Merge remote-tracking branch 'origin/test_branch' into test_branch

ConvLab-3

@@ -0,0 +1 @@
+Subproject commit 60f4e5641f93e99b8d61b49cf5fd6dc818a83c4c

MachineLearningNLG.py

@@ -1,15 +1,21 @@
 from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, pipeline
 import requests
+import os
+import time
+
 
 class MachineLearningNLG:
     def __init__(self):
         self.model_name = "./nlg_model"  # Ścieżka do wytrenowanego modelu
+        if not os.path.exists(self.model_name):
+            raise ValueError(
+                f"Ścieżka {self.model_name} nie istnieje. Upewnij się, że model został poprawnie zapisany.")
         self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
         self.model = AutoModelForSeq2SeqLM.from_pretrained(self.model_name)
         self.generator = pipeline('text2text-generation', model=self.model, tokenizer=self.tokenizer)
 
     def translate_text(self, text, target_language='pl'):
-        url = 'https://translate.googleapis.com/translate_a/single?client=gtx&sl=auto&tl={}&dt=t&q={}'.format(target_language, text)
+        url = f'https://translate.googleapis.com/translate_a/single?client=gtx&sl=auto&tl={target_language}&dt=t&q={text}'
         response = requests.get(url)
         if response.status_code == 200:
             translated_text = response.json()[0][0][0]
@@ -19,13 +25,22 @@ class MachineLearningNLG:
 
     def nlg(self, system_act):
         input_text = f"generate text: {system_act}"
+        start_time = time.time()
         result = self.generator(input_text)
+        response_time = time.time() - start_time
         response = result[0]['generated_text']
         translated_response = self.translate_text(response, target_language='pl')
         return translated_response
 
+    def generate(self, action):
+        return self.nlg(action)
 
+    def init_session(self):
+        pass
+
+
+# Przykład użycia
 if __name__ == "__main__":
     nlg = MachineLearningNLG()
-    system_act = "inform(date.from=15.07, date.to=22.07)"
+    system_act = "inform(people.kids.ages=[4,9])"
     print(nlg.nlg(system_act))

Main.py

@@ -9,7 +9,7 @@ if __name__ == "__main__":
     nlu = NaturalLanguageAnalyzer()
     dst = DialogueStateTracker()
     policy = DialoguePolicy()
-    nlg = MachineLearningNLG()  # Używamy nowego komponentu NLG
+    nlg = MachineLearningNLG()
 
     agent = PipelineAgent(nlu=nlu, dst=dst, policy=policy, nlg=nlg, name='sys')
     response = agent.response(text)

train_nlg.py

@@ -50,10 +50,10 @@ training_args = Seq2SeqTrainingArguments(
     per_device_eval_batch_size=16,
     predict_with_generate=True,
     learning_rate=5e-5,
-    num_train_epochs=3,
+    num_train_epochs=10,
     evaluation_strategy="epoch",
     save_strategy="epoch",
-    save_total_limit=1,
+    save_total_limit=None,  # Wyłącz rotację punktów kontrolnych
     load_best_model_at_end=True,
 )
 
@@ -68,3 +68,7 @@ trainer = Seq2SeqTrainer(
 
 # Trening modelu
 trainer.train()
+
+# Zapisanie wytrenowanego modelu
+trainer.save_model("./nlg_model")
+tokenizer.save_pretrained("./nlg_model")