cost estimates with reproducible cost methodology
Version 11 · Nov 3, 2026, 09:30 AM
- Correction: Corrected the description of AWQ (it scales salient channels rather than keeping them in higher precision), a misattributed FlashAttention speedup figure (7.6x on attention, not 2-4x), and the GPT-4 launch date in the cost table, and removed or labelled several unsourced figures (plus 12 minor fixes)
Version 10 · Nov 2, 2026, 11:51 AM
- Improve internal links (2026-10-13)
Version 9 · Oct 13, 2026, 10:08 AM
- Added 6 contextual internal links
Version 8 · Jul 23, 2026, 03:04 PM