aminediroHF/scaling-rl-dp2eai-2026
1
Add 'Thank you / Questions?' closing slide after Ongoing work
Slide 19: center 'one device per (dp,ep,sp,cp)' under the cube
Speaker notes: why 100k+ (agentic harness); MFU/TPS+OOM metrics, SFTTrainer proxy
Apply review: PPO clip (s4), SP schema simplified (s17), drop s20+renumber, end CTA
Add Kashif Rasul as co-author (title + footers)
Add DP2E-AI 2026 presentation (static deck)
initial commit
