Hi everyone, im trying to load my index

Any idea why?

what kind of index is it?

DocummentSummaryIndex

first generated with .from_documents()

so you'll also need to save/persist a vector store as well I think 🤔 Which might be why you get empty responses

self.index = DocumentSummaryIndex.from_documents(
documents=self.llamaindex_documents,
service_context=self.service_context,
storage_context=self.storage_context,
show_progress=True,
)
self.index.set_index_id(self.chat_id)
self.redis.add_index(self.chat_id)

self.query_engine = self.index.as_query_engine(
response_mode="tree_summarize",
streaming=True,
use_async=True,
summary_template=self.summary_template,
)
self.index.storage_context.persist("index")

Ah, so the vector store is probably saved locally in the index folder

self.stoarge_context = StorageContext.from_defaults(docstore=docstore, index_store=index_store, persist_dir="./index") would probably work

In that example is saved only using the docstore and indexstore

Not my intention

Im trying to persist the index in the RedisIndexStore

it persists automatically actually (except for the vector store, because you did not provide one), so no need to call persist

Now that you say it its true that is making an index folder

yea, that example needs to be updated I guess :PSadge:

lol in the example, the response is empty when querying the loaded vector index too

makes sense

tbh didnt read until the end

As developers do 😉

So solution here is to either use the default vector store that was saved to disk, or use some other vector store (redis, qdrant, weaviate, pinecone, etc.)

Will try to add a Redis vector store but i dont see why is needed to work

The summaries of each document are embedded

so that they can be retrieved

Without the vector store, there are no summaries to retrieve

hence the empty response

Really? I expected the summary to be saved in the docstore

So the SummaryIndex is just a Vector store index in disguise

Its a combination

Interesting

Summaries are embedded, but the original documents live in the docstore

Makes sense

Will try to solve

Can i update the docs?

I mean, does the community have acces to the make changes in the docs?

definitely! You can fork the repo on github, make changes, and then make a PR with your fork

happy to merge any changes

Perfect

node_ids = self._index_struct.summary_id_to_node_ids[summary_id]
KeyError: 'llama_index/vector_4e8af3bf-d8ab-4b2b-a268-7b608b6a5010'

Does this error ring any bell? @Logan M

hmm, this tells me maybe the index store did not save/load properly? I should probably try this myself with a redis setup

if i comment line 165 and 166 # if query_result.ids is not None:
# top_k_summary_ids = query_result.ids

on retrievers.py it works

So i guess something is wrong when doing the query of the Reddis vector store