Updated

.ipynb_checkpoints/model_loader-checkpoint.py

@@ -279,27 +279,27 @@ def load_T5_model_classification(checkpoint, num_labels, half_precision, full =
     # Load model and tokenizer
 
     if "ankh" in checkpoint :
-        model = T5EncoderModel.from_pretrained(checkpoint)
-        tokenizer = AutoTokenizer.from_pretrained(checkpoint)
+        model = T5EncoderModel.from_pretrained(checkpoint,resume_download=True)
+        tokenizer = AutoTokenizer.from_pretrained(checkpoint,resume_download=True)
 
     elif "prot_t5" in checkpoint:
         # possible to load the half precision model (thanks to @pawel-rezo for pointing that out)
         if half_precision and deepspeed:
             #tokenizer = T5Tokenizer.from_pretrained('Rostlab/prot_t5_xl_half_uniref50-enc', do_lower_case=False)
             #model = T5EncoderModel.from_pretrained("Rostlab/prot_t5_xl_half_uniref50-enc", torch_dtype=torch.float16)#.to(torch.device('cuda')
-            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False)
-            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'))
+            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False,resume_download=True)
+            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'),resume_download=True)
         else:
             model = T5EncoderModel.from_pretrained(checkpoint)
             tokenizer = T5Tokenizer.from_pretrained(checkpoint)
                 
     elif "ProstT5" in checkpoint:
         if half_precision and deepspeed: 
-            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False)
-            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'))
+            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False,resume_download=True)
+            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'),resume_download=True)
         else:
-            model = T5EncoderModel.from_pretrained(checkpoint)
-            tokenizer = T5Tokenizer.from_pretrained(checkpoint) 
+            model = T5EncoderModel.from_pretrained(checkpoint,resume_download=True)
+            tokenizer = T5Tokenizer.from_pretrained(checkpoint,resume_download=True) 
     
     # Create new Classifier model with PT5 dimensions
     class_config=ClassConfig(num_labels=num_labels)

model_loader.py

@@ -279,27 +279,27 @@ def load_T5_model_classification(checkpoint, num_labels, half_precision, full =
     # Load model and tokenizer
 
     if "ankh" in checkpoint :
-        model = T5EncoderModel.from_pretrained(checkpoint)
-        tokenizer = AutoTokenizer.from_pretrained(checkpoint)
+        model = T5EncoderModel.from_pretrained(checkpoint,resume_download=True)
+        tokenizer = AutoTokenizer.from_pretrained(checkpoint,resume_download=True)
 
     elif "prot_t5" in checkpoint:
         # possible to load the half precision model (thanks to @pawel-rezo for pointing that out)
         if half_precision and deepspeed:
             #tokenizer = T5Tokenizer.from_pretrained('Rostlab/prot_t5_xl_half_uniref50-enc', do_lower_case=False)
             #model = T5EncoderModel.from_pretrained("Rostlab/prot_t5_xl_half_uniref50-enc", torch_dtype=torch.float16)#.to(torch.device('cuda')
-            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False)
-            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'))
+            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False,resume_download=True)
+            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'),resume_download=True)
         else:
             model = T5EncoderModel.from_pretrained(checkpoint)
             tokenizer = T5Tokenizer.from_pretrained(checkpoint)
                 
     elif "ProstT5" in checkpoint:
         if half_precision and deepspeed: 
-            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False)
-            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'))
+            tokenizer = T5Tokenizer.from_pretrained(checkpoint, do_lower_case=False,resume_download=True)
+            model = T5EncoderModel.from_pretrained(checkpoint, torch_dtype=torch.float16).to(torch.device('cuda'),resume_download=True)
         else:
-            model = T5EncoderModel.from_pretrained(checkpoint)
-            tokenizer = T5Tokenizer.from_pretrained(checkpoint) 
+            model = T5EncoderModel.from_pretrained(checkpoint,resume_download=True)
+            tokenizer = T5Tokenizer.from_pretrained(checkpoint,resume_download=True) 
     
     # Create new Classifier model with PT5 dimensions
     class_config=ClassConfig(num_labels=num_labels)