Llama 3.1 405B now runs at 969 tokens/s on Cerebras Inference

(cerebras.ai)

427 points benchmarkist | 1 comments | 19 Nov 24 00:15 UTC | HN request time: 0.201s | source

1. qwertox ◴[19 Nov 24 07:03 UTC] No.42180721[source]▶

I'd like to see a tokens / second / watt comparison.