Upload model

huggingface.py

@@ -1,5 +1,4 @@
-from __future__ import annotations
-from pathlib import Path
+import inspect
 from torch import nn
 from transformers import AutoConfig, AutoModel, AutoTokenizer
 from transformers import PretrainedConfig, PreTrainedModel, PreTrainedTokenizer
@@ -19,23 +18,31 @@ def make_config_class(model_args: dict, model_type: str) -> PretrainedConfig:
     return Config
 
 
-def make_model_class(base_class: nn.Module, config_to_args: callable = None) -> PreTrainedModel:
-    if config_to_args is None:
-        def config_to_args(config):
-                args_dict = {}
-
-                for attr_name in dir(config):
-                    if not attr_name.startswith("_") and not callable(getattr(config, attr_name)):
-                        args_dict[attr_name] = getattr(config, attr_name)
-
-                return args_dict
-        
+def make_model_class(base_class: nn.Module, config_attributes: list[str] = None) -> PreTrainedModel:
+    base_init_signature = inspect.signature(base_class.__init__)
+    base_params = set(base_init_signature.parameters.keys()) - {"self"}
+    
     class Model(PreTrainedModel):
         config_class: PretrainedConfig
 
         def __init__(self, config, **kwargs):
             super().__init__(config, **kwargs)
-            self._model = base_class(**config_to_args(config), **kwargs)
+            
+            if config_attributes is not None:
+                model_kwargs = {a: getattr(config, a) for a in config_attributes if hasattr(config, a)}
+            else:
+                model_kwargs = {}
+
+                for param_name in base_params:
+                    if hasattr(config, param_name):
+                        model_kwargs[param_name] = getattr(config, param_name)
+            
+            filtered_kwargs = {k: v for k, v in kwargs.items() if k in base_params}
+            
+            if "config" in base_params:
+                self._model = base_class(config, **model_kwargs, **filtered_kwargs)
+            else:
+                self._model = base_class(**model_kwargs, **filtered_kwargs)
 
         def forward(self, *args, **kwargs):
             return self._model(*args, **kwargs)

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9bd62368812f60243bbcb3ce4b2999a16bd2726a093923f48c7193995ab6f84
+oid sha256:5906761113ebf5b48e15bb2a1143f7d02c8fd959c946be09b2b3bcb2ca195a6e
 size 228