fix(config_base): Fixes Config class not being inheritted by PretrainedConfig.

archai/nlp/models/config_base.py

@@ -9,39 +9,22 @@ from typing import Any, Dict, Mapping, Optional, List, Union
 
 import torch
 
+from transformers import PretrainedConfig
 
-class Config:
+
+class Config(PretrainedConfig):
     """Base configuration class, used to define some common attributes
         and shared methods for loading and saving configurations.
 
     """
 
-    hyperparameter_map: Dict[str, str] = {}
-
-    def __getattribute__(self, key: str) -> Any:
-        if key != 'hyperparameter_map' and key in super().__getattribute__('hyperparameter_map'):
-            key = super().__getattribute__('hyperparameter_map')[key]
-        
-        return super().__getattribute__(key)
-
-    def __setattr__(self, key: str, value: Any) -> None:
-        if key in super().__getattribute__('hyperparameter_map'):
-            key = super().__getattribute__('hyperparameter_map')[key]
-
-        super().__setattr__(key, value)
-
     def __init__(self, **kwargs) -> None:
         """Initializes the class by verifying whether keyword arguments
             are valid and setting them as attributes.
 
         """
 
-        # Non-default attributes
-        for key, value in kwargs.items():
-            try:
-                setattr(self, key, value)
-            except AttributeError as e:
-                raise e
+        super().__init__(**kwargs)
 
     def _map_to_list(self,
                      variable: Union[int, float, List[Union[int, float]]],
@@ -58,14 +41,6 @@ class Config:
 
         return [variable] * size
 
-    def to_dict(self) -> Dict[str, Any]:
-        config_dict = {}
-
-        for key, value in self.__dict__.items():
-            config_dict[key] = value
-
-        return config_dict
-
 
 class SearchConfigParameter:
     """Base search configuration parameter class, used to define whether

archai/nlp/models/hf_gpt2/config_hf_gpt2.py

@@ -7,15 +7,15 @@
 from typing import List, Optional, Union
 
 from archai.nlp.models.config_base import Config, SearchConfigParameter, SearchConfig
-from transformers import CONFIG_MAPPING, PretrainedConfig
+from transformers import CONFIG_MAPPING
 
 
-class HfGPT2Config(Config, PretrainedConfig):
+class HfGPT2Config(Config):
     """Huggingface's Open AI GPT-2 default configuration.
 
     """
 
-    hyperparameter_map = {
+    attribute_map = {
         'n_token': 'vocab_size',
         'tgt_len': 'n_positions',
         'd_model': 'n_embd',
@@ -24,7 +24,7 @@ class HfGPT2Config(Config, PretrainedConfig):
         'dropatt': 'attn_pdrop',
         'weight_init_std': 'initializer_range'
     }
-    hyperparameter_map.update(CONFIG_MAPPING['gpt2']().attribute_map)
+    attribute_map.update(CONFIG_MAPPING['gpt2']().attribute_map)
 
     def __init__(self,
                  n_token: Optional[int] = 10000, # changed from 50257 for model's production

archai/nlp/models/hf_transfo_xl/config_hf_transfo_xl.py

@@ -7,21 +7,21 @@
 from typing import List, Optional
 
 from archai.nlp.models.config_base import Config, SearchConfigParameter, SearchConfig
-from transformers import CONFIG_MAPPING, PretrainedConfig
+from transformers import CONFIG_MAPPING
 
 
-class HfTransfoXLConfig(Config, PretrainedConfig):
+class HfTransfoXLConfig(Config):
     """Huggingface's Transformer-XL default configuration.
 
     """
 
-    hyperparameter_map = {
+    attribute_map = {
         'n_token': 'vocab_size',
         'weight_init_type': 'init',
         'weight_init_range': 'init_range',
         'weight_init_std': 'init_std',
     }
-    hyperparameter_map.update(CONFIG_MAPPING['transfo-xl']().attribute_map)
+    attribute_map.update(CONFIG_MAPPING['transfo-xl']().attribute_map)
 
     def __init__(self,
                  n_token: Optional[int] = 267736,