This PR aims to greatly disambiguate code logic by using distinct naming for various object. Currently there are 10+ names like embedding_xxx for class, object, dict and files with a lot of clashes.
With this change:
Use catalog to replace embeddings_config object and represent the all available (may not be used) model/index config
Use registry to replace embedddings_map object which holds all the model/client objects
Use env to replace embeddings_spec which is the additional config per environment
Make all the field naming the same across yaml file and dataclass object. This will help for parsing and error checking.
Use A and A_dict to differentiate dataclass object vs yaml dict.
shifucun
commented
1 month ago
This PR aims to greatly disambiguate code logic by using distinct naming for various object. Currently there are 10+ names like embedding_xxx for class, object, dict and files with a lot of clashes.
With this change:
catalogto replaceembeddings_configobject and represent the all available (may not be used) model/index configregistryto replaceembedddings_mapobject which holds all the model/client objectsenvto replaceembeddings_specwhich is the additional config per environment