Hi, In AI Workbench, the Hybrid RAG example app doesn’t work when used with DGX Spark. After cloning the project in AI Workbench and set the NVIDIA API Key, the build fails. Pls advise how to fix.
For a Spark specific RAG example using AI Workbench, check out our playbook RAG Application in AI Workbench | DGX Spark
Thank you for your response. We’re looking for a RAG solution on DGX Spark that can operate without access to Internet. We evaluated the RAG Application in AI Workbench | DGX Spark .
It’s an Agentic RAG solution that requires access to a 3rd party service (TAVILY). This does not meet our requirements as we do not want user queries to be sent to a 3rd party service. We modified the Agentic RAG code so it would not send the queries to 3rd party service but this will not work.
The Hybrid RAG example looks promising as it doesn’t require sending user queries to a 3rd party service. We hope nVidia could provide a solution to fix this build issue on DGX Spark.
Unfortunately, it seems the Hybrid RAG project is not compatible with ARM machines. You would need to get Hugging Face to update their containers
Could you kindly recommend a different Hybrid RAG project for DGX Spark operating in an air-gapped setup? The project doesn’t have to use AI Workbench.