Once I have already created a Knowledge

rrohan._.s

Once I have already created a Knowledge Graph Index using the "from_documents" method, how can I just load the index from the existing Neo4j knowledge graph that I've created?

24 comments

rrohan._.s

I know the Property Graph Index has a "from_existing" method but I don't see that method in the Knowledge Graph Index

LLogan M

Knowledge graph index is kind of deprecated.

But in any case, unless you built the graph with llama-index, I think the only choice is text to cypher

rrohan._.s

Is there a more updated graph rag index I should use?

LLogan M

Property graph index is the go to

LLogan M

It's much more customizable

rrohan._.s

Does property graph also have the "include embeddings" option like the Knowledge graph index does?

LLogan M

It assumes embeddings. Probably a good idea to read up on the (very detailed) docs

LLogan M

https://docs.llamaindex.ai/en/stable/module_guides/indexing/lpg_index_guide/?h=property

rrohan._.s

https://docs.llamaindex.ai/en/stable/examples/property_graph/property_graph_neo4j/
I followed this guide to use the Property graph with Neo4j, but used ollama to run my own model instead of OpenAI. After running for a couple hours to create the graph, it errored out with this message:

Plain Text

neo4j.exceptions.ClientError: {code: Neo.ClientError.Procedure.ProcedureCallFailed} {message: Failed to invoke procedure `db.create.setNodeVectorProperty`: Caused by: java.lang.IllegalArgumentException: Vector must only contain finite values. Provided: SequenceValueVectorCandidate[sequence=List{}]}

LLogan M

Seems like an issue with embeddings?

rrohan._.s

I used Neo4j to store the embedding vectors but maybe that causes problems?

rrohan._.s

I followed the guide pretty much exactly with my own docs so I'm not sure what's wrong

LLogan M

I guess I just meant, it sounds like your embedding model returned inf values

rrohan._.s

Do you have any suggestions to ensure that the embeddings don't return inf values?

LLogan M

Not sure 🤔 Normally this shouldn't happen actually. But I guess it depends what embedding model you were using? Which embed model was this?

rrohan._.s

I've tested with mxbai-embed-large:latest and all-minilm:33m

rrohan._.s

Both had the same issue

rrohan._.s

Is this code just only compatible with OpenAI embedding models?

LLogan M

But what embedding model class? HuggingFaceEmbedding?

LLogan M

It should work with any embedding model. But it seems like the one you are using has a bug

rrohan._.s

ollama_embedding = OllamaEmbedding(
model_name="all-minilm:33m",
base_url="http://localhost:11434"
)

rrohan._.s

So maybe Ollama Embedding class is the problem? I've tested multiple sentence transformers and they appear to all have the same issue

LLogan M

Might be an issue with ollama yea 🤔 Have you updated the ollama server recently?

rrohan._.s

Yea I'm using ollama v1.4.8

Add a reply

Find answers from the community

Once I have already created a Knowledge