Update app.py

app.py

@@ -1,11 +1,12 @@
-from flask import Flask, request, jsonify, Response, stream_with_context
+from flask import Flask, request, jsonify, Response
 import hmac, hashlib, secrets, time
 import requests
 import os
+import json
 
 app = Flask(__name__)
 
-# 🔑 Secret key for API authentication (Load from environment in production)
+# Secret key for API authentication (Load from environment in production)
 SECRET_KEY = os.getenv("SECRET_KEY")
 
 # Track API statistics
@@ -50,35 +51,37 @@ def validate_api_key(api_key):
     expected_signature = hmac.new(SECRET_KEY.encode(), random_part.encode(), hashlib.sha256).hexdigest()[:16]
     return expected_signature == received_signature
 
-# Streaming AI response
-def generate_stream_response(query: str):
+# Stream response from DeepInfra
+def stream_response(query: str):
     json_data = {
         'model': 'meta-llama/Llama-3.3-70B-Instruct-Turbo',
         'messages': [
-            {'role': 'system', 'content': 'You are TrueSyncAI, a pioneering AI startup founded by Sujal Rajpoot and Anuj Rajpoot.'},
+            {'role': 'system', 'content': 'You are TrueSyncAI, a pioneering AI startup founded by Sujal Rajpoot and Anuj Rajpoot. Your mission is to bring virtual versions of individuals into the real world, redefining human-AI interaction.'},
             {'role': 'user', 'content': query}
         ],
-        'stream': True,  # Enable streaming
-        'max_tokens': 2048,
-        'temperature': 0.2,
-        'top_p': 1
+        'stream': True, 'max_tokens': 2048, 'temperature': 0.2, 'top_p': 1
     }
-    
     try:
-        with requests.post(
+        response = requests.post(
             'https://api.deepinfra.com/v1/openai/chat/completions',
             json=json_data, stream=True, timeout=None
-        ) as response:
-            response.raise_for_status()
-            for line in response.iter_lines():
-                if line:
+        )
+        response.raise_for_status()
+        
+        def generate():
+            for value in response.iter_lines(decode_unicode=True, chunk_size=1000):
+                if value:
                     try:
-                        yield f"{line.decode('utf-8')}\n"
-                    except:continue
-    except requests.exceptions.RequestException:
-        yield "API Server is under maintenance. Please Try After Some Time. Thank You for using TrueSyncAI Chat API. Have a great day."
-
-# Streaming Chat Endpoint
+                        content = json.loads(value[5:])['choices'][0]['delta']['content']
+                        yield content + '\n'
+                    except:
+                        continue
+        
+        return Response(generate(), content_type='text/plain')
+    except:
+        return jsonify({"error": "API Server is under maintenance. Please try again later."})
+    
+# Chat Endpoint with streaming response
 @app.route("/v1/chat/completions", methods=["POST"])
 def chat():
     global request_count
@@ -93,8 +96,7 @@ def chat():
         return jsonify({"error": "Message cannot be empty"}), 400
     
     request_count += 1
-
-    return Response(stream_with_context(generate_stream_response(message)), content_type="text/plain")
+    return stream_response(message)
 
 if __name__ == "__main__":
-    app.run(host="0.0.0.0", port=7860)  # Hugging Face Spaces default port
+    app.run(host="0.0.0.0", port=7860, threaded=True)  # Enable threaded mode for better concurrency