Nemotron-4 340B Synthetic (2024年)（ネモトロン4 340B）

概要

Nemotron-4 340B は2024年6月NVIDIA発表・Industry-leading synthetic data generation 98% 2024年NVIDIA Nemotron-4 340B Open Model License synthetic data generation LLM position確立。Industry-leading 340B parameters + Base + Instruct + Reward triplet + designed for synthetic data + 9T tokens trained + NVIDIA Open Model License。

主な特徴・仕組み

• Industry-leading NVIDIA AI 2024年6月 release
• Industry-leading Nemotron-4 340B synthetic data generation LLM signature
• Industry-leading synthetic data generation 98% of fine-tuning data signature
• Industry-leading 340B parameters MoE-like architecture signature
• Industry-leading Base + Instruct + Reward triplet release signature (3-model family)
• Industry-leading designed for synthetic data signature (NVIDIA explicit purpose)
• Industry-leading 9T tokens trained signature
• Industry-leading 50+ languages + 40+ programming languages
• Industry-leading FP8 inference on 8x H100 + 16x H100 signature
• Industry-leading NVIDIA Open Model License signature (research + commercial OK)
• Industry-leading HuggingFace + NVIDIA NIM ecosystem
• Industry-leading 2024年6月 paper + NVIDIA AI Foundation announcement
• Industry-leading NVIDIA 1993 Santa Clara California Jensen Huang 31 years

まとめ

Nemotron-4 340B = 2024年6月NVIDIA発表のpremium Open Model License synthetic data generation LLM。Industry-leading synthetic data generation 98% of fine-tuning data + 340B parameters MoE-like architecture + Base + Instruct + Reward triplet release 3-model family + designed for synthetic data NVIDIA explicit purpose + 9T tokens trained + 50+ languages + 40+ programming languages + FP8 inference on 8x H100 + 16x H100 + NVIDIA Open Model License research + commercial OK + HuggingFace + NVIDIA NIM ecosystem + 2024年6月 paper + NVIDIA AI Foundation announcement + NVIDIA 1993 Santa Clara California Jensen Huang 31 years position確立。