Update app.py

app.py

@@ -87,28 +87,65 @@ def scrape_articles(articles: List[Dict]) -> List[Dict]:
     
     return articles
 
+# Load prompt templates
+with open("prompts.yaml", 'r') as stream:
+    prompt_templates = yaml.safe_load(stream)
+
+# Initialize the models
+qwen_model = HfApiModel(
+    max_tokens=2096,
+    temperature=0.5,
+    model_id='Qwen/Qwen2.5-Coder-32B-Instruct',
+    custom_role_conversions=None,
+)
+
+deepseek_model = HfApiModel(
+    max_tokens=2096,
+    temperature=0.3,  # Lower temperature for more focused summaries
+    model_id='deepseek-ai/DeepSeek-R1-Distill-Qwen-32B',
+    custom_role_conversions=None,
+)
+
 @tool
 def summarize_news(articles: List[Dict]) -> str:
-    """Creates a summary of the news articles followed by a list of sources.
-    
-    Args:
-        articles: List of article dictionaries containing title, source, date, link, snippet, and full_content
-    
-    Returns:
-        A string containing a summary followed by article references
-    """
+    """Creates a summary of the news articles followed by a list of sources."""
     if not articles or not isinstance(articles, list):
         return "No articles to summarize"
     
-    # Collect all content for the overall summary
-    all_content = [article.get('full_content', article['snippet']) for article in articles]
+    # Prepare content for summarization
+    content_to_summarize = ""
+    for article in articles:
+        content = article.get('full_content', article['snippet'])
+        content_to_summarize += f"Title: {article['title']}\nSource: {article['source']}\nContent: {content}\n\n"
+    
+    # Use DeepSeek model to generate a concise summary
+    summary_prompt = f"""Please provide a concise summary of these news articles, focusing on the key points and main developments. Then list the individual articles with brief descriptions.
+
+Articles to summarize:
+{content_to_summarize}
+
+Format the output as:
+📰 Summary: [overall summary]
+
+🔍 Key Articles:
+1. [Title] - [brief description]
+[Read more link + date]
+"""
     
-    # Create a high-level summary from content
+    try:
+        summary = deepseek_model.complete(summary_prompt).strip()
+        return summary
+    except Exception as e:
+        # Fallback to original summary format if DeepSeek fails
+        return original_summary_format(articles)
+
+def original_summary_format(articles: List[Dict]) -> str:
+    # Original summary format as fallback
     summary = "📰 Summary:\n"
     summary += "Latest news covers " + ", ".join(set(article['source'] for article in articles)) + ". "
-    summary += "Key points: " + ". ".join(all_content[:2]) + "\n\n"
+    all_snippets = [article.get('full_content', article['snippet']) for article in articles]
+    summary += "Key points: " + ". ".join(all_snippets[:2]) + "\n\n"
     
-    # List individual articles
     summary += "🔍 Articles:\n"
     for idx, article in enumerate(articles, 1):
         title = article['title']
@@ -123,24 +160,12 @@ def summarize_news(articles: List[Dict]) -> str:
     
     return summary
 
-# Load prompt templates
-with open("prompts.yaml", 'r') as stream:
-    prompt_templates = yaml.safe_load(stream)
-
-# Initialize the model
-model = HfApiModel(
-    max_tokens=2096,
-    temperature=0.5,
-    model_id='Qwen/Qwen2.5-Coder-32B-Instruct',
-    custom_role_conversions=None,
-)
-
 final_answer = FinalAnswerTool()
 
 # Create the agent with all tools
 agent = CodeAgent(
-    model=model,
-    tools=[fetch_news, scrape_articles, summarize_news, final_answer],  # Added scrape_articles
+    model=qwen_model,  # Use Qwen model for main agent
+    tools=[fetch_news, scrape_articles, summarize_news, final_answer],
     max_steps=6,
     verbosity_level=1,
     grammar=None,