Update app.py

app.py

@@ -83,12 +83,15 @@ def generate_response(prompt, model, tokenizer):
 
 #Funktion, die der trainer braucht, um das Training zu evaluieren - mit einer Metrik
 def compute_metrics(eval_pred):
+    metric = evaluate.load("glue", "mrpc")
     logits, labels = eval_pred
     predictions = np.argmax(logits, axis=-1)
     #Call compute on metric to calculate the accuracy of your predictions.
     #Before passing your predictions to compute, you need to convert the predictions to logits (remember all Transformers models return logits):
     return metric.compute(predictions=predictions, references=labels)
 
+    
+
 #oder mit allen Metriken
 def compute_metrics_alle(eval_pred):
     metrics = ["accuracy", "recall", "precision", "f1"] #List of metrics to return
@@ -184,7 +187,12 @@ lm_datasets = tokenized_datasets.map(
     batch_size=1000,
     num_proc=4,
 )
-print ("lm datasets")
+
+data_collator = DataCollatorWithPadding(tokenizer=tokenizer)
+
+
+print ("###############lm datasets####################")
+print (tokenizer.decode(lm_datasets["train"][1]["input_ids"])
 #die Daten wurden nun "gereinigt" und für das Model vorbereitet.
 #z.B. anschauen mit: tokenizer.decode(lm_datasets["train"][1]["input_ids"])
 
@@ -210,6 +218,8 @@ training_args = TrainingArguments(
     overwrite_output_dir = 'True',
     per_device_train_batch_size=batch_size,  #batch_size = 2 for full training
     per_device_eval_batch_size=batch_size,
+    num_train_epochs=5, 
+    logging_steps=5000,
     evaluation_strategy = "epoch",  #oder  steps
     logging_strategy="epoch",   #oder steps
     #logging_steps=10,
@@ -221,18 +231,20 @@ training_args = TrainingArguments(
     #logging_steps=2,  # set to 1000 for full training
     #save_steps=16,    # set to 500 for full training
     #eval_steps=4,     # set to 8000 for full training
-    #warmup_steps=1,   # set to 2000 for full training
+    warmup_steps=100,   # set to 2000 for full training
     #max_steps=16,     # delete for full training
     # overwrite_output_dir=True,
     #save_total_limit=1,
     #fp16=True,
-    #save_strategy = "no",
+    save_strategy = "no",
     optim="adamw_torch",
     #load_best_model_at_end=False,
     #load_best_model_at_end=True
     #push_to_hub=True,
 )
 
+
+
 #Trainer zusammenstellen
 print ("################################")
 print ("trainer")
@@ -242,6 +254,7 @@ trainer = Trainer(
         args=training_args,
         train_dataset=lm_datasets["train"],
         eval_dataset=lm_datasets["test"],
+        data_collator=data_collator,
         #tokenizer=tokenizer,
         compute_metrics=compute_metrics,
 )
@@ -324,6 +337,12 @@ print("Evaluate:")
 trainer.evaluate(eval_dataset=lm_datasets["test"])
 print("Done Eval")
 
+print('################ Test Trained Model ###################')
+#predictions = trainer.predict(lm_datasets["test"])
+#preds = np.argmax(predictions.predictions, axis=-1)
+
+
+
 ###################################################
 #Save to a place -????? Where????
 #print("Save to ???")
@@ -334,7 +353,7 @@ print("Done Eval")
 #####################################
 #Push to Hub
 print ("################################")
-print("push to hub")
+print("###################push to hub###################")
 print("push to hub - Model")
 login(token=os.environ["HF_WRITE_TOKEN"]) 
 trainer.push_to_hub("alexkueck/li-tis-tuned-2")