remove _class_register altogether

Signed-off-by: dafnapension <dafnashein@yahoo.com>

Author: dafnapension

.github/workflows/library_tests.yml

@@ -37,6 +37,7 @@ jobs:
     - run: curl -LsSf https://astral.sh/uv/install.sh | sh
     - run: uv pip install --upgrade --system torch --index-url https://download.pytorch.org/whl/cpu
     - run: uv pip install --system -c constraints.txt -e ".[tests]"
+    - run: uv pip install --system protobuf
     - run: |
         pip install --only-binary :all: spacy
 

docs/conf.py

@@ -116,7 +116,6 @@ def autodoc_skip_member(app, what, name, obj, would_skip, options):
         class_name = obj.__qualname__.split(".")[0]
         if (
             class_name
-            and Artifact.is_registered_class_name(class_name)
             and class_name != name
         ):
             return True

prepare/metrics/custom_f1.py

@@ -433,4 +433,7 @@ class NERWithoutClassReporting(NER):
     global_target=global_target,
 )
 
-add_to_catalog(metric, "metrics.ner", overwrite=True)
+if __name__ == "__main__" or __name__ == "custom_f1":
+    # because a class is defined in this module, need to not add_to_catalog just for importing that module in order to retrieve the defined class
+    # and need to prepare for case when this module is run directly from python (__main__) or, for example, from test_preparation (custom_f1)
+    add_to_catalog(metric, "metrics.ner", overwrite=True)

src/unitxt/artifact.py

@@ -1,4 +1,3 @@
-import difflib
 import inspect
 import json
 import os
@@ -23,7 +22,6 @@
     separate_inside_and_outside_square_brackets,
 )
 from .settings_utils import get_constants, get_settings
-from .text_utils import is_camel_case
 from .type_utils import isoftype, issubtype
 from .utils import (
     artifacts_json_cache,
@@ -134,21 +132,11 @@ def maybe_recover_artifacts_structure(obj):
     return obj
 
 
-def get_closest_artifact_type(type):
-    artifact_type_options = list(Artifact._class_register.keys())
-    matches = difflib.get_close_matches(type, artifact_type_options)
-    if matches:
-        return matches[0]  # Return the closest match
-    return None
-
 
 class UnrecognizedArtifactTypeError(ValueError):
     def __init__(self, type) -> None:
         maybe_class = type.split(".")[-1]
         message = f"'{type}' is not a recognized artifact 'type'. Make sure a the class defined this type (Probably called '{maybe_class}' or similar) is defined and/or imported anywhere in the code executed."
-        closest_artifact_type = get_closest_artifact_type(type)
-        if closest_artifact_type is not None:
-            message += f"\n\nDid you mean '{closest_artifact_type}'?"
         super().__init__(message)
 
 
@@ -161,7 +149,7 @@ def __init__(self, dic) -> None:
 
 
 class Artifact(Dataclass):
-    _class_register = {}
+    # _class_register = {}
 
     __type__: str = Field(default=None, final=True, init=False)
     __title__: str = NonPositionalField(
@@ -220,7 +208,7 @@ def fix_module_name_if_not_in_path(module):
         if file_components[0] == "":
             file_components = file_components[1:]
         file_components[-1] = file_components[-1].split(".")[0]  #omit the .py
-        if len(module.__package__) == 0:
+        if not getattr(module, "__package__", None) or len(module.__package__) == 0:
             return file_components[-1]
         package_components = module.__package__.split(".")
         assert all(p_c in file_components for p_c in package_components)
@@ -252,29 +240,9 @@ def get_module_class(cls, artifact_type:str):
         return artifact_type.rsplit(".", 1)
 
 
-    @classmethod
-    def register_class(cls, artifact_class):
-        assert issubclass(
-            artifact_class, Artifact
-        ), f"Artifact class must be a subclass of Artifact, got '{artifact_class}'"
-        assert is_camel_case(
-            artifact_class.__name__
-        ), f"Artifact class name must be legal camel case, got '{artifact_class.__name__}'"
-
-        if cls.is_registered_type(cls.get_artifact_type()):
-            assert (
-                str(cls._class_register[cls.get_artifact_type()]) == cls.get_artifact_type()
-            ), f"Artifact class name must be unique, '{cls.get_artifact_type()}' is already registered as {cls._class_register[cls.get_artifact_type()]}. Cannot be overridden by {artifact_class}."
-
-            return cls.get_artifact_type()
-
-        cls._class_register[cls.get_artifact_type()] = cls.get_artifact_type()   # for now, still maintain the registry from qualified to qualified
-
-        return cls.get_artifact_type()
 
     def __init_subclass__(cls, **kwargs):
         super().__init_subclass__(**kwargs)
-        cls.register_class(cls)
 
     @classmethod
     def is_artifact_file(cls, path):
@@ -284,18 +252,6 @@ def is_artifact_file(cls, path):
             d = json.load(f)
         return cls.is_artifact_dict(d)
 
-    @classmethod
-    def is_registered_type(cls, type: str):
-        return type in cls._class_register
-
-    @classmethod
-    def is_registered_class_name(cls, class_name: str):
-        for k in cls._class_register:
-            _, artifact_class_name = cls.get_module_class(k)
-            if artifact_class_name == class_name:
-                return True
-        return False
-
     @classmethod
     def get_class_from_artifact_type(cls, type:str):
         module_path, class_name = cls.get_module_class(type)
@@ -309,27 +265,20 @@ def get_class_from_artifact_type(cls, type:str):
         return klass
 
 
-
     @classmethod
     def _recursive_load(cls, obj):
         if isinstance(obj, dict):
-            new_d = {}
-            for key, value in obj.items():
-                new_d[key] = cls._recursive_load(value)
-            obj = new_d
+            obj = {key: cls._recursive_load(value) for key, value in obj.items()}
+            if cls.is_artifact_dict(obj):
+                try:
+                    artifact_type = obj.pop("__type__")
+                    artifact_class = cls.get_class_from_artifact_type(artifact_type)
+                    obj = artifact_class.process_data_after_load(obj)
+                    return artifact_class(**obj)
+                except (ImportError, AttributeError) as e:
+                    raise UnrecognizedArtifactTypeError(artifact_type) from e
         elif isinstance(obj, list):
-            obj = [cls._recursive_load(value) for value in obj]
-        else:
-            pass
-        if cls.is_artifact_dict(obj):
-            cls.verify_artifact_dict(obj)
-            try:
-                artifact_type = obj.pop("__type__")
-                artifact_class = cls.get_class_from_artifact_type(artifact_type)
-                obj = artifact_class.process_data_after_load(obj)
-                return artifact_class(**obj)
-            except (ImportError, AttributeError) as e:
-                raise UnrecognizedArtifactTypeError(artifact_type) from e
+            return [cls._recursive_load(value) for value in obj]
 
         return obj
 
@@ -389,7 +338,7 @@ def verify_data_classification_policy(self):
 
     @final
     def __post_init__(self):
-        self.__type__ = self.register_class(self.__class__)
+        self.__type__ = self.__class__.get_artifact_type()
 
         for field in fields(self):
             if issubtype(

src/unitxt/catalog/cards/safety/airbench2024.json

@@ -1,15 +1,15 @@
 {
-    "__type__": "task_card",
+    "__type__": "unitxt.card.TaskCard",
     "loader": {
-        "__type__": "multiple_source_loader",
+        "__type__": "unitxt.loaders.MultipleSourceLoader",
         "sources": [
             {
-                "__type__": "load_hf",
+                "__type__": "unitxt.loaders.LoadHF",
                 "path": "stanford-crfm/air-bench-2024",
                 "name": "default"
             },
             {
-                "__type__": "load_hf",
+                "__type__": "unitxt.loaders.LoadHF",
                 "path": "stanford-crfm/air-bench-2024",
                 "data_files": {
                     "judge_prompts": "judge_prompt_final.csv"
@@ -22,7 +22,7 @@
     },
     "preprocess_steps": [
         {
-            "__type__": "select_fields",
+            "__type__": "unitxt.operators.SelectFields",
             "fields": [
                 "cate-idx",
                 "judge_prompt"
@@ -32,7 +32,7 @@
             ]
         },
         {
-            "__type__": "join_streams",
+            "__type__": "unitxt.stream_operators.JoinStreams",
             "left_stream": "test",
             "right_stream": "judge_prompts",
             "how": "inner",
@@ -43,7 +43,7 @@
         }
     ],
     "task": {
-        "__type__": "task",
+        "__type__": "unitxt.task.Task",
         "input_fields": {
             "cate-idx": "str",
             "l2-name": "str",
@@ -59,10 +59,10 @@
         ]
     },
     "templates": {
-        "__type__": "templates_dict",
+        "__type__": "unitxt.templates.TemplatesDict",
         "items": {
             "default": {
-                "__type__": "input_output_template",
+                "__type__": "unitxt.templates.InputOutputTemplate",
                 "input_format": "{prompt}\n",
                 "output_format": ""
             }

src/unitxt/catalog/formats/chat_api_with_tokenizer_chat_template.json

@@ -1,4 +1,4 @@
 {
-    "__type__": "hf_system_format",
+    "__type__": "unitxt.formats.HFSystemFormat",
     "model_name": "ibm-granite/granite-3.1-2b-instruct"
 }

src/unitxt/register.py

@@ -1,9 +1,7 @@
-import importlib
-import inspect
 import os
 from pathlib import Path
 
-from .artifact import Artifact, Catalogs
+from .artifact import Catalogs
 from .catalog import EnvironmentLocalCatalog, GithubCatalog, LocalCatalog
 from .error_utils import Documentation, UnitxtError, UnitxtWarning
 from .settings_utils import get_constants, get_settings
@@ -89,27 +87,6 @@ def _reset_env_local_catalogs():
             _register_catalog(EnvironmentLocalCatalog(location=path))
 
 
-def _register_all_artifacts():
-    dir = os.path.dirname(__file__)
-    file_name = os.path.basename(__file__)
-
-    for file in os.listdir(dir):
-        if (
-            file.endswith(".py")
-            and file not in constants.non_registered_files
-            and file != file_name
-        ):
-            module_name = file.replace(".py", "")
-
-            module = importlib.import_module("." + module_name, __package__)
-
-            for _name, obj in inspect.getmembers(module):
-                # Make sure the object is a class
-                if inspect.isclass(obj):
-                    # Make sure the class is a subclass of Artifact (but not Artifact itself)
-                    if issubclass(obj, Artifact) and obj is not Artifact:
-                        Artifact.register_class(obj)
-
 
 class ProjectArtifactRegisterer(metaclass=Singleton):
     def __init__(self):
@@ -118,7 +95,6 @@ def __init__(self):
 
         if not self._registered:
             _register_all_catalogs()
-            _register_all_artifacts()
             self._registered = True