Can't get Azure GPT Model to work for TTYG ChatGPT Retrival Plugin

Question

I can't get my Azure OpenAI configured correctly, while I'm able to get everything (for the Talk To Your Graph Features, TTYG) working with my OpenAI API key. I can also confirm that my Azure OpenAI key/config is working for other tools.

Instructions I've followed:

• https://graphdb.ontotext.com/documentation/10.6/talk-to-graph.html?highlight=lab#what-s-in-this-document
• https://graphdb.ontotext.com/documentation/10.6/gpt-queries.html#configuring
• https://github.com/Ontotext-AD/chatgpt-retrieval-plugin?tab=readme-ov-file#quickstart

Here is the Azure OpenAI set up and error message. (When I replace the configuration with my OpenAI key, it works.)

Shell script variables:

export BEARER_TOKEN="<my bearer token>

export OPENAI_API_KEY="<my azure api key>"

export OPENAI_API_BASE="https://test-1.openai.azure.com/"

export OPENAI_API_TYPE="azure"

export OPENAI_EMBEDDINGMODEL_DEPLOYMENTID="test-1-text-embedding-ada-002"

GraphDB java properties:

graphdb.gpt.token: <my azure api key>

graphdb.gpt.model: test-1-text-embedding-ada-002

graphdb.gpt.url: https://test-1.openai.azure.com/openai/deployments/test-1-text-embedding-ada-002/chat/completions?api-version=2023-03-15-preview

graphdb.gpt.auth: api-key

Error message:

See attachment and below, I was unable to find the complete message in the logs. When I replace the above with my OpenAI key and bearer token, it works.

error message

"Handler dispatch failed; nested exception is. Could not initialize class com.graphdb.gpt.Gpt..."

Answer 1

This is because the values for graphdb.gpt.model - at least as of 10.6.3 are checked against a rather tight set of predetermined models.

You don't need this, as you are setting the model inside the graphdb.gpt.url anyway:

graphdb.gpt.url: https://test-1.openai.azure.com/openai/deployments/test-1-text-embedding-ada-002/chat/completions?api-version=2023-03-15-preview

The same setup should work if you remove the graphdb.gpt.model parameter.