app

.ipynb_checkpoints/app-checkpoint.py

@@ -57,7 +57,7 @@ max_length = max_seq_length
 num_workers = 2
 
 # I just reused our previous research code with some modifications.
-dir_main = "./"
+dir_main = "."
 name_model_mtr = "ChemLlama_Medium_30m_vloss_val_loss=0.029_ep_epoch=04.ckpt" 
 
 dir_model_mtr = f"{dir_main}/SolLlama-mtr/{name_model_mtr}"
@@ -73,7 +73,8 @@ num_workers = 2
 ## FT
 
 ver_ft = 0
-dir_model_ft_to_save = f"{dir_main}/SolLlama-mtr"
+# dir_model_ft_to_save = f"{dir_main}/SolLlama-mtr"
+ir_model_ft_to_save = f"{dir_main}SolLlama-mtr"
 # name_model_ft = 'Solvent.pt'
 name_model_ft = f"{solute_or_solvent}.pt"
 
@@ -130,7 +131,7 @@ trainer = L.Trainer(
 
 
 # Predict
-local_model_ft = utils_sl.load_model_ft_with_epoch(
+local_model_ft = utils_sl.load_model_ft_with(
     class_model_ft=model_ft, 
     dir_model_ft=dir_model_ft_to_save,
     name_model_ft=name_model_ft

.ipynb_checkpoints/utils_sl-checkpoint.py

@@ -78,20 +78,16 @@ def model_evalulator_sol(
 # from .model_finetune import CustomFinetuneModel
 # import model_finetune_sol
 import torch
-def load_model_ft_with_epoch(class_model_ft,
+def load_model_ft_with(class_model_ft,
                              dir_model_ft:str,
                              name_model_ft:str):
     # dir_model_ft level 1
     # ex /main/model_mtr/model_mtr_ep/dataset
-    
-
-    # extension = '.ckpt'
-    extension = '.pt'
 
     dir_target_model_ft = f"{dir_model_ft}/{name_model_ft}"
 
     
-    loaded_state_dict = torch.load(dir_target_model_ft)
+    loaded_state_dict = torch.load(dir_target_model_ft, map_location=torch.device('cpu'))
     class_model_ft.load_state_dict(loaded_state_dict['state_dict'])
     
     return class_model_ft # now is model_ft

app.py

@@ -57,7 +57,7 @@ max_length = max_seq_length
 num_workers = 2
 
 # I just reused our previous research code with some modifications.
-dir_main = "./"
+dir_main = "."
 name_model_mtr = "ChemLlama_Medium_30m_vloss_val_loss=0.029_ep_epoch=04.ckpt" 
 
 dir_model_mtr = f"{dir_main}/SolLlama-mtr/{name_model_mtr}"
@@ -73,7 +73,8 @@ num_workers = 2
 ## FT
 
 ver_ft = 0
-dir_model_ft_to_save = f"{dir_main}/SolLlama-mtr"
+# dir_model_ft_to_save = f"{dir_main}/SolLlama-mtr"
+ir_model_ft_to_save = f"{dir_main}SolLlama-mtr"
 # name_model_ft = 'Solvent.pt'
 name_model_ft = f"{solute_or_solvent}.pt"
 
@@ -130,7 +131,7 @@ trainer = L.Trainer(
 
 
 # Predict
-local_model_ft = utils_sl.load_model_ft_with_epoch(
+local_model_ft = utils_sl.load_model_ft_with(
     class_model_ft=model_ft, 
     dir_model_ft=dir_model_ft_to_save,
     name_model_ft=name_model_ft

utils_sl.py

@@ -78,20 +78,16 @@ def model_evalulator_sol(
 # from .model_finetune import CustomFinetuneModel
 # import model_finetune_sol
 import torch
-def load_model_ft_with_epoch(class_model_ft,
+def load_model_ft_with(class_model_ft,
                              dir_model_ft:str,
                              name_model_ft:str):
     # dir_model_ft level 1
     # ex /main/model_mtr/model_mtr_ep/dataset
-    
-
-    # extension = '.ckpt'
-    extension = '.pt'
 
     dir_target_model_ft = f"{dir_model_ft}/{name_model_ft}"
 
     
-    loaded_state_dict = torch.load(dir_target_model_ft)
+    loaded_state_dict = torch.load(dir_target_model_ft, map_location=torch.device('cpu'))
     class_model_ft.load_state_dict(loaded_state_dict['state_dict'])
     
     return class_model_ft # now is model_ft