Troubleshooting API connection issues

PPico

running into an issue of not connecting, api is in my env, I checked with echo, not sure why its nto working

Attachment

Screenshot_2024-11-11_at_12.11.47_PM.png

25 comments

LLogan M

That prints your key?

PPico

yes it does

LLogan M

Is the key valid? You can test this directly too

Plain Text

from llama_index.embeddings.openai import OpenAIEmbedding

# test directly
embed_model = OpenAIEmbedding(api_key=api_key)
resp = embed_model.get_text_embedding("hello world")

# can also pass in directly 
index = VectorStoreIndex(nodes=nodes, embed_model=embed_model)

PPico

yeah that seemed to work

PPico

thank you so much

PPico

i am now in the final portion of this program and now it wont give me a response

Attachment

PPico

any idea why?

LLogan M

Seems a similar issue, it's not finding your key for whatever reason

Similar solution too

Plain Text

from llama_index.llms.openai import OpenAI

llm = OpenAI(model="gpt-4o-mini" api_key="...")

query_engine = index.as_query_engine(llm=llm)

PPico

heres my engine, do i just put the llm=llm into the selector?

Attachment

LLogan M

yes

PPico

i really appreciate the help, thank you

PPico

lemme try that

LLogan M

you'll also want to make sure your tools/other query engines have an llm too

PPico

interesting so I shouldnt run from defaults then?

LLogan M

from defaults is fine. For the selector, the llm goes in there LLMSingleSelector.from_defaults(llm=llm)

Your query engine tools are also fine, nearly

LLogan M

but you are passing in an index instead of query engine for the vector tool

LLogan M

and when you create the vector index query engine/summary query engine, pass the llm in there

LLogan M

hope that makes sense

PPico

got a little bit closer, let me try to put the llm in both engines

Attachment

PPico

sickkk i got it to summarize

PPico

thanks so much

LLogan M

normally you don't have to do this 😅 It should be picking it up from os.environ in your notebook, but I guess something is wonky

LLogan M

Yea happy to help!

PPico

a little bit late, but i was curious, if i wanted to hookup a csv instead of a document how easy could it be?

LLogan M

CSV's are pretty tricky. If its just a list of QA pairs for example, embedding every row makes sense (which is what simpledirectoryreader does by default)

If they are highly numerical/relational, you might want text2sql

Add a reply

Find answers from the community

Troubleshooting API connection issues