Diagnostics

qarac/models/QaracEncoderModel.py

@@ -47,8 +47,12 @@ class QaracEncoderModel(transformers.PreTrainedModel):
             Vector representing the document
 
         """
+        print('Encoder',self.encoder.device)
+        print('Head',self.head.device)
         if attention_mask is None and 'attention_mask' in input_ids:
             (input_ids,attention_mask) = (input_ids['input_ids'],input_ids['attention_mask'])
+        print('input_ids',input_ids.device)
+        print('attention_mask',attention_mask.device)
         return self.head(self.encoder(input_ids,
                                       attention_mask).last_hidden_state,
                          attention_mask)

qarac/models/QaracTrainerModel.py

@@ -99,8 +99,4 @@ class QaracTrainerModel(torch.nn.Module):
         
         consistency = self.cosine(s0,s1)
         return (encode_decode,question_answering,reasoning,consistency)
-    
-    def devices(self):
-        print("Question encoder",self.question_encoder.device)
-        print('Answer Encoder',self.answer_encoder.device)
-        print('Decoer',self.decoder.device)
+    

scripts.py

@@ -125,7 +125,6 @@ def train_models(path,progress=gradio.Progress(track_tqdm=True)):
                                                                tokenizer)
     
     trainer.to(device)
-    trainer.devices()
     loss_fn = CombinedLoss()
     loss_fn.cuda()
     optimizer = torch.optim.NAdam(trainer.parameters(),lr=5.0e-5)