Delete convert_weights.py

convert_weights.py

@@ -1,81 +0,0 @@
-import torch
-from config import BBBConfig
-from modeling import BBBModelForSequenceClassification
-import os
-
-BASE_ARCH_PARAMS = {
-  "gnn_layers": 4,
-  "gnn_hidden": 512,
-  "dropout": 0.1,
-  "num_heads": 8,
-  "input_dim": 768,
-  "num_features": 2265,
-  "proj_dim": 1024,
-  "neurons_fc": [256],
-  "num_labels": 1
-}
-
-def convert_model(checkpoint_path: str, task_name: str, problem_type: str, save_directory: str):  
-  config_params = BASE_ARCH_PARAMS.copy()
-  config_params["task"] = task_name
-  config_params["problem_type"] = problem_type
-  
-  config = BBBConfig(**config_params)
-  
-  print(f"Instantiating HF model")
-  hf_model = BBBModelForSequenceClassification(config)
-  hf_model.eval()
-
-  if not os.path.exists(checkpoint_path):
-    return
-    
-  print(f"Loading original checkpoint from {checkpoint_path}")
-  old_state_dict = torch.load(checkpoint_path, map_location="cpu")
-
-  new_state_dict = {}
-  for key, value in old_state_dict.items():
-    if key.startswith("gats.") or \
-       key.startswith("bns.") or \
-       key.startswith("proj_gnn.") or \
-       key.startswith("proj_feat."):
-      
-      # Add the "base_model." prefix
-      new_key = "base_model." + key
-      new_state_dict[new_key] = value
-      
-    elif key.startswith("fc."):
-      # The 'fc' layer is already in the correct place
-      new_state_dict[key] = value
-      
-    else:
-      print(f"[Warning] Unmapped key found: {key}")
-      new_state_dict[key] = value
-
-  print("State dict key names adjusted.")
-
-  try:
-    hf_model.load_state_dict(new_state_dict, strict=True)
-    print("State dict loaded successfully into HF")
-  except RuntimeError as e:
-    print("\n--- ERROR LOADING STATE DICT ---")
-    print("Verify that the parameters in BASE_ARCH_PARAMS are correct.")
-    print(e)
-    return
-
-  print(f"Saving HF-formatted model to {save_directory}")
-  hf_model.save_pretrained(save_directory)
-
-if __name__ == "__main__":
-  convert_model(
-    checkpoint_path="./model_classification.pth",
-    task_name="classification",
-    problem_type="single_label_classification",
-    save_directory="./classification"
-  )
-  
-  convert_model(
-    checkpoint_path="./model_regression.pth",
-    task_name="regression",
-    problem_type="regression",
-    save_directory="./regression"
-  )