Update app.py

app.py

@@ -1,6 +1,5 @@
 #Using codes from killerz3/PodGen & eswardivi/Podcastify
 import json
-import spaces
 import httpx
 import os
 import re
@@ -10,24 +9,23 @@ import torch
 import tempfile
 import gradio as gr
 import gradio_client
+from openai import OpenAI
 from pydub import AudioSegment
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
 from moviepy.editor import AudioFileClip, concatenate_audioclips
 
 system_prompt = '''
-    You are an talkshow podcast generator. You have to create short conversations between Alice and Bob that gives an overview of the News given by the user.
+    You are an talk-show podcast generator. You have to create short conversations between Xiaoxiao and Yunjian that gives an overview of the News given by the user.
     Please provide the script and output strictly in the following JSON format:
     {
       "title": "[string]",
       "content": {
-        "Alice_0": "[string]",
-        "BOB_0": "[string]",
+        "Xiaoxiao: "[string]",
+        "Yunjian": "[string]",
         ...
       }
     }
-    #Please note that the [string] you generate now must be in based on the tone of people's daily life.
-    #No more than five rounds of conversation, be concise.
+    #Please note that the [string] you generate now must be in easy-and-understandable Chinese.
+    #Be concise.
 '''
 
 DESCRIPTION = '''
@@ -52,15 +50,8 @@ footer {
 }
 """
 
-MODEL_ID = "01-ai/Yi-1.5-6B-Chat"
-
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_ID,
-    torch_dtype=torch.float16,
-    device_map="auto"
-).eval()
-
-tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+apikey = os.environ.get("API_KEY")
+client = OpenAI(api_key=apikey, base_url="https://api.deepseek.com")
 
 
 def validate_url(url):
@@ -98,7 +89,7 @@ async def gen_show(script):
     for key, text in content.items():
         speaker = key.split('_')[0]  # Extract the speaker name
         index = key.split('_')[1]    # Extract the dialogue index
-        voice = "en-US-JennyNeural" if speaker == "Alice" else "en-US-GuyNeural"
+        voice = "zh-CN-XiaoxiaoNeural" if speaker == "Xiaoxiao" else "zh-CN-YunjianNeural"
 
         # Create temporary file for each speaker's dialogue
         temp_file = tempfile.NamedTemporaryFile(suffix='.mp3', delete=False)
@@ -128,26 +119,6 @@ async def gen_show(script):
 
     return output_filename
 
-@spaces.GPU
-def generator(messages):
-    input_ids = tokenizer.apply_chat_template(
-        conversation=messages,
-        add_generation_prompt=True,
-        tokenize=True,
-        return_tensors='pt'
-    )
-    
-    output_ids = model.generate(
-        input_ids.to('cuda'), 
-        eos_token_id=tokenizer.eos_token_id,
-        max_new_tokens=4096,
-        temperature=0.5,
-        repetition_penalty=1.2,
-        )
-    
-    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."""
@@ -172,8 +143,15 @@ async def main(link):
         {"role": "system", "content": system_prompt},
         {"role": "user", "content": prompt},
     ]
-    
-    generated_script = extract_content(generator(messages))
+    completion = client.chat.completions.create(
+        model="deepseek-chat",
+        messages=messages,
+        max_tokens=4096,
+        temperature=0.7,
+        stream=False
+    )
+
+    generated_script = completion.choices[0].message.content
 
     print("Generated Script:"+generated_script)