Hi there!
From docs “The recipe key allows for different types of config and model architecture. Currently, only “default.v1” is supported.”
So currently we can only use one recipe and have to create different configs with the same recipe?
Hi there!
From docs “The recipe key allows for different types of config and model architecture. Currently, only “default.v1” is supported.”
So currently we can only use one recipe and have to create different configs with the same recipe?
Can you please give more context?
Hi,
Rasa 3.0 as this is a new implementation… written here here. Why? Using different configs…
@statimo It is also mentioned in the doc “Currently, only “default.v1” is supported.” so 1 config and 1 default.v1 is supported. But you can use I guess, give it a try. I will personally not recommend that.
Hi @statimo what this means is that only one type of configuration is supported. This has not changed from previous versions, currently the type of configuration looks something like this:
pipeline:
- <tokenizer>
- <featurizer>
- ...
- <entity_extractor>
- ...
- <intent_classifier>
- ...
policies:
- <policy1>
- <policy2>
- ...
You can still have many different configurations that follow this pattern, as you did before. Any configuration that worked in the previous versions will also work in Rasa 3.x (provided that the format and deprecated components are adjusted accordingly).
The reason we defined this type of configuration is to allow us in future to define different styles of configuration (where a style is defined by a recipe). This has not happened yet, but enabling this enables quicker prototyping and research
Hope that makes sense, please ask followup questions if something is unclear!