优化日志输出

app.py

@@ -32,6 +32,7 @@ initQ = False
 
 def run_sakurallm():
     import subprocess
+    import threading
 
     def run_command(command):
         process = subprocess.Popen(
@@ -43,17 +44,22 @@ def run_sakurallm():
             universal_newlines=True  # 兼容不同平台上的换行符
         )
 
-        # 实时打印 stdout 的输出
-        # for stdout_line in iter(process.stdout.readline, ""):
-        #     print(stdout_line, end="")  # 避免多打印一个换行符
+        def log_stream(stream, prefix):
+            for line in iter(stream.readline, ""):
+                print(f"{prefix}{line}", end="")
+            stream.close()
 
-        # process.stdout.close()
-        # process.wait()
-
-        # stderr = process.stderr.read()
-        # process.stderr.close()
+        # 使用线程实时打印 stdout 和 stderr，避免阻塞主线程且防止缓冲区填满
+        t_stdout = threading.Thread(target=log_stream, args=(process.stdout, "[LLAMA] "))
+        t_stderr = threading.Thread(target=log_stream, args=(process.stderr, "[LLAMA_ERR] "))
+        
+        t_stdout.daemon = True
+        t_stderr.daemon = True
+        
+        t_stdout.start()
+        t_stderr.start()
 
-        # return process.returncode, stderr
+        return process
 
     # command = ["ping", "-c", "4", "qq.com"]
     command = [
@@ -78,8 +84,8 @@ def run_sakurallm():
         '--port',
         '8080'
     ]
-        # 不知道为什么 ping 可以打印流式输出, llamacpp server 不能行
-    # returncode, stderr = run_command(command)
+    
+    print("Starting SakuraLLM server...")
     run_command(command)
 
     # if returncode != 0: