Open Models Almanac
Methodsinference & servinginference quantization

specific method · filed under inference & serving

Mamba SSM-cache quantization

The Mamba state-space-model cache is converted from its original FP32 precision to lower precision.

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

Also called SSM cache quantization, quantize the Mamba SSM cache.

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

we quantize the Mamba SSM cache from the original FP32 precision to lower precisions.

usedinference servingin Nemotron 3 UltraNVIDIA
1 more from this model

we quantize the Mamba SSM cache from the original FP32 precision to lower precisions

usedpost trainingin Nemotron 3 UltraNVIDIA
↑