Model techniques map
Techniquesoptimizationtraining precision

specific method · filed under optimization

NVFP4 pre-training

Pre-training uses an NVFP4 recipe, with the evidence describing it as a compute-efficiency approach.

Also called NVFP4 pre-training recipe, Pre-training with NVFP4 quantization, NVFP4 recipe.

sources
5
models
2
lab adopt it
1
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 2core 3

Documented in

Further reading

Picked by hand, not extracted: where to read more, not evidence for anything on this page.

Evidence

5 spans quoted from the sources, strongest treatment first.

it is trained using an NVFP4 pre-training recipe to maximize compute efficiency

coretraining objectivein Nemotron 3 UltraNVIDIA

trained using an NVFP4 pre-training recipe to maximize compute efficiency

coreoptimizationin Nemotron 3 UltraNVIDIA

trained Nemotron 3 Ultra using the same NVFP4 pretraining recipe

coreoptimizationin Nemotron 3 UltraNVIDIA

NVFP4 pre-training

usedtraining objectivein Nemotron 3 UltraNVIDIA

The pre-training phase used an NVFP4 recipe.

usedoptimizationin Nemotron 3.5 LightningNVIDIA

Filed alongside

Other methods under optimization :: training precision.