2. CRS Models at Module Level

Apart from loading models with an one-liner similar to the module-level example above, it is also flexible to create model variants by using different module-level combinations. Here we specify a UniCRS variant by using a new ChatGPT-based generator module to build a unicrs_model_variant:

from recwizard import FillBlankPipeline, FillBlankConfig, UnicrsRec, ChatgptGen

unicrs_model_variant = recwizard.AutoModel(
    config=FillBlankConfig(),
    rec_module=UnicrsRec.from_pretrained('RecWizard/unicrs-rec-redial'),
    gen_module=ChatgptGen.from_pretrained('RecWizard/chatgpt-gen-fillblank')
)

Then, this initialized unicrs_model_variant can be used in a similar way as Case 1:

unicrs_model_variant.response("Recommend me some popular and classic movies, \
                I like <entity> Titanic </entity>.")

We have the response from the UniCRS variant with ChatGPT generator module soon:

"Sure, if you enjoyed 'Titanic,' you might like these other popular and
     classic movie like The Shawshank Redemption (1994)"

So, this example shows that we can load the CRS modules first (they are typically recommender modules and generator modules). Then, feed the instantiated modules into a higher-level RecWizard model, where the model knows how to manage the usages and the results from different modules. If you are willing to know more about the model / module concepts in RecWizard, please check this concept section.