I wish I got to see the talk How We Scaled Kimi K2.5 [S81695] live and in person at GTC
The session was interesting because Zhilin Yang explained the concepts in a manner that engaged technical and non technical audiences. Furthermore, the presentation had a little of everything for everyone:
- Low level Neural Net Optimizations
- Training optimizations
- Agentic swarms as a concept
Some screenshots I took
The slide deck too if you want to look at it
S81695_1773825130217001pmt7.pdf (10.4 MB)