At GTC Berlin, NVIDIA added more open-weight models to its library, pitched at Europe’s appetite for sovereign AI.
What’s new
- Refreshed Nemotron Retriever embedding, parsing and reranking models for agentic RAG.
- New Alpamayo models for autonomous vehicles, plus healthcare and robotics models.
- The idea is that most companies can’t train from scratch, so they fine-tune open weights, like Mistral’s, on their own data.
My take
Embedding and reranking models aren’t flashy, but they’re what makes “search your own docs” actually work. Open weights for that layer is good news for anyone building local RAG.
Have a blessed day! And Happy Hacking!
