Document prompt requires documents to have metadata variables: ['source']. Received document with missing metadata: ['source'].

[echoopen]

I'm very sorry, but I'm having this problem with the pinecone example: "Document prompt requires documents to have metadata variables: ['source']. Received document with missing metadata: ['source']."

[willydouhard]

Don't be sorry! It looks like the documents you embedded and persisted in your pinecone miss an attribute. If you use the pdf-qa example does it work? (it is the same as the pinecone example but it also creates the vectors from the uploaded pdf)

[echoopen]

The pdf example works, but the pdf example has this code: docs = text_splitter.splitdocuments (documents) for i, doc in enumerate(docs). doc.metadata["source"] = f "source{i}" return docs

But I use DirectoryLoader to load the txt text into docs, after using the above code, still report this error, if it is convenient, convenient to give me a simple DirectoryLoader will be stored in the vector Pinecone example? Mine is docsearch = Pinecone.from_texts(split_docs, embeddings, index_name=index_name)

[willydouhard]

I found this example showing how to useDirectoryLoader with sources.