Revision History
Tülu 3 · 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 author count and affiliations; SFT prompt figures (939,344 is the SFT mix drawn from 23.3M prompts, and ShareGPT was excluded, not used); the core skill count (seven, not eleven); the DPO data description (GPT-4o judge only, 22-model off-policy pool, no HelpSteer); RLVR scope (no code verifier); the 405B GPU split and invented inference-bottleneck claims; the 405B post date relative to DeepSeek R1; a TechCrunch misquote; and the Qwen 2.5 and Hermes 3 comparison rows (plus 23 minor fixes)
Version 7 · Nov 5, 2026, 09:14 AM
- Improve internal links (2026-10-13)
Version 5 · Oct 13, 2026, 10:03 AM
- Added 6 contextual internal links
Version 4 · Jul 23, 2026, 03:30 PM