search

NVIDIA / NeMo-Guardrails

NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.
Other
4.02k stars 367 forks source link

Issue with azure openAI Api key #470

Open Pratekh opened 5 months ago

Pratekh commented 5 months ago

response = rails.generate(messages=[{ "role": "user", "content": "What is the capital of France?" }]) print(response["content"])

WARNING: nemoguardrails.actions.action_dispatcher:Error while execution generate_user_intent: Error code: 404 - {'error': {'code': '404', 'message': 'Resource not found'}}

I'm sorry, an internal error has occurred.

drazvan commented 5 months ago

@Pratekh : can you provide more details on the config (e.g., the content of config.yml)? This is probably a configuration issue.

Pratekh commented 5 months ago 
%%writefile config/config.yml
type: main
engine: azure
model: gpt-35-turbo-16k
parameters:
    azure_endpoint: https://abc.openai.azure.com/
    api_version: 2023-07-01-preview
    deployment_name: abc
    api_key: 00000000000000000000000
Pratekh commented 5 months ago

How can I fix this issue?

rohitk-cognizant commented 4 months ago

Any update on this issue, we are also using Azure Open AI, getting same error while running nemoguardrails server

drazvan commented 4 months ago

@rohitk-cognizant : to debug this, can you try to get a functional code where you initialize the LLM separately, in a "pure LangChain way". 

model = AzureOpenAI(...)
print(model.invoke("test prompt")

And then try to pass that directly to the LLMRails instance

rails = LLMRails(config=config, llm=model)
response = rails.generate(messages=[{
"role": "user",
"content": "What is the capital of France?"
}])
print(response["content"])

If this works, then I can point you to where exactly you can check how the AzureOpenAI engine is initialized and we can check the difference in parameters.