Concept detail
MuonClip / QK-Clip
0 tagged videos · 1 connection videos · degree 0
Publication timeline
- 2026-031 video
Source videos
Observed neighbors
stable training of very large transformers
QK-Clip is the stabilization mechanism that makes the token-efficiency gain from Muon usable at 1T scale, so efficiency and stability are coupled rather than independent.
1 source videos
How We Scaled Kimi K2.5 | Zhilin Yang's full GTC 2026 KeynoteRelated tagged insights
No core insights are tagged with this concept.