Update story_generator.py

story_generator.py

@@ -29,12 +29,12 @@ def generate_story(theme, reading_level, max_new_tokens=400, temperature=0.7):
     with torch.no_grad():
         output = story_model.generate(
             input_ids,
-            max_new_tokens=max_new_tokens,  # ⏩ Reduced to 250 for speed (previously 700)
+            max_new_tokens=max_new_tokens,
             temperature=temperature,
-            top_k=20,  # ⏩ Lowered from 50 → More focused, fewer unnecessary words
-            top_p=0.7,  # ⏩ Reduced randomness (previously 0.95)
+            top_k=20,
+            top_p=0.7,
             do_sample=True,
-            early_stopping=True,  # ⏩ Stops at logical sentence breaks
+            early_stopping=True,
             pad_token_id=story_tokenizer.pad_token_id,
             eos_token_id=story_tokenizer.eos_token_id,
             attention_mask=input_ids.ne(story_tokenizer.pad_token_id)
@@ -50,12 +50,12 @@ def generate_questions(story, max_new_tokens=150, temperature=0.7):
     with torch.no_grad():
         output = question_model.generate(
             input_ids,
-            max_new_tokens=max_new_tokens,  # ⏩ Reduced to 100 for speed (previously 300)
+            max_new_tokens=max_new_tokens,
             temperature=temperature,
-            top_k=20,  # ⏩ More focused question generation
-            top_p=0.7,  # ⏩ Less randomness
+            top_k=20,
+            top_p=0.7,
             do_sample=True,
-            early_stopping=True,  # ⏩ Stops at logical breakpoints
+            early_stopping=True,
             pad_token_id=question_tokenizer.pad_token_id,
             eos_token_id=question_tokenizer.eos_token_id,
             attention_mask=input_ids.ne(question_tokenizer.pad_token_id)