Show HN: Where a hand-written GPU kernel beats the library, and where it can'tkyrieblunders.bearblog.dev·1 pts·vishal-padia·0
I made a kernel 2.2x faster. It made my training loop 3x slowerkyrieblunders.bearblog.dev·24 pts·vishal-padia·3
Wrote FA2 using CuteDSL, what worked and what didn't workkyrieblunders.bearblog.dev·2 pts·vishal-padia·0