mem0ai
relevant_memories in mem0/proxy/main.py misisng "result" key when using Ollama and chat completion
🐛 Describe the bug
When using mem0 with the chat completion feature with the following Ollama config
config = {
"llm": {
"provider": "ollama",
"config": {
"model": "llama3.1",
"temperature": 0.1,
"max_tokens": 128000,
"ollama_base_url": "http://localhost:11434",
}
},
"graph_store": {
"provider": "neo4j",
"config": {
"url": "neo4j://localhost:7687",
"username": "neo4j",
"password": "***"
},
"llm" : {
"provider": "ollama",
"config": {
"model": "llama3.1",
"temperature": 0.0,
"max_tokens": 128000,
"ollama_base_url": "http://localhost:11434",
}
}
},
"vector_store": {
"provider": "qdrant",
"config": {
"collection_name": "mem0-test",
"host": "localhost",
"port": 6333,
"embedding_model_dims": 768,
"on_disk": True,
}
},
"embedder": {
"provider": "ollama",
"config": {
"model": "nomic-embed-text",
"ollama_base_url": "http://localhost:11434",
}
},
"version": "v1.1",
}
The retrieved memories in relevant_memories that get sent to _format_query_with_memories are one layer further down under a key called "results". When I change the code on line 181 to go through all memories in relevant_memories['results'] everything works as expected.
- memories_text = "\n".join(memory["memory"] for memory in relevant_memories)
+ memories_text = "\n".join(memory["memory"] for memory in relevant_memories['results'])
full function code with change for completeness:
def _format_query_with_memories(self, messages, relevant_memories):
memories_text = "\n".join(memory["memory"] for memory in relevant_memories['results'])
return f"- Relevant Memories/Facts: {memories_text}\n\n- User Question: {messages[-1]['content']}"
Not sure how this should be handled properly to work with all other providers as I'm fairly new to Mem0, but thought I at least should register an issue since I couldn't find anything regarding this so far.
My code for reference:
Code
import logging
from mem0.proxy.main import Mem0
handler = logging.FileHandler(filename="ai-assitant.log", mode='a')
logging.basicConfig(handlers=[handler],
format='%(asctime)s,%(msecs)d %(name)s %(levelname)s %(message)s',
datefmt='%H:%M:%S',
level=logging.INFO)
config = {
"llm": {
"provider": "ollama",
"config": {
"model": "llama3.1",
"temperature": 0.1,
"max_tokens": 128000,
"ollama_base_url": "http://localhost:11434",
}
},
"graph_store": {
"provider": "neo4j",
"config": {
"url": "neo4j://localhost:7687",
"username": "neo4j",
"password": "***"
},
"llm" : {
"provider": "ollama",
"config": {
"model": "llama3.1",
"temperature": 0.0,
"max_tokens": 128000,
"ollama_base_url": "http://localhost:11434",
}
}
},
"vector_store": {
"provider": "qdrant",
"config": {
"collection_name": "mem0-test",
"host": "localhost",
"port": 6333,
"embedding_model_dims": 768,
"on_disk": True,
}
},
"embedder": {
"provider": "ollama",
"config": {
"model": "nomic-embed-text",
"ollama_base_url": "http://localhost:11434",
}
},
"version": "v1.1",
}
client = Mem0(config=config)
user_id = "aiquen"
# Ask the user for an input
message = input("Welcome to Mem0 AI-Assistant, how can I help you? > ")
while True:
# Use the input to re-create messages list each time
messages = [
{
"role": "user",
"content": f"{message}"
}
]
# Create a chat_completion
chat_completion = client.chat.completions.create(messages=messages, user_id=user_id, model="ollama/llama3.1")
# Print the answer from chat_completion
print(f"Assistant: {chat_completion.choices[0].message.content}")
message = input("> ")
@aiqueneldar I have raised the PR for this issue.
I also have a suggestion for your code: to print the AI's response in real-time, you might want to use the following approach:
for chunk in chat_completion:
if chunk.choices[0].delta.content is not None:
print(chunk.choices[0].delta.content, end="")
Instead of:
print(f"Assistant: {chat_completion.choices[0].message.content}")