Update app.py

app.py

@@ -166,8 +166,8 @@ def messages():
 
     try:
         json_body = request.json
-        model = json_body.get('model', 'claude-3-5-sonnet-20240620')
-        stream = json_body.get('stream', False)
+        model = json_body.get('model', 'claude-3-opus-20240229')
+        stream = json_body.get('stream', True)
         messages = json_body.get('messages', [])
 
         # 分块处理
@@ -183,10 +183,18 @@ def messages():
         total_output_tokens = 0
         full_response = []
 
+        def send_event(event_type, data):
+            event = create_event(event_type, data)
+            logger.info(f"Sending {event_type} event", extra={
+                'event_type': event_type,
+                'data': {'content': event}
+            })
+            return event
+
         def generate():
             nonlocal total_output_tokens
 
-            yield from send_event("message_start", {
+            yield send_event("message_start", {
                 "type": "message_start",
                 "message": {
                     "id": msg_id,
@@ -200,6 +208,8 @@ def messages():
                 },
             })
 
+            yield send_event("content_block_start", {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}})
+
             for chunk_index, chunk in enumerate(chunks):
                 chunk_input = "\n\n".join([normalize_content(msg['content']) for msg in chunk])
                 chunk_input_tokens = calculate_tokens(chunk_input)
@@ -211,7 +221,7 @@ def messages():
                 # 动态调整超时时间
                 timeout_seconds = max(30, min(300, chunk_input_tokens // 1000 * 30))
 
-                yield from send_event("chunk_start", {
+                yield send_event("chunk_start", {
                     "type": "chunk_start",
                     "chunk_index": chunk_index,
                     "total_chunks": total_chunks,
@@ -228,6 +238,14 @@ def messages():
                     if data.get('final', False):
                         response_event.set()
 
+                def on_connect():
+                    logger.info(f"Connected to Perplexity AI for chunk {chunk_index + 1}/{total_chunks}", extra={
+                        'event_type': 'chunk_connection_established',
+                        'data': {'chunk_index': chunk_index, 'total_chunks': total_chunks}
+                    })
+                    sio.emit('perplexity_ask', (chunk_input, get_emit_data()))
+
+                sio.on('connect', on_connect)
                 sio.on('query_progress', on_query_progress)
 
                 def timeout_handler():
@@ -243,7 +261,6 @@ def messages():
 
                 try:
                     sio.connect('wss://www.perplexity.ai/', **connect_opts, headers=sio_opts['extraHeaders'])
-                    sio.emit('perplexity_ask', (chunk_input, get_emit_data()))
 
                     while not response_event.is_set() and not timeout_event.is_set():
                         sio.sleep(0.1)
@@ -252,7 +269,7 @@ def messages():
                             full_response.append(new_chunk)
                             chunk_tokens = calculate_tokens(new_chunk)
                             total_output_tokens += chunk_tokens
-                            yield from send_event("content_block_delta", {
+                            yield send_event("content_block_delta", {
                                 "type": "content_block_delta",
                                 "index": 0,
                                 "delta": {"type": "text_delta", "text": new_chunk},
@@ -260,7 +277,7 @@ def messages():
 
                 except Exception as e:
                     logger.error(f"Error during chunk {chunk_index + 1}/{total_chunks} processing: {str(e)}", exc_info=True)
-                    yield from send_event("content_block_delta", {
+                    yield send_event("content_block_delta", {
                         "type": "content_block_delta",
                         "index": 0,
                         "delta": {"type": "text_delta", "text": f"Error processing chunk {chunk_index + 1}/{total_chunks}: {str(e)}"},
@@ -270,26 +287,30 @@ def messages():
                     if sio.connected:
                         sio.disconnect()
 
-                yield from send_event("chunk_end", {
+                yield send_event("chunk_end", {
                     "type": "chunk_end",
                     "chunk_index": chunk_index,
                     "total_chunks": total_chunks,
                 })
 
-            yield from send_event("content_block_stop", {"type": "content_block_stop", "index": 0})
-            yield from send_event("message_delta", {
+            yield send_event("content_block_stop", {"type": "content_block_stop", "index": 0})
+            yield send_event("message_delta", {
                 "type": "message_delta",
                 "delta": {"stop_reason": "end_turn", "stop_sequence": None},
                 "usage": {"input_tokens": total_input_tokens, "output_tokens": total_output_tokens},
             })
-            yield from send_event("message_stop", {"type": "message_stop"})
+            yield send_event("message_stop", {"type": "message_stop"})
 
         if stream:
             return Response(generate(), content_type='text/event-stream')
         else:
             # 非流式处理
-            for _ in generate():
-                pass  # 处理所有生成的事件，但不发送
+            full_response = []
+            for event in generate():
+                if 'content_block_delta' in event:
+                    data = json.loads(event.split('data: ')[1])
+                    full_response.append(data['delta']['text'])
+            
             return jsonify({
                 "content": [{"text": ''.join(full_response), "type": "text"}],
                 "id": msg_id,
@@ -306,8 +327,8 @@ def messages():
 
     except Exception as e:
         logger.error(f"Request error: {str(e)}", exc_info=True)
-        log_request(request.remote_addr, request.path, 400)
-        return jsonify({"error": str(e)}), 400
+        log_request(request.remote_addr, request.path, 500)
+        return jsonify({"error": str(e)}), 500
 
 def get_emit_data():
     return {
@@ -325,6 +346,7 @@ def get_emit_data():
         "query_source": "home"
     }
 
+
 def handle_non_stream(previous_messages, msg_id, model, input_tokens):
     try:
         response_event = Event()