Abstracting model code (#36)

* Initial commit for refactoring skip model code using class factories and abstract classes.

Signed-off-by: Kira Selby <kaselby@uwaterloo.ca>

* Minor fixes

Signed-off-by: Kira Selby <kaselby@uwaterloo.ca>

* Add missing lines

Signed-off-by: Kira Selby <kaselby@uwaterloo.ca>

---------

Signed-off-by: Kira Selby <kaselby@uwaterloo.ca>

Author: kaselby

run_benchmark.py

@@ -397,7 +397,7 @@ def main():
     checkpoint = config._name_or_path
     tokenizer = AutoTokenizer.from_pretrained(checkpoint, trust_remote_code=True)
     tokenizer.pad_token = tokenizer.eos_token
-    model_name = checkpoint.split("-")[0].capitalize()
+    model_name = checkpoint.split("/")[1].split("-")[0].capitalize()
 
     # Get test prompts
     test_prompts = get_diverse_test_prompts()

src/configuration_skip.py

@@ -0,0 +1,57 @@
+from transformers import  PretrainedConfig
+import os
+from typing import Union, Any, Type
+
+
+
+def build_skip_config(base_config_class: type[PretrainedConfig], model_type_name: str) -> type[PretrainedConfig]:
+    class SkipConnectionConfig(base_config_class):
+        model_type: str = model_type_name
+        has_no_defaults_at_init: bool = True
+
+        def __init__(self, 
+                    sparsity: float,
+                    predictor_loss_type: str = "bce",
+                    predictor_temperature: float = 1.0,
+                    predictor_loss_alpha: float = 1.0,
+                    predictor_loss_weight: float = 0.1,
+                    use_optimized_weight_cache: bool = True,
+                    **kwargs):
+            self._sparsity = sparsity
+            self.predictor_loss_type = predictor_loss_type
+            self.predictor_temperature = predictor_temperature
+            self.predictor_loss_alpha = predictor_loss_alpha
+            self.predictor_loss_weight = predictor_loss_weight
+            self.use_optimized_weight_cache = use_optimized_weight_cache
+            super().__init__(**kwargs)
+        
+        @property
+        def sparsity(self):
+            return self._sparsity
+        
+        @sparsity.setter
+        def sparsity(self, value):
+            self._sparsity = value
+
+        @classmethod
+        def from_json_file(cls, json_file: Union[str, os.PathLike]):
+            """
+            Instantiates a [`PretrainedConfig`] from the path to a JSON file of parameters.
+
+            Args:
+                json_file (`str` or `os.PathLike`):
+                    Path to the JSON file containing the parameters.
+
+            Returns:
+                [`PretrainedConfig`]: The configuration object instantiated from that JSON file.
+
+            """
+            config_dict = cls._dict_from_json_file(json_file)
+            return cls(**config_dict)
+
+        @classmethod
+        def from_dict(cls, config_dict: dict[str, Any], **kwargs):
+            if "name_or_path" in kwargs and ("name_or_path" in config_dict or "_name_or_path" in config_dict):
+                del kwargs["name_or_path"]
+            return super().from_dict(config_dict, **kwargs)
+    return SkipConnectionConfig