changelogs.mdx

title	description	icon
Changelogs		signs-post

v0.0.18

Currently under beta testing:

• 0.0.18-beta1

Model Chaining

Model chaining allows you to chain models together so that the output of one model can be the input to another model. For all the models in the chain, all the request and entities in the model input are the same.

ModelChainComponent

ChainedModelInput

This is the input for any models within a model chain. It's a child class of the ModelInput and has two additional properties:

• upstream_model_output: the output of the upstream model
• upstream_model_name: the name of the upstream model

Support dict as model output

Starting from v0.0.18, Wyvern will support serializable dictionary as model output.

If the model output is a dictionary, wyvern will generate the same number of model events as the number of keys in the dictionary. For each model event (ModelEventData), a new optional property model_key is added to reprent the key of in the dictionary. The ModelEventData.model_output will be the value of the key in the model output.

Model output cache

Now ModelComponent initialization can take cache_output as an optional input. If cache_output is set to True, the model output will be cached. The default value is True.

WyvernRequest has a new global map model_output_map which is used to cache model output. The key of the map is the model name. The value is another map (the inner map). They key of the inner map is the entity identifier and the value is the model output value.

Component.get_model_output

Within any component, now you can use self.get_model_output to get the model cached model output.