Community Discussion · Policy

NVIDIA Open-Sources Embed Models: The "Open Weights" Trap in Community Governance

PR MergedJul 182026/07/18 77 views

I noticed an interesting detail: The Nemotron 3 Embed series released by NVIDIA this time is labeled "open weights" rather than "open source." On Hugging Face, the model card says "open weights," and the license permits commercial use, but the code, training scripts, and datasets are not publicly available together. For community veterans, the difference between these few words implies two completely different governance models.

Policy

1 replies

Ctrl + Enter to reply

Long JiJul 30(edited)

[quote="zhang_jinyu, post:1, topic:932"]

I noticed an interesting detail: NVIDIA's newly released Nemotron 3 Embed series is labeled "open weights" rather than "open source." On Hugging Face, the model card says "open weights," the license allows commercial use, but code, training scripts, and datasets are not publicly available. For community veterans, the difference between these words implies two completely different governance models.

From technical metrics, the 8B version topping the RTEB leaderboard is indeed impressive. Embedding ...

[/quote]

This news is worth noting. The practical gap between open weights and open source is amplified in infrastructure models like Embeddings—once developers rely on them, subsequent migration costs get stuck on CUDA.