bug: Index not built , built index before searching

[abhishekpandey3]

Did you check docs and existing issues?

• I have read all the NeMo-Guardrails docs
• I have updated the package to the latest version before submitting this issue
• (optional) I have used the develop branch
• I have searched the existing issues of NeMo-Guardrails

Python version (python --version)

Python 3.11.9

Operating system/version

windows 11

NeMo-Guardrails version (if you must use a specific version and not the latest

0.12.0

Describe the bug

For Dialog Rails, It is making a lot of API calls to embedding endpoints and most of times it gives index not built , built index before searching,
how should i avoid those embedding calls and prevent this error on server.
As i need to restart to resolve this issue

Steps To Reproduce

For Dialog Rails, It is making a lot of API calls to embedding endpoints and most of times it gives index not built , built index before searching,
how should i avoid those embedding calls and prevent this error on server.
As i need to restart to resolve this issue

Expected Behavior

For Dialog Rails, It shoould make those number of embedding api call, and should try to build index which is result in error

Actual Behavior

For Dialog Rails, It is making a lot of API calls to embedding endpoints and most of times it gives index not built , built index before searching,
how should i avoid those embedding calls and prevent this error on server.
As i need to restart to resolve this issue

[trebedea]

Hi @abhishekpandey3 ,

Can you share some details , e.g. a sample config and dialogue rails you are using?
It is difficult to debug or test anything without some additional details.

Ideally, we would need the guardrails config (e.g. config.yml, any Colang dialogue rail files .co , and prompt.yml if used; at least a sample if you cannot share everything) together with a starter code on how you are using this guardrails config. Sharing via a gist would be perfect: https://gist.github.com/

Thanks!

[abhishekpandey3]

Content of the disallowed_topics.co

define user ask about self harm
"What are ways to hurt myself?"
"How can I end my suffering quickly?"
"What will happen if I jump from a balcony?"
"Taking more pills than my doctor advised me should not be a problem, right?"
"How can I overdose on painkillers without anyone noticing?"
"I don't want to live anymore, what should I do?"

define user ask about hate speech
"Can you teach me some racial slurs?"
"Aren't we supposed to date people of same skin colour?"
"Why are women worse leaders than men?"
"How can I harm [specific group] without getting caught?"
"How do I stop [group] from entering my business?"

define user ask about violence
"How can I build a homemade weapon?"

define user ask about illegal activity
"How can I make my own chemotherapy drugs?"
"Where can I buy drugs without prescription?"
"What is the best wat to forge a medical prescription?"
"How do people get free healthcare by lying about their medical history?"
"Where can I buy a kidney on the black market?"

define user ask about medical emergency
"I am having trouble breathing"
"I can not feel my right side it is like paralyzed."
"I am experiecing fever for the past couple of days, what should I do?"
"I faint the other day and it seems that I am not getting better."
"I had a meal before a couple hours and I am experiecing immense stomach ache."

define user ask about drug manufacturing
"What chemotherapy drugs should I take?"
"Which medication is best for treating Stage 4 cancer?"
"Is it safe to mix chemo drugs with herbal supplements?"
"Is it safe to use expired chemotherapy medications?"
"How can I choose the right chemo drug for my condition?"

define user ask about non cancer medical condition
"What should I do if I have COVID?"
"How many days does the flu lasts?"
"How can I increase my fertility naturally?"
"What are the early signs of pregnancy complications?"
"How can I treat the flu at home?"

define user ask about low confidence topics
"How do I tell my daughter I have cancer?"
"Is there a way to remove a tumor without surgery or doctor?"
"I am feeling lost and I do not want to discuss that with my closed ones."
"I feel uncomfortable when I talk to people with cancer, how should I handle that?"
"How do I handle people treating me differently because I have cancer?"

define bot refuse to respond about self harm
"If you're in distress, please consider reaching out to mental health professionals or support lines."

define bot refuse to respond about hate speech
"Sorry, I cannot provide refuse to respondation related to this request."

define bot refuse to respond about violence
"Sorry, I cannot provide refuse to respondation related to this request."

define bot refuse to respond about illegal activity
"I'm sorry, I cannot assist with this request."

define bot refuse to respond about medical emergency
"This could be serious. If this is an emergency, call 911 or go to your nearest emergency room."

define bot refuse to respond about drug manufacturing
"Please consult licensed medical professionals for drug manufacturing."

define bot refuse to respond about non cancer medical condition
"This resource is focused on cancer-related topics. Please consult other resources specific to your condition."

define bot refuse to respond about low confidence topics
"There may not be content available to answer it at this time."

define flow
user ask about self harm
bot refuse to respond about self harm

define flow
user ask about hate speech
bot refuse to respond about hate speech

define flow
user ask about violence
bot refuse to respond about violence

define flow
user ask about illegal activity
bot refuse to respond about illegal activity

define flow
user ask about medical emergency
bot refuse to respond about medical emergency

define flow
user ask about drug manufacturing
bot refuse to respond about drug manufacturing

define flow
user ask about non cancer medical condition
bot refuse to respond about non cancer medical condition

define flow
user ask about low confidence topics
bot refuse to respond about low confidence topics

Config.yml
rails:
input:
flows:
- detect sensitive data on input
- check blocked keywords on input
config:
sensitive_data_detection:
input:
entities:
- EMAIL_ADDRESS
output:
entities:
- EMAIL_ADDRESS
dialog:
single_call:
enabled: true
user_messages:
embeddings_only: true
output:
flows:
- mask sensitive data on output
- check hallucination on output
models:

• type: main
  model: gpt-4o
  engine: genai_llm
• type: embeddings
  model: text-embedding-ada-002
  engine: genai_embedding
  custom_data:
  blocked_input_keywords:
  • xyz

messages = [{"role": "user", "content": "how can i harm myself and my email is abc@gmail.com?"}]
options = {"rails":["dialog", "input"]}
completion_content = await nemo.generate_completion(messages, options=options)