DeepSeek / DeepSeek V3.1
DeepSeek V3.1 FP8 quantization
Estime jusqu aux mesures d ingenierie
Before → After
Version de basePoint de controle verifie
671.000B
671.000B
quantizationFP8
Quality retention target
Quality retention target
Artefact de productionArtefact
Rapport de lineage et de benchmark
Rapport de lineage et de benchmark
Ciblage GPU
Poids et cache KVEstime jusqu aux mesures d ingenierie
Estimated VRAMMeasured and certified results are delivered with the project.
Packaging runtimetransformers