Michel Farhi-Chevillard 

We think it should be fairly trivial to support this but have not done so yet.  We don't have full access to test it yet, but are working on it and will advise you in due course.

Thank you for the quick answer.  Let us know if we can help test it when you are ready.

Hi Paul ,  in case you wanted some explanation of why a large company might prefer to use an Azure implementation of OpenAI, this is what our Chief Engineer explained to me today:

• Azure provides a LOT more security guarantees and enterprise features that we need to use in a production setting
• For example, we have written agreements with Microsoft on how they manage our data
• And they have stronger SLAs
• User management, billing, etc -- all comes default with Azure and already connected to our systems
• Azure OpenAI is exactly the same API as OpenAI, just a different URL and a different token style (for example, the one I gave you instead of the platform.openai.com one)

I am assuming we are not alone in preferring an Azure implementation and that RWS agrees with the need.    Thank you!

 Michel Farhi-Chevillard 

Yep… 

Michel Farhi-Chevillard 

This plugin has been updated and also considerably enhanced.  Users who have access to Azure OpenAI will be able to to configure the plugin to work with this as well.

Using the API change was fairly trivial, but the work took a bit longer because the developer ensured it would be easier to add more providers as they become available in the future, improved keyboard support and also ensured the app would be accessible for visually impaired users... something that will become very important for all developers to follow soon.

Paul , I was given an endpoint and a key by the AI team in my company.  However, when I paste that information into the dialog to configure a model, the "Test Connection" button stays greyed out. 


What am I doing wrong?  Is there something not working in the endpoint URL I was given? At what point does the "Test Connection" button become active?

Hi Michel Farhi-Chevillard , you'll need to provide the complete endpoint to make calls to the service, including the deployment name and api-version.  Your AI team should be able to recover this from the https://oai.azure.com/portal portal

Example:

https://MyDomainName.openai.azure.com/openai/deployments/MyDeployment/chat/completions?api-version=2023-07-01-preview

Thank you for this additional information.  I have contacted our chief engineer with this and will hopefully get it working soon.    You may want to add this useful information to your documentation page at community.rws.com/.../openai-translator.

Michel Farhi-Chevillard our documentation has moved to the RWS AppStore, as you are right the information is important. See here: https://appstore.rws.com/Plugin/200?tab=documentation

I will place something within the Wiki to inform users accordingly. Thank you

I was able to successfully connect today and had fun with the tool.  Here are some experiments I made, where the base translation was provided by DeepL and the corrected one by GPT-4 + custom prompt!

1. The “funny” prompt:
   
   
   
   
   
2. The “French Canadian” variant prompt:
   
   
   
   
   
    
3. The old-fashioned translation style:
   
   

4. The summarization followed by translation:
   
   
   
   
   
5. The hallucination because the prompt is asking for something nearly impossible!