Azure Open AI

Using Azure Open AI

1. Provision an endpoint in Azure, and the models you want to use.

2. Export and edit the provider, changing the URL as required.

3. Create a new secret only credential within Control Hub.

   • For support setting up a secret only credential see our guide on creating secret only credentials.

4. Update the provider with that correct path.

Troubleshooting

Sometimes the Tiktoken's tonkenizer is unable to resolve which causes an error on every call. "Failed to resolve 'openaipublic.blob.core.wi". This is caused by networking rules in certain environments.

The following steps resolve the issue by removing the need for the above call to occur in the first place.

1. Create a new folder named tiktoken_cache in any location.

   • E.g. C:\tiktoken_cache. We recommend having this outside of any InsightMaker folders.

2. Set an environment variable called TIKTOKEN_CACHE_DIR that points to the folder you created.

   1. Search "Edit the system environment variables on windows.

   2. Select Environment Variables...

   3. Within System Variables select New.

   4. Variable Name: Enter TIKTOKEN_CACHE_DIR

   5. Variable Value: Enter the path to the folder you created.

   6. Select Ok and Exit.

3. Download the tiktoken tokenizer. The tokenizer you need depends on the model you are running:

   • GPT 3.5 or 4: https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken

   • GPT 4o or 4o-mini: https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken

If you are can't download it from these links please try on an unrestricted network or hotspot.

1. Once downloaded, rename the tokenizer file:

   • From cl100k_base.tiktoken to 9b5ad71b2ce5302211f9c61530b329a4922fc6a4

   • From o200k_base.tiktoken to fb374d419588a4632f3f557e76b4b70aebbca790

     • These do not need an extension.

2. Move the downloaded and renamed item to the folder created in step 1.