Closed ElieTaillard closed 4 months ago
@ElieTaillard Our MistralAI connector is intended to be used with the MistralAI platform see: https://docs.mistral.ai/api/.
For Mistral models deployed to Azure we are working on a new connector. This is the tracking task #3992. This is one of our top priorities but we don't have an ETA yet.
Thank you for the information
Description: I encountered an issue while deploying the Mistral Large model on Azure through an AI hub. When using the method
builder.AddMistralChatCompletion(mistralModelName, mistralApiKey, mistralUri), "Mistral")
in my kernel builder, everything seems fine. However, when I call the methodGetStreamingChatMessageContentsAsync
, I receive a 405 Method Not Allowed error.Reproduction Steps:
builder.AddMistralChatCompletion(mistralModelName, mistralApiKey, mistralUri, "Mistral")
with the retrieved API key and URI in the kernel builder.GetStreamingChatMessageContentsAsync
in the application.Expected Behavior: The
GetStreamingChatMessageContentsAsync
method should call the correct endpoint and function without any errors.Actual Behavior: The application calls the endpoint
https://Mistral-large-****-serverless.*****.inference.ai.azure.com/chat/completions
and receives a 405 Method Not Allowed error.Workaround: I noticed that only the endpoint with
/v1/chat/completions
works. To circumvent this issue, I added/v1
to the URI parameter.Possible Cause: It appears there is an issue with the Mistral connector in the Semantic Kernel. The connector might be missing the /v1 segment in the endpoint path. Additionally, the deployed Mistral model might not accept requests without the /v1 in the URL.
Environment: