NVIDIA's DeepSeek-V4-Pro-0813-nvfp4-DSpark release packages DeepSeek's main model and speculative-decoding draft component in a consistent numeric format. The draft component proposes several candidate text tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → for the main model to verify, allowing multiple accepted tokens to emerge from a generation step. According to the model card, this checkpoint keeps the backbone weights identical to NVIDIA's earlier NVFP4 release and additionally converts the draft experts to NVFP4. NVIDIA attributes the draft model itself to DeepSeek; its contribution here is quantizationShrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.Full definition → and packaging. The release addresses deployment consistency rather than introducing a newly trained reasoning modelA model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.Full definition →.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
