SOLVED

Bot Framework Dispatch CLI replacement with Orchestrator

Copper Contributor

Hey everyone,

 

As the Dispatch CLI will be deprecated on December 31, 2021, we've started evaluating Orchestrator as an alternative to it. Information about Dispatch CLI here.

 

We were using Dispatch CLI basically as a high-level router for our bot that will route the occurrence (text) to the correct resource (Luis, QnA) and produce a resource-specific intent. Also, we are retraining it periodically so the dispatcher will always have the up-to-date data from all resources so it will route to the correct resource.

 

So, wanted to ask – what is the strategy of using an orchestrator in the current flow? We are able to retrain it with BF CLI but we don't wanna restart the bot after refreshing the orchestrator. Also, how can we use the same trained orchestrator model in multiple bots? (as I understood it's not a trained Luis app)

 

7 Replies

@MykolaMedynskiy  - We will also check it from our end and let you know.

@MykolaMedynskiy

Could you please elaborate your requirement and share the repro steps if any?

Hey @Nivedipa-MSFT,

 

So the requirements are:

– periodically refresh orchestrator model

– use it in the bot for intent recognition

 

And the problem is refreshing the model without restarting the application, since the orchestrator model is not a Luis App, as was with Dispatch.

@MykolaMedynskiy 

We are checking this internally.We will get back to you once we have any update.

@Nivedipa-MSFT we've checked all the documentation and haven't found the way to implement it.

We can see that there is a refresh option, but how to use it and not affect the bot application files? Can it be kept separately from the bot project and be queried via some API?

best response confirmed by MykolaMedynskiy (Copper Contributor)
Solution

@MykolaMedynskiy 

1.Orchestrator is a local resource, so is deployed to the target environment. It should require significantly less training than a LUIS model, so should really only be updated along with your source code. But yes, since the model is loaded into memory you need to restart the bot.

2.Deploy the trained model to the desired environments

 

Please go through below doc for implementation details:

https://docs.microsoft.com/en-us/azure/bot-service/bot-builder-tutorial-orchestrator

 

Thanks, 

Nivedipa

------------------------------------------------------------------------------------------------------------------------ 

If the response is helpful, please click "**Mark as Best Response**" and like it. You can share your feedback via Microsoft Teams Developer Feedback link. Click here to escalate. 

 

1 best response

Accepted Solutions
best response confirmed by MykolaMedynskiy (Copper Contributor)
Solution

@MykolaMedynskiy 

1.Orchestrator is a local resource, so is deployed to the target environment. It should require significantly less training than a LUIS model, so should really only be updated along with your source code. But yes, since the model is loaded into memory you need to restart the bot.

2.Deploy the trained model to the desired environments

 

Please go through below doc for implementation details:

https://docs.microsoft.com/en-us/azure/bot-service/bot-builder-tutorial-orchestrator

 

Thanks, 

Nivedipa

------------------------------------------------------------------------------------------------------------------------ 

If the response is helpful, please click "**Mark as Best Response**" and like it. You can share your feedback via Microsoft Teams Developer Feedback link. Click here to escalate. 

 

View solution in original post