Revision History
Mamba 2 · 8 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 Mamba-2 head-sharing description (B and C are shared across heads, not per head), removed a fabricated 8.5 vs 9.1 perplexity comparison (paper: 6.09 vs 6.13 at 2.7B), noted that Jamba 1.5 and Hymba use Mamba-1 rather than Mamba 2, fixed Jamba parameter counts and memory claims, replaced unsupported Hunyuan-TurboS speed figures with the reported 1.8x, corrected Hymba's benchmark claim, Jamba 1.6 customer figures, and Cartesia Sonic's latency (135 ms) (plus 28 minor fixes)
Version 8 · Nov 2, 2026, 05:06 PM
- Update: fix a broken link on the first name Albert (pointed to the ALBERT language model)
Version 7 · Nov 2, 2026, 08:51 AM
- Correction: Codestral Mamba is a pure Mamba 2 model with no attention layers, and Hymba (November 2024) preceded the Nemotron-H paper (April 2025).
Version 6 · Oct 31, 2026, 03:47 AM
- Improve internal links (2026-10-13)
Version 5 · Oct 13, 2026, 10:20 AM
- Added 6 contextual internal links
Version 4 · Jul 23, 2026, 03:29 PM