Update app.py

app.py

@@ -7,7 +7,7 @@ import os
 TITLE = "✍️ AI Story Outliner"
 DESCRIPTION = """
 Enter a prompt and get 10 unique story outlines from a CPU-friendly AI model.
-The app uses **Tencent's Hunyuan-1.8B** to generate creative outlines formatted in Markdown.
+The app uses **GPT-2**, a reliable and well-established model, to generate creative outlines.
 
 **How it works:**
 1.  Enter your story idea.
@@ -25,36 +25,21 @@ examples = [
 ]
 
 # --- Model Initialization ---
-# This section loads a smaller, CPU-friendly model.
-# It will automatically use the HF_TOKEN secret when deployed on Hugging Face Spaces.
+# This section loads a smaller, stable, and CPU-friendly model that requires no authentication.
 generator = None
 model_error = None
 
 try:
     print("Initializing model... This may take a moment.")
     
-    # Explicitly load the token from environment variables (for HF Spaces secrets).
-    # This makes the authentication more robust, overriding any bad default credentials.
-    hf_token = os.environ.get("HF_TOKEN")
-
-    # Add a check to see if the token was loaded correctly.
-    if hf_token:
-        print("✅ HF_TOKEN secret found.")
-    else:
-        print("⚠️ HF_TOKEN secret not found. Please ensure it is set in your Hugging Face Space settings.")
-        # Raise an error to stop the app from proceeding without a token.
-        raise ValueError("Hugging Face token not found. Please set the HF_TOKEN secret.")
-
-    # Using a smaller model from the user's list.
-    # Passing the token explicitly to ensure correct authentication.
+    # Using 'gpt2', a stable and widely supported model that does not require a token.
     generator = pipeline(
         "text-generation",
-        model="tencent/Hunyuan-1.8B-Instruct",
-        torch_dtype=torch.bfloat16, # Use bfloat16 for better performance if available
-        device_map="auto", # Will use GPU if available, otherwise CPU
-        token=hf_token
+        model="gpt2",
+        torch_dtype=torch.float32, # Use float32 for wider CPU compatibility
+        device_map="auto" # Will use GPU if available, otherwise CPU
     )
-    print("✅ Tencent/Hunyuan-1.8B-Instruct model loaded successfully!")
+    print("✅ GPT-2 model loaded successfully!")
 
 except Exception as e:
     model_error = e
@@ -76,31 +61,42 @@ def generate_stories(prompt: str) -> list[str]:
         # Return a list of 10 empty strings to clear the outputs
         return [""] * 10
 
-    # This prompt format is specific to the Hunyuan model.
-    system_instruction = "You are an expert storyteller. Your task is to take a user's prompt and write a short story as a Markdown outline. The story must have a dramatic arc and be the length of a song. Use emojis to highlight the story sections."
-    
-    story_prompt = f"<|im_start|>system\n{system_instruction}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
+    # A generic story prompt that works well with models like GPT-2.
+    story_prompt = f"""
+Story Idea: "{prompt}"
+
+Create a short story outline based on this idea.
+
+### 🎬 The Hook
+A dramatic opening.
+
+### 🎼 The Ballad
+The main story, told concisely.
+
+### 🔚 The Finale
+A clear and satisfying ending.
+---
+"""
 
     # Parameters for the pipeline to generate 10 diverse results.
     params = {
         "max_new_tokens": 250,
         "num_return_sequences": 10,
         "do_sample": True,
-        "temperature": 0.8,
+        "temperature": 0.9,
+        "top_k": 50,
         "top_p": 0.95,
-        "pad_token_id": generator.tokenizer.eos_token_id # Suppress warning
+        "pad_token_id": generator.tokenizer.eos_token_id
     }
     
     # Generate 10 different story variations
     outputs = generator(story_prompt, **params)
 
-    # Extract the generated text and clean it up.
+    # Extract the generated text. GPT-2 will continue from the prompt.
     stories = []
     for out in outputs:
-        # Remove the prompt part from the full generated text
         full_text = out['generated_text']
-        assistant_response = full_text.split("<|im_start|>assistant\n")[-1]
-        stories.append(assistant_response)
+        stories.append(full_text)
 
     # Ensure we return exactly 10 stories, padding if necessary.
     while len(stories) < 10: