Update app.py

app.py

@@ -1,6 +1,6 @@
 #Using codes from killerz3/PodGen & eswardivi/Podcastify
+#For ZeroGPU limit, I roll back to inference API. You can use local or HF model also, remove the relative comment sign, it works;
 import json
-import spaces
 import httpx
 import os
 import re
@@ -9,9 +9,9 @@ import edge_tts
 import torch
 import tempfile
 import gradio as gr
-import gradio_client
+from huggingface_hub import AsyncInferenceClient
 from pydub import AudioSegment
-from transformers import AutoModelForCausalLM, AutoTokenizer
+#from transformers import AutoModelForCausalLM, AutoTokenizer
 
 from moviepy.editor import AudioFileClip, concatenate_audioclips
 
@@ -52,16 +52,22 @@ footer {
 }
 """
 
-MODEL_ID = "01-ai/Yi-1.5-6B-Chat"
+MODEL_ID = "01-ai/Yi-1.5-34B-Chat"
+Client = AsyncInferenceClient(MODEL_ID)
 
+
+"""
+# USING LOCAL MODEL
 model = AutoModelForCausalLM.from_pretrained(
-    MODEL_ID,
-    torch_dtype=torch.float16,
-    device_map="auto"
+     MODEL_ID,
+     torch_dtype=torch.float16,
+     device_map="auto"
 ).eval()
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 
+"""
+
 
 def validate_url(url):
     try:
@@ -127,8 +133,9 @@ async def gen_show(script):
         print(f"Deleted temporary file: {temp_file}")
 
     return output_filename
-
-@spaces.GPU
+    
+"""
+# USING LOCAL MODEL
 def generator(messages):
     input_ids = tokenizer.apply_chat_template(
         conversation=messages,
@@ -148,6 +155,8 @@ def generator(messages):
     results = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
     print(results)
     return results
+"""
+
 
 def extract_content(text):
     """Extracts the JSON content from the given text."""
@@ -166,15 +175,22 @@ async def main(link):
     if "Error" in text:
         return text, None
 
-    prompt = f"News: {text}"
-
-    messages = [
-        {"role": "system", "content": system_prompt},
-        {"role": "user", "content": prompt},
-    ]
-    
-    generated_script = extract_content(generator(messages))
+    prompt = f"News: {text} json:"
+    formatted_prompt = system_prompt + "\n\n\n" + text
+    # messages = [
+    #     {"role": "system", "content": system_prompt},
+    #     {"role": "user", "content": prompt},
+    # ]
 
+    answer = Client.text_generation(
+        prompt=formatted_prompt, 
+        max_new_tokens=4096,
+        details=True,
+        temperature=0.7,
+        return_full_text=False)
+    print(answer)
+    #generated_script = extract_content(generator(messages))
+    generated_script = extract_content(answer)
     print("Generated Script:"+generated_script)
 
     # Check if the generated_script is empty or not valid JSON