Revision History
Model Parallelism · 15 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 Llama 3 405B utilization figure (41% MFU, not 57%), removed the false claim that DeepSeek-V3 was trained with tensor parallelism (the report says it was trained without it), fixed the 5D parallelism attribution, and corrected the description of PipeDream-2BW and PipeDream-Flush as two distinct schedules (plus 21 minor fixes)
Version 15 · Nov 2, 2026, 10:06 AM
- Update: added a video.
Version 14 · Nov 1, 2026, 08:14 AM
- Start the article with its lead paragraph: moved the first paragraph above the "What is model parallelism?" heading. No text changed.
Version 13 · Oct 29, 2026, 09:24 AM
- Link [TorchTitan] to the new torchtitan article
Version 12 · Oct 16, 2026, 10:57 PM
- Improve internal links (2026-10-13)
Version 10 · Oct 13, 2026, 10:04 AM