Skip to content

Comment on Surpassing vLLM with a Generated Inference Stack

Comments

Luke: Do you have benchmarks for BF16?

Unfortunately, not at present; we went for FP8 because we believed it was generally the best tradeoff of quality and speed. Allowed faster iteration as well.

We believe our improvements would hold on BF16, but let me check.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.