DataGemma Fine-tuned Gemma 2 models that integrate retrieval techniques to ground responses in real-world data Read blog The world’s first open models designed to help address the challenges of hallucination by grounding LLMs in the vast, real-world statistical data of Google's Data Commons. Read docs View technical report Watch Model series Retrieval-Interleaved Generation (RIG)This approach uses a variant of Gemma 2 that is fine-tuned to recognize when it needs to replace a generated number with more accurate information from Data Commons. Retrieval-Augmented Generation (RAG)This approach uses a variant of Gemma 2 that retrieves relevant information from Data Commons and then uses that information to create an extended prompt for the Gemini 1.5 Pro model. Run or download RIG Run RIG in Colab Download RIG from Kaggle Download RIG from Hugging Face Run or download RAG Run RAG in Colab Download RAG from Kaggle Download RAG from Hugging Face Capabilities Generate answers with real dataExplore and uncover verifiable insights by simply asking questions in plain language. Evaluate AI data grounding techniquesInvestigate ways to guide generative AI model output with retrieval-augmented and data-interleaved techniques.
