Nemotron-4 340B

by Nvidiaopen weights

Nemotron family

Parameters

340B

Context window

4K tokens

Released

2024-06-14

Input price / 1M tok

—

Output price / 1M tok

—

Modality

tools

About Nemotron-4 340B

Nvidia's flagship open-weight model. Notable for its purpose-built synthetic-data pipeline — most training data was generated by prior Nemotron / Llama runs. Primarily used as a teacher for distilling smaller models, not for direct production.

Strengths

Open weights
Purpose-built for synthetic data generation
Strong at instruction following

Weaknesses

4K context (very small)
Not tuned for chat
Massive to self-host

Best for

Synthetic data generationDistillationResearch

Vendor page →HuggingFace →