Node

I am inserting images into my index with a mimetype, and then when querying the index am getting this

Plain Text

"Error generating summary: 'NodeWithScore' object has no attribute 'image_mimetype'"

Plain Text

image_doc = ImageDocument(
            doc_id=f"{image_request.userId}_{image_id}",
            image=image_data,
            text=image_request.text_request.text
            if image_request.text_request
            else "",
            image_mimetype=image_request.mimetype,
        )

the mimetype is there and correct, but it doesn't end up in my vector database somehow and now is causing an error?

31 comments

LLogan M

NodeWithScore only has a node and score attribute

To access it, I think you'd need to node_with_score.node.image_mimetype

ZZachHandley

that's not my error

ZZachHandley

it's Llama Index

ZZachHandley

cause I definitely am not trying to access the node.image_mimetype lol

ZZachHandley

working on building my own query engine I think because the issue is I really just need it to stop thinking about if an image is relevant or not and just know that all images are relevant that it can see in the vector database but to pick the best ones to achieve the task

LLogan M

If you have the full traceback, it can probably help track down the issue

ZZachHandley

lemme get it, that's FastAPI

ZZachHandley

all I'm getting is a basic error message :thinkRoll:

Hmmmm

Sus

Lol

Maybe because you have it wrapped in a try/except somewhere

ZZachHandley

Attachment

ZZachHandley

LLogan M

So the traceback is being swallowed

good point

one sec

@Logan M

I do insert the image_mimetype and I do check, it's definitely being set on insert

ZZachHandley

but it's definitely not making it to Qdrant

ZZachHandley

so ImageDocument is also removing my image_mimetype somewhere too

LLogan M

Hmm, so 1) I wonder if the node parser is removing it

2) the llamaindex code probably shouldn't be passing in NodeWithScore objects there, feels like a bug

ZZachHandley

yeahhhhh so is there a way for me to just retrieve everything filtered for a user and then have it just select the top_k most relevant ones but guarantee it selects them?

LLogan M

filters are applied before top k is found

ZZachHandley

understood, but that doesn't change anything because they're all the same User ID and it still is giving me nothin even with an AI generated metadata

ZZachHandley

Plain Text

{"id_": "131f30e5-472e-4aeb-b04f-446118e10915", "embedding": null, "metadata": {"user_id": "65a0349f048a9118ff98", "text": "The image shows a black-and-white portrait of a young man with a thoughtful expression, possibly symbolizing introspection or a serious, contemplative personality."}, "excluded_embed_metadata_keys": [], "excluded_llm_metadata_keys": [], "relationships": {"1": {"node_id": "65a0349f048a9118ff98_65ab0bea34fd19fa6f92", "node_type": "4", "metadata": {"user_id": "65a0349f048a9118ff98", "text": "The image shows a black-and-white portrait of a young man with a thoughtful expression, possibly symbolizing introspection or a serious, contemplative personality."}, "hash": "44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a", "class_name": "RelatedNodeInfo"}}, "hash": "44136fa355b3678a1146ad16f7e8649e94fb4fc21fe77e8310c060f61caaff8a", "text": "", "start_char_idx": 0, "end_char_idx": 0, "text_template": "{metadata_str}\n\n{content}", "metadata_template": "{key}: {value}", "metadata_seperator": "\n", "image": "/9j/4AAQSkZJRgABAQAASABIAAD/4QBwRXhpZgAATU0AKgAAAAgABAESAAMAAAABAAEAAAFCAAQAAAABAAACDgFDAAQAAAABAAACqIdpAAQAAAABAAAAPgAAAAAAA6ABAAMAAAABAAEAAKACAAQAAAABAAACDaADAAQAAAABAAACpwAAAAD/7QA4UGhvdG9zaG9wIDMuMAA4QklNBAQAAAAAAAA4QklNBCUAAAAAABDUHYzZjwCyBOmACZjs+EJ+/8AAEQgCpwINAwERAAIRAQMRAf/EAB8AAAEFAQEBAQEBAAAAAAAAAAABAgMEBQYHCAkKC//EALUQAAIBAwMCBAMFBQQEAAABfQECAwAEEQUSITFBBhNRYQcicRQygZGhCCN

ZZachHandley

Plain Text

The image shows a black-and-white portrait of a young man with a thoughtful expression, possibly symbolizing introspection or a serious, contemplative personality.

ZZachHandley

that was my own generation

ZZachHandley

using the multimodal_agent.acompletion endpoint

ZZachHandley

not what it gave me.

ZZachHandley

@jerryjliu0

Add a reply

Find answers from the community

Node