Open Models Almanac
Methodssoftware implementationkernel & quantization library

implementation detail · filed under software implementation

FlashInfer NVLinkOneSided all-to-all

FlashInfer's NVLinkOneSided implementation used for all-to-all communication, selected for end-to-end serving performance.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See kernel & quantization library for the methods filed beside it.

Also called FlashInfer's NVLinkOneSided, FlashInfer’s NVLinkOneSided implementation.

sources
2
model
1
lab adopt it
1
strongest
used

Documented in

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
used 2

NVIDIA-Nemotron-3-Ultra-550B-A55B

NVIDIAadopted2 sources

adopted FlashInfer’s NVLinkOneSided implementation... as the best end-to-end choice for Nemotron 3 Ultra serving on GB200 NVL72

usedsoftware implementationin Nemotron 3 UltraNVIDIA
1 more from this model

adopted FlashInfer's NVLinkOneSided implementation as the best end-to-end choice for Nemotron 3 Ultra serving on GB200 NVL72

usedinference servingin Nemotron 3 UltraNVIDIA
↑