Create app.py

app.py

@@ -0,0 +1,389 @@
+import gradio as gr
+import os
+import requests
+import json
+import time
+from dotenv import load_dotenv
+
+# Load .env file (if it exists)
+load_dotenv()
+
+def create_deepseek_interface():
+    # Get API keys from environment variables
+    api_key = os.getenv("FW_API_KEY")
+    serphouse_api_key = os.getenv("SERPHOUSE_API_KEY")
+    
+    if not api_key:
+        print("Warning: FW_API_KEY environment variable is not set.")
+    if not serphouse_api_key:
+        print("Warning: SERPHOUSE_API_KEY environment variable is not set.")
+    
+    # Keyword extraction function (LLM-based)
+    def extract_keywords_with_llm(query):
+        if not api_key:
+            return "FW_API_KEY not set for LLM keyword extraction.", query
+        
+        # Extract keywords using LLM (DeepSeek model)
+        url = "https://api.fireworks.ai/inference/v1/chat/completions"
+        payload = {
+            "model": "accounts/fireworks/models/llama4-maverick-instruct-basic",
+            "max_tokens": 200,
+            "temperature": 0.1,  # Low temperature for consistent results
+            "messages": [
+                {
+                    "role": "system",
+                    "content": "You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside tags, and then provide your solution or response to the problem. Extract key search terms from the user's question that would be effective for web searches. Provide these as a search query with words separated by spaces only, without commas. For example: 'Prime Minister Han Duck-soo impeachment results'"
+                },
+                {
+                    "role": "user",
+                    "content": query
+                }
+            ]
+        }
+        headers = {
+            "Accept": "application/json",
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {api_key}"
+        }
+        
+        try:
+            response = requests.post(url, headers=headers, json=payload)
+            response.raise_for_status()
+            result = response.json()
+            
+            # Extract keywords from response
+            keywords = result["choices"][0]["message"]["content"].strip()
+            
+            # Use original query if keywords are too long or improperly formatted
+            if len(keywords) > 100:
+                return f"Extracted keywords: {keywords}", query
+                
+            return f"Extracted keywords: {keywords}", keywords
+            
+        except Exception as e:
+            print(f"Error during keyword extraction: {str(e)}")
+            return f"Error during keyword extraction: {str(e)}", query
+    
+    # Search function using SerpHouse API
+    def search_with_serphouse(query):
+        if not serphouse_api_key:
+            return "SERPHOUSE_API_KEY is not set."
+        
+        try:
+            # Extract keywords
+            extraction_result, search_query = extract_keywords_with_llm(query)
+            print(f"Original query: {query}")
+            print(extraction_result)
+            
+            # Basic GET method seems best after analyzing documentation
+            url = "https://api.serphouse.com/serp/live"
+            
+            # Check if query is in Korean
+            is_korean = any('\uAC00' <= c <= '\uD7A3' for c in search_query)
+            
+            # Simplified parameters
+            params = {
+                "q": search_query,
+                "domain": "google.com",
+                "serp_type": "web",  # Changed to basic web search
+                "device": "desktop",
+                "lang": "ko" if is_korean else "en"
+            }
+            
+            headers = {
+                "Authorization": f"Bearer {serphouse_api_key}"
+            }
+            
+            print(f"Calling SerpHouse API with basic GET method...")
+            print(f"Search term: {search_query}")
+            print(f"Request URL: {url} - Parameters: {params}")
+            
+            # Execute GET request
+            response = requests.get(url, headers=headers, params=params)
+            response.raise_for_status()
+            
+            print(f"SerpHouse API response status code: {response.status_code}")
+            search_results = response.json()
+            
+            # Check response structure
+            print(f"Response structure: {list(search_results.keys()) if isinstance(search_results, dict) else 'Not a dictionary'}")
+            
+            # Parse and format search results (in Markdown)
+            formatted_results = []
+            formatted_results.append(f"## Search term: {search_query}\n\n")
+            
+            # Handle various possible response structures
+            organic_results = None
+            
+            # Possible response structure 1
+            if "results" in search_results and "organic" in search_results["results"]:
+                organic_results = search_results["results"]["organic"]
+                
+            # Possible response structure 2
+            elif "organic" in search_results:
+                organic_results = search_results["organic"]
+                
+            # Possible response structure 3 (nested results)
+            elif "results" in search_results and "results" in search_results["results"]:
+                if "organic" in search_results["results"]["results"]:
+                    organic_results = search_results["results"]["results"]["organic"]
+            
+            # Process organic results if available
+            if organic_results and len(organic_results) > 0:
+                # Output response structure
+                print(f"First organic result structure: {organic_results[0].keys() if len(organic_results) > 0 else 'empty'}")
+                
+                for i, result in enumerate(organic_results[:5], 1):  # Show only top 5 results
+                    title = result.get("title", "No title")
+                    snippet = result.get("snippet", "No content")
+                    link = result.get("link", "#")
+                    displayed_link = result.get("displayed_link", link)
+                    
+                    # Format in Markdown (including number and link)
+                    formatted_results.append(
+                        f"### {i}. [{title}]({link})\n\n"
+                        f"{snippet}\n\n"
+                        f"**Source**: [{displayed_link}]({link})\n\n"
+                        f"---\n\n"
+                    )
+                
+                print(f"Found {len(organic_results)} search results")
+                return "".join(formatted_results)
+            
+            # Handle case with no results or unexpected structure
+            print("No search results or unexpected response structure")
+            print(f"Detailed response structure: {search_results.keys() if hasattr(search_results, 'keys') else 'Unclear structure'}")
+            
+            # Find error messages in response
+            error_msg = "No search results found or response format is different than expected"
+            if "error" in search_results:
+                error_msg = search_results["error"]
+            elif "message" in search_results:
+                error_msg = search_results["message"]
+                
+            return f"## Results for '{search_query}'\n\n{error_msg}"
+            
+        except Exception as e:
+            error_msg = f"Error during search: {str(e)}"
+            print(error_msg)
+            import traceback
+            print(traceback.format_exc())
+            
+            # Add API request details for debugging (in Markdown)
+            return f"## Error Occurred\n\n" + \
+                   f"An error occurred during search: **{str(e)}**\n\n" + \
+                   f"### API Request Details:\n" + \
+                   f"- **URL**: {url}\n" + \
+                   f"- **Search Term**: {search_query}\n" + \
+                   f"- **Parameters**: {params}\n"
+    
+    # Function to call DeepSeek API with streaming
+    def query_deepseek_streaming(message, history, use_deep_research):
+        if not api_key:
+            yield history, "Environment variable FW_API_KEY is not set. Please check the environment variables on the server."
+            return
+        
+        search_context = ""
+        search_info = ""
+        if use_deep_research:
+            try:
+                # Start search (first message)
+                yield history + [(message, "🔍 Extracting optimal keywords and searching the web...")], ""
+                
+                # Execute search - add logs for debugging
+                print(f"Deep Research activated: Starting search for '{message}'")
+                search_results = search_with_serphouse(message)
+                print(f"Search results received: {search_results[:100]}...")  # Output first part of results
+                
+                if not search_results.startswith("Error during search") and not search_results.startswith("SERPHOUSE_API_KEY"):
+                    search_context = f"""
+Here are recent search results related to the user's question. Use this information to provide an accurate response with the latest information:
+{search_results}
+Based on the above search results, answer the user's question. If you cannot find a clear answer in the search results, use your knowledge to provide the best answer.
+When citing search results, mention the source, and ensure your answer reflects the latest information.
+"""
+                    search_info = f"🔍 Deep Research feature activated: Generating response based on relevant web search results..."
+                else:
+                    print(f"Search failed or no results: {search_results}")
+            except Exception as e:
+                print(f"Exception occurred during Deep Research: {str(e)}")
+                search_info = f"🔍 Deep Research feature error: {str(e)}"
+        
+        # Prepare conversation history for API request
+        messages = []
+        for user, assistant in history:
+            messages.append({"role": "user", "content": user})
+            messages.append({"role": "assistant", "content": assistant})
+        
+        # Add system message with search context if available
+        if search_context:
+            # DeepSeek model supports system messages
+            messages.insert(0, {"role": "system", "content": search_context})
+        
+        # Add new user message
+        messages.append({"role": "user", "content": message})
+        
+        # Prepare API request
+        url = "https://api.fireworks.ai/inference/v1/chat/completions"
+        payload = {
+            "model": "accounts/fireworks/models/llama4-maverick-instruct-basic",
+            "max_tokens": 20480,
+            "top_p": 1,
+            "top_k": 40,
+            "presence_penalty": 0,
+            "frequency_penalty": 0,
+            "temperature": 0.6,
+            "messages": messages,
+            "stream": True  # Enable streaming
+        }
+        headers = {
+            "Accept": "application/json",
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {api_key}"
+        }
+        
+        try:
+            # Request streaming response
+            response = requests.request("POST", url, headers=headers, data=json.dumps(payload), stream=True)
+            response.raise_for_status()  # Raise exception for HTTP errors
+            
+            # Add message and start with initial response
+            new_history = history.copy()
+            
+            # Include search_info in starting message if available
+            start_msg = search_info if search_info else ""
+            new_history.append((message, start_msg))
+            
+            # Full response text
+            full_response = start_msg
+            
+            # Process streaming response
+            for line in response.iter_lines():
+                if line:
+                    line_text = line.decode('utf-8')
+                    
+                    # Remove 'data: ' prefix
+                    if line_text.startswith("data: "):
+                        line_text = line_text[6:]
+                    
+                    # Check for stream end message
+                    if line_text == "[DONE]":
+                        break
+                    
+                    try:
+                        # Parse JSON
+                        chunk = json.loads(line_text)
+                        chunk_content = chunk.get("choices", [{}])[0].get("delta", {}).get("content", "")
+                        
+                        if chunk_content:
+                            full_response += chunk_content
+                            # Update chat history
+                            new_history[-1] = (message, full_response)
+                            yield new_history, ""
+                    except json.JSONDecodeError:
+                        continue
+            
+            # Return final response
+            yield new_history, ""
+            
+        except requests.exceptions.RequestException as e:
+            error_msg = f"API error: {str(e)}"
+            if hasattr(e, 'response') and e.response and e.response.status_code == 401:
+                error_msg = "Authentication failed. Please check your FW_API_KEY environment variable."
+            yield history, error_msg
+    
+    # Create Gradio interface
+    with gr.Blocks(theme="soft", fill_height=True) as demo:
+        # Header section
+        gr.Markdown(
+            """
+            # 🤖 Llama-4-Maverick-17B + Research 
+            ### Llama-4-Maverick-17B Model + Real-time 'Deep Research' Agentic AI System @ https://discord.gg/openfreeai
+            """
+        )
+        
+        # Main layout
+        with gr.Row():
+            # Main content area
+            with gr.Column():
+                # Chat interface
+                chatbot = gr.Chatbot(
+                    height=500,
+                    show_label=False,
+                    container=True
+                )
+                
+                # Add Deep Research toggle and status display
+                with gr.Row():
+                    with gr.Column(scale=3):
+                        use_deep_research = gr.Checkbox(
+                            label="Enable Deep Research",
+                            info="Utilize optimal keyword extraction and web search for latest information",
+                            value=False
+                        )
+                    with gr.Column(scale=1):
+                        api_status = gr.Markdown("API Status: Ready")
+                
+                # Check and display API key status
+                if not serphouse_api_key:
+                    api_status.value = "⚠️ SERPHOUSE_API_KEY is not set"
+                if not api_key:
+                    api_status.value = "⚠️ FW_API_KEY is not set"
+                if api_key and serphouse_api_key:
+                    api_status.value = "✅ API keys configured"
+                
+                # Input area
+                with gr.Row():
+                    msg = gr.Textbox(
+                        label="Message",
+                        placeholder="Enter your prompt here...",
+                        show_label=False,
+                        scale=9
+                    )
+                    submit = gr.Button("Send", variant="primary", scale=1)
+                
+                # Clear conversation button
+                with gr.Row():
+                    clear = gr.ClearButton([msg, chatbot], value="🧹 Clear Conversation")
+                    
+                # Example queries
+                gr.Examples(
+                    examples=[
+                        "Explain the difference between Transformers and RNNs in deep learning.",
+                        "Write a Python function to find prime numbers within a specific range.",
+                        "Summarize the key concepts of reinforcement learning."
+                    ],
+                    inputs=msg
+                )
+                
+                # Error message display
+                error_box = gr.Markdown("")
+        
+        # Connect buttons to functions
+        submit.click(
+            query_deepseek_streaming,
+            inputs=[msg, chatbot, use_deep_research],
+            outputs=[chatbot, error_box]
+        ).then(
+            lambda: "",
+            None,
+            [msg]
+        )
+        
+        # Allow Enter key submission
+        msg.submit(
+            query_deepseek_streaming,
+            inputs=[msg, chatbot, use_deep_research],
+            outputs=[chatbot, error_box]
+        ).then(
+            lambda: "",
+            None,
+            [msg]
+        )
+        
+    return demo
+
+# Run interface
+if __name__ == "__main__":
+    demo = create_deepseek_interface()
+    demo.launch(debug=True)