Hey guys I am encountering a weird issue

That's strange! What is the version that you are trying with?

Latest one from Github

I could fix it minutes ago with this inside _get_prefix_messages_with_context:

context_str = context_str_template.replace("{system_prompt}", system_prompt)
return [ChatMessage(content=context_str, role="system")] + prefix_messages

but I'm not a professional engineer tbh, so you might have better solutions, I used nice prompt engineering with ChatGPT to address the issue.

It should be like this only.
context_str = context_str_template.format(system_prompt=system_prompt)

If it is working in some responses and for some cases it is throwing error. I believe there could be some missing part in the library.


Can you share your code. I can try to check

Yea would be nice if this was repoducabile locally 🤔 Seems very weird

So, I am working on a Slack app.

This is my chat.py:

def initialize_chat_engine(user_id):
yerevan_project_index, project_name = user_engines.get(user_id, (None, None))
if yerevan_project_index is None:
raise ValueError("Yerevan account's settings have not been configured for this user.")

## CONTEXT MODE
memory = ChatMemoryBuffer.from_defaults(token_limit=2000)
chat_engine = yerevan_project_index.as_chat_engine(
chat_mode="context",
memory=memory,
service_context = service_context,
refine_template=DEFAULT_REFINE_PROMPT,
system_prompt=(f'Your name is Yerevan, who assisting only with questions related to {project_name}')
response_synthesizer=response_synthesizer,
similarity_top_k=5
)
return chat_engine

This is my slack application's snippet were messages are handled:

def process_event(data):
logging.debug("Processing event...")
text = data['event'].get('text')
channel_id = data['event'].get('channel')
user_id = data['event'].get('user')

if not text or not channel_id or not user_id:
return


user_workspace = get_workspace_name(user_id)
user_name = get_user_name(user_id)

with user_data_lock:
if user_id not in user_data:
message_text = "Please connect to your Yerevan account."
try:
client.chat_postMessage(channel=channel_id, text=message_text)
except SlackApiError as e:
logging.error(f"Slack API error occurred: {e}")
return
else:
cipher_suite = Fernet(user_data[user_id]['key'].encode('utf-8'))

try:
chat_engine = initialize_chat_engine(user_id)
except Exception as e:
logging.error(f"Error occurred while initializing chat engine: {e}")
return

response = chat_engine.chat(text)
logging.debug(f"Response: {response}")

if isinstance(response, AgentChatResponse):
message_text = response.response
else:
message_text = "Unexpected response type."


record_interaction(user_workspace, user_name, text, message_text)

try:
client.chat_postMessage(channel=channel_id, text=message_text)
except SlackApiError as e:
logging.error(f"Slack API error occurred: {e}")

ChatGPT's response that worked for me:

Thank you for providing the output. It seems that the context_str_template looks correct with the appropriate placeholders {context_str} and {{system_prompt}}. The error is likely happening because of how the format method is used in the _get_prefix_messages_with_context function.

In Python, when you have double curly braces {{}} in a format string, it's used to escape the curly braces and treat them as literal characters in the formatted string. So, {{system_prompt}} becomes {system_prompt} in the formatted string. When you later try to format the context_str_template again, Python is looking for a key named system_prompt in the template string. This seems to be causing the issue.

You can modify the code to handle this by using the formatted string once, and replace the placeholder with the actual system prompt. Here's the modified _get_prefix_messages_with_context method:



Try making this change and test the code again to see if the error persists.