Open-Source Model
A machine learning model whose source code, training data, and weights are publicly available for inspection, modification, and redistribution.
Example: NVIDIA Nemotron-3 Family
NVIDIA recently released the Nemotron-3 family of open-source models, featuring three sizes:
- Nano: 30-billion parameters (3-billion active via Mixture-of-Experts architecture).
- Super: 100-billion parameters (10-billion active).
- Ultra: 500-billion parameters (50-billion active).
The gary-explains channel reviewed the Nemotron-3 Nano in th
Example: Google EmbeddingGemma 2
Google has launched EmbeddingGemma 2, a compact open-source embedding model designed for multimodal Retrieval Augmented Generation (RAG).
- Architecture: 740 million parameters.
- License: Apache 2.0.
- Capabilities: Uniquely maps diverse data types—including text, code, images, video, and audio—into a single, unified cross-modal embedding space.
- Use Case: Optimized for on-device multimodal RAG applications.
See EmbeddingGemma 2: On-Device Multimodal RAG with Unified Cross-Modal Embeddings for detailed analysis.