Add support for Azure OpenAI for local deployment

[jexp]

They should be compatible with the OpenAI models so nothing to change in the code, just in the utility function where we create the LLM

https://python.langchain.com/v0.1/docs/integrations/chat/azure_chat_openai/

support these environment variables

AZURE_OPENAI_API_KEY="..."
AZURE_OPENAI_ENDPOINT="https://<your-endpoint>.openai.azure.com/"
AZURE_OPENAI_API_VERSION="2023-06-01-preview"
AZURE_OPENAI_CHAT_DEPLOYMENT_NAME="xxx"

[arthcras]

To add this option to the knowledge graph builder would definitely help users to solve data confidentially issues

[ionuttiplea]

I did replace the openAI model with AzureOpenAI locally and it worked I guess that making an integration to specifically support Azure OpenAI ( at least for those running the app locally ) would be great :)

[jexp]

Yes it’s on the plan. mostly additional config for baseurl, version, deployed model

[maxgosk]

I did replace the openAI model with AzureOpenAI locally and it worked I guess that making an integration to specifically support Azure OpenAI ( at least for those running the app locally ) would be great :)

Hi! Could you share more how you did this please, thank you

[ionuttiplea]

I did replace the openAI model with AzureOpenAI locally and it worked I guess that making an integration to specifically support Azure OpenAI ( at least for those running the app locally ) would be great :)

Hi! Could you share more how you did this please, thank you

Sure! Disclaimer though, I didn't think of extending what exists with AzureOpenAI, I only replaced the model used for GPT

Is quite simple in /backend/src/shared/common_fn.py , line 135 I replaced the openAI model with:

elif "gpt" in model_version: llm = AzureChatOpenAI( openai_api_version=os.environ["AZURE_OPENAI_API_VERSION"], azure_deployment=os.environ["AZURE_OPENAI_CHAT_DEPLOYMENT_NAME"] )

Also you would require the following .env variables, if you used Azure Open AI through langchain you could reuse that config

AZURE_OPENAI_ENDPOINT=AZURE_OPENAI_DEPLOYMENT_BASE_URL AZURE_OPENAI_API_KEY=AZURE_OPENAI_API_KEY AZURE_OPENAI_CHAT_DEPLOYMENT_NAME=GPT_4O # The name of the deployment as you find it in Azure

MODEL_NAME=gpt-4o # the name of the model deployed

AZURE_OPENAI_API_VERSION=2024-02-15-preview # this is quite tricky to get right, you should find it on your deployment details

AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT_NAME=ada-002 # optional, can be used for embeddings OPENAI_API_TYPE=azure # I think this one is the same for everyone

[maxgosk]

I did replace the openAI model with AzureOpenAI locally and it worked I guess that making an integration to specifically support Azure OpenAI ( at least for those running the app locally ) would be great :)

Hi! Could you share more how you did this please, thank you

Sure! Disclaimer though, I didn't think of extending what exists with AzureOpenAI, I only replaced the model used for GPT

Is quite simple in /backend/src/shared/common_fn.py , line 135 I replaced the openAI model with:

elif "gpt" in model_version: llm = AzureChatOpenAI( openai_api_version=os.environ["AZURE_OPENAI_API_VERSION"], azure_deployment=os.environ["AZURE_OPENAI_CHAT_DEPLOYMENT_NAME"] )

Also you would require the following .env variables, if you used Azure Open AI through langchain you could reuse that config

AZURE_OPENAI_ENDPOINT=AZURE_OPENAI_DEPLOYMENT_BASE_URL AZURE_OPENAI_API_KEY=AZURE_OPENAI_API_KEY AZURE_OPENAI_CHAT_DEPLOYMENT_NAME=GPT_4O # The name of the deployment as you find it in Azure

MODEL_NAME=gpt-4o # the name of the model deployed

AZURE_OPENAI_API_VERSION=2024-02-15-preview # this is quite tricky to get right, you should find it on your deployment details

AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT_NAME=ada-002 # optional, can be used for embeddings OPENAI_API_TYPE=azure # I think this one is the same for everyone

Thank you I make it work earlier doing the same!