dosubot[bot]
commented
1 month ago Hey @theconnectionist, I'm here to help you with any bugs, questions, or contributions you have. Let's tackle this together!
The error "TypeError: 'coroutine' object is not iterable" occurs because the code is trying to iterate over a coroutine object, which is not directly iterable. This typically happens when an asynchronous function is called without awaiting its result.
In your case, the issue likely arises because the OllamaEmbedding class's asynchronous methods are not being awaited properly when use_async is set to True.
To resolve this issue, ensure that the asynchronous methods are awaited correctly. Here is a general approach to fix this:
-
Ensure that the embedding model's asynchronous methods are awaited:
- If
OllamaEmbeddinghas asynchronous methods likeaget_text_embedding_batch, make sure they are awaited properly.
- If
-
Modify the
PropertyGraphIndexinitialization to handle asynchronous embedding:- Ensure that the embedding process within
PropertyGraphIndexcorrectly handles asynchronous calls.
- Ensure that the embedding process within
Here is an example of how you might modify the embedding process to handle asynchronous calls:
from llama_index.embeddings.ollama import OllamaEmbedding
from llama_index.core import PropertyGraphIndex
import asyncio
# Initialize OllamaEmbedding
ollama_embedding = OllamaEmbedding(
model_name="gemma2:2b-text-q2_K",
base_url="http://localhost:11434",
ollama_additional_kwargs={"mirostat": 0}
)
# Ensure async embedding is handled properly
async def create_index():
index = PropertyGraphIndex(
nodes=nodes,
property_graph_store=GraphRAGStore(),
kg_extractors=[kg_extractor],
show_progress=True,
embed_model=ollama_embedding,
embed_kg_nodes=True,
use_async=True
)
return index
# Run the async function
index = asyncio.run(create_index())By properly awaiting asynchronous methods, you can avoid the "TypeError: 'coroutine' object is not iterable" error. This approach ensures that the embedding process is handled asynchronously and correctly integrated into the PropertyGraphIndex initialization.
To continue talking to Dosu, mention @dosu.
logan-markewich
theconnectionist
mribbons
Question Validation
Question
I'm running GraphRAG_v1.ipynb. I ran it successfully with the OpenAI llm and embedding models (i.e all I did was provide my openai key). Next I tried the same notebook with Ollama based llm and embedding models with a locally running server.
The Ollama llm had no problem running as part of the notebook. Next I tried using OllamaEmbedding, initialized as follows:
and tested/verified as follows:
Next I used ollama_embedding in PropertyGraphIndex:
That is consistently erroring out as shown below. The only way to work around this is to set use_async=False. However, I imagine that will slow down my indexing operations. I couldn't find anything relevant in previous issues or at discord.
My Question: Why is this error occurring when use_async is True. Did I do something wrong? Or, is it is a bug in the underlying implementation?