Hey there I am migrating from llama

wwalid

Hey there, I am migrating from llama-index 0.4 to 0.6 and I am having trouble translating the syntax from the old version to the new one

How would one write the following in the newer versions?

Plain Text

# Indexing
# this should directly index documents into Elasticsearch
client = ElasticsearchVectorClient()
GPTOpensearchIndex(documents, client=client, chunk_size_limit=1024)

# Querying
# this should ask the query 'q' on the Elasticsearch index, using the qa & refinement templates provided.
# and with the LLM Predictor provided
client = ElasticsearchVectorClient()
index = GPTOpensearchIndex([], client=client)
llm_predictor = LLMPredictor(llm=ChatOpenAI(
    temperature=0, model_name="gpt-3.5-turbo"))

similarity_top_k = 1
index.query(q, similarity_top_k=similarity_top_k,
               llm_predictor=llm_predictor,
               text_qa_template=CHAT_QA_PROMPT,
               refine_template=CHAT_REFINE_PROMPT)

13 comments

wwalid

I've tried the following for indexing and it doesn't seem to work properly

Plain Text

documents = SimpleDirectoryReader('./data').load_data()
client = ElasticsearchVectorClient()
vector_store = OpensearchVectorStore(client=client)
storage_context = StorageContext.from_defaults(vector_store=vector_store)
service_context = ServiceContext.from_defaults(chunk_size_limit=1024)

VectorStoreIndex.from_documents(documents, storage_context=storage_context, service_context=service_context)

LLogan M

dang, I just had it all typed out and it dissapeared 😦 Will type again.... lol

wwalid

What I wrote above works now, I had something faulty in other code

I wonder if that is the correct way of doing it though?

LLogan M

That is the correct way I think! You are ahead of me lol I should have read that haha

For the query, it will look like this

Plain Text

query_engine = index.as_query_engine(similarity_top_k=1, text_qa_template=CHAT_QA_PROMPT, refine_template=CHAT_REFINE_PROMPT)
response = query_engine.query("my query")

wwalid

Ah nice! thanks a lot @Logan M , you're always helpful!

In your example, how would you initialize that index with Elasticsearch as the backend?

I'm trying this but apparently there's nothing called from_vector_store

Plain Text

client = ElasticsearchVectorClient(account_id=account_id,
                                app_id=app_id)

vector_store = OpensearchVectorStore(client)
index = VectorStoreIndex.from_vector_store(vector_store=vector_store)

Also, the main refactor here is:

that storage backends are provided through the storage_context parameter
configurations specific to the query/indexer are provided through the service_context parameter

Is my understanding correct?

LLogan M

the from_vector_store is very new, I can't remember if it's on the latest pypi version yet or not 👀 But that's mainly used to connect to a vector db that's already be populated. Setting up the storage like you did before is the initial way to do it

LLogan M

And yea, that's the main refactor there! Additionaly, you might have noticed the as_query_engine() thing too, which kind of seperates the query from the index a bit

wwalid

Yeah, I want to use a storage that is already populated.

I basically have two main endpoints in my app, one for indexing, and the other for querying.

And the querying endpoint should use the pre-populated Elasticsearch index

wwalid

Any alternatives to from_vector_store that can use the prepopulated index?

LLogan M

the alternative is setting up the vector store with the storage context again (so you setup the client to connect to the existing vector db created with llama-index) and then you can do something like this

index = VectorStoreIndex([], storage_context=storage_context)

wwalid

Ah interesting, so it's the old way (where we had to pass an empty list)

Makes sense 🤔 thanks a lot man!

LLogan M

Yea! from_vector_store is just making that look less awkward (and automates the storage_context setup)

wwalid

Noice! much cleaner than the old interface yet sadly, a very confusing and breaking change

Add a reply

Find answers from the community

Hey there I am migrating from llama