In your video you say "sentence_file_path". However shouldn't it be "embed_file_path" ? create_tree_ah_index function should have the GCS bucket of the embedded data and not the text with teh ids right ?
Great video! What is the difference between Vertex Search service VS Vector Search for RAG application? which one is better in terms of handling better retrieval of relevant documents for RAG application where we deal with 100+ PDF documents? Can you share some insights?
Thanks for the tutorial! Instead of going through the ids in the json file to fetch the sentences, is it possible to integrate those directly as metadata in the index?
Thanks, this is tremendously helpful One point to note - you need to upload the embed file, not the sentence file -> upload_file(bucket_name,embed_file_path)
Great Video. One question, I noticed you used a different model (gecko) to Gemini Pro for the embeddings. Is this ok to do? I assumed the models needed to be the same for both training and inference? Thanks again
Text embedding models are independent of LLMs. You only have to ensure that the same embedding model is used for indexing the documents and the query. This is critical to retrieving the context based on the similarity.