Cerebras
Software
Building fast voice-driven interactive demos
"I built demos in the past against Cerebras directly, where the really fun thing is exactly what they're showing there. When you hook up a voice model and you can literally say to your screen, 'Actually show me a chart like this.' That's super fun."
More source quotes
Running AI search/retrieval models at high token throughput (3,000+ tokens/sec)
"That model on Cerebras today already runs at 3,000 tokens per second. We were targeting context running at like 15,000 or 20,000 tokens per second on that chip, and that's going to come this year."
AI inference workloads
"I'm a big fan of Cerebras. In inference, it's a godsend. It's super fast and it's made certain compromises that make it very very good for that."