Actually probably easier to continue in here
@Akinus21 What about providing an empty OpenAI key
something like
import os
os.environ["OPENAI_API_KEY"] = ""
Getting the same error. I'm going to try with the openai module:
Import openai
openai.api_key = ""
I've had to do that in the past, but for the life of me I can't figure out which module is still needing that variable.
Maybe there's something off in your queryconfigs?
def ask_gpt_custom(prompt):
graph = build_index(prompt)
query_configs = [
{
"index_struct_type": "tree",
"query_mode": "embedding",
"query_kwargs": {
"child_branch_factor": 2
}
},
]
response = graph.query(
prompt,
query_configs=query_configs
)
return f'{response}'oh didn't even see you already posted it
and this all goes well until the graph.query I guess
The only thing I could think is that one of the LangChain or Llama-Index modules I am importing uses OpenAI by default for something.
Could it be in the sentence transfomations?
#### Import Data ####
import os
import time
import base64
import hashlib
from datetime import datetime
import pytz
from googleapiclient.discovery import build
from google_auth_oauthlib.flow import InstalledAppFlow
from google.auth.transport.requests import Request
from google.oauth2.credentials import Credentials
from googleapiclient.errors import HttpError
from googleapiclient.errors import Error
from email.message import EmailMessage
import logging
import openai
from llama_index import download_loader, LLMPredictor, PromptHelper, ServiceContext
from dotenv import load_dotenv
import sys
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from langchain.llms.base import LLM
from llama_index import GPTTreeIndex, SimpleDirectoryReader, LangchainEmbedding, GPTListIndex
from transformers import pipeline
from typing import Mapping, Any
import signal
from llama_index.indices.composability import ComposableGraph
Yeah I was already checking if it wasn't your embedding model or anything but that looks good too
I mean
INFO:sentence_transformers.SentenceTransformer:Load pretrained SentenceTransformer: sentence-transformers/all-mpnet-base-v2
INFO:sentence_transformers.SentenceTransformer:Use pytorch device: cpu
It looks like its loading in a HuggingFace model without problem
@Akinus21 are you using the OpenAI library for anything? It's in your imports
Only to provide the 'openai.api_key = os.getenv('OPEN_API_KEY')' line, hoping that would satisfy the error
oh I completely read over that
Alright I'm gonna run it too and I'll see what happens
note: the Google search id and api keys are no longer valid
You are able to build the indices, right? The exception happens at query time? @Akinus21
It doesnt error on this graph = build_index(prompt) right
It seems like. I have no error information indicating problems with building the indices. Although, I have before, but I think I need to solve the OpenAI key thing before I can get to that.
The background story here is:
I run this script on a server. It fires every 5 seconds. It queries a gmail inbox for messages with a particular pattern. I did not pay due attention, and long story short, it ran every 5 seconds for an entire day and errored out before it could delete the email so it just ran and ran. It would NOT have been an issue, except, it was querying ChatGPT without me knowing why, and so my cost was very high by the end of the day.
I cannot test by adding a valid OpenAI key, to see if there are any errors past that point, because I've reached the cap. So I have no idea if there are any errors in the code past the point that it throws the OpenAI error. I cannot figure out where that error is occurring, because I don't know what's calling it.
All this, sums up why I needed help in the discord, because it's beyond my skillset to figure this one out.
Very long shot, but could it be because of a line without service_context?
line 293: index = GPTTreeIndex.from_documents(documents)
I reckon that if you do not provide a service context it will re-create the TreeIndex everytime the attachments folder gets updated
That's the only thing I can think of right now honestly
Still got the same error, but what occurred to me is I might be able to figure out the calls if I uninstall the openai module:
Traceback (most recent call last):
File "/home/gabri/AkoGPT/AkoGPT_main.py", line 31, in <module>
from llama_index import download_loader, LLMPredictor, PromptHelper, ServiceContext
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/__init__.py", line 15, in <module>
from llama_index.embeddings.openai import OpenAIEmbedding
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/embeddings/openai.py", line 6, in <module>
import openai
ModuleNotFoundError: No module named 'openai'I hope it gives some more clarity because I'm kind of out of ideas too tbh
looks like it's in the embeddings somewhere in Llama-Index
What about specifying a custom model_name to the HuggingFaceEmbedding?
wait that doesnt make sense bc its huggingface
Maybe @Logan M has any idea...?
@OverclockedClock @Akinus21
Ok so the issue is, you are using a cutsomLLM and embeddings from huggingface, but it's still asking for an openai key?
I think you can reveal the issue pretty quickly by setting the openai key to some fake value. Then if it actually tries to call openai, it will error out and hopefully you can track down where/why it's being called
oh, it looks like @OverclockedClock suggested that already LOL and it didn't help track down the issue?
in the graph query configs, try passing in the service_context to each query_kwargs maybe?
@Logan M I can't find any documentation on that, would it simply be:
query_configs = [
{
"index_struct_type": "tree",
"query_mode": "embedding",
"query_kwargs": {
"child_branch_factor": 2
"service_context" = service_context
}
},
]?
Almost, just change it slightly and swap out the equals sign
"child_branch_factor": 2,
"service_context": service_context
I think I've narrowed it down to my CustomLLM class code
here's the error now:
Traceback (most recent call last):
File "/home/gabri/AkoGPT/AkoGPT_main.py", line 592, in <module>
get_emails()
File "/home/gabri/AkoGPT/AkoGPT_main.py", line 524, in get_emails
response = ask_gpt_custom(body_decoded)
File "/home/gabri/AkoGPT/AkoGPT_main.py", line 372, in ask_gpt_custom
graph = build_index(prompt)
File "/home/gabri/AkoGPT/AkoGPT_main.py", line 307, in build_index
index = GPTListIndex.load_from_disk(index_file)
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/indices/base.py", line 369, in load_from_disk
return cls.load_from_string(file_contents, kwargs) File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/indices/base.py", line 345, in load_from_string return cls.load_from_dict(result_dict, kwargs)
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/indices/base.py", line 322, in load_from_dict
return cls(index_struct=index_struct, docstore=docstore, **kwargs)
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/indices/list/base.py", line 54, in init
super().init(
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/indices/base.py", line 69, in init
self._service_context = service_context or ServiceContext.from_defaults()
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/indices/service_context.py", line 69, in from_defaults
llm_predictor = llm_predictor or LLMPredictor()
File "/home/gabri/.local/lib/python3.10/site-packages/llama_index/llm_predictor/base.py", line 164, in init
self._llm = llm or OpenAI(temperature=0, model_name="text-davinci-003")
File "pydantic/main.py", line 341, in pydantic.main.BaseModel.init
pydantic.error_wrappers.ValidationError: 1 validation error for OpenAI
Ohhh you need to pass in the service context when you load from disk 👀👀
Making progress! I passed the service_context for every index build, the graph build, and the graph query. Finally moving on to the next errors.
Thank you all for the help. I finally got it to not use OpenAI API key. You all were awesome in helping me.
Now I just have to figure out this error when I query the graph:
str type expected (type=type_error.str)
Wooo! Glad to hear you managed to fix it! What was using the OpenAI key in the end?
I was declaring my custom LLM wrong, so it was using the default LLM which is OpenAI's.
