sync: PR #21 - MiMo compat fix, endpoint edit dedup, anti-stall Windows compat, AGENTS.md/CLAUDE.md

translate-proxy.py

@@ -5677,23 +5677,40 @@ class Handler(http.server.BaseHTTPRequestHandler):
                 break
             self._forward_oa_compat(upstream, stream, model, chat_body, body, input_data, fwd, target, tracker)
 
+    @staticmethod
+    def _is_mimo_provider():
+        return "xiaomimimo.com" in TARGET_URL
+
     def _build_chat_body(self, model, messages, body, stream):
         chat_body = {"model": model, "messages": messages}
+        is_mimo = self._is_mimo_provider()
         for k in ("temperature", "top_p"):
             if k in body:
                 chat_body[k] = body[k]
-        chat_body["max_tokens"] = max(body.get("max_output_tokens", 0), 64000)
+        max_tok = max(body.get("max_output_tokens", 0), 64000)
+        if is_mimo:
+            chat_body["max_completion_tokens"] = max_tok
+        else:
+            chat_body["max_tokens"] = max_tok
         tools = oa_convert_tools(body.get("tools"))
         if tools:
             chat_body["tools"] = tools
         if body.get("tool_choice"):
             chat_body["tool_choice"] = body["tool_choice"]
         chat_body["stream"] = stream
-        if not REASONING_ENABLED or REASONING_EFFORT == "none":
-            chat_body["enable_thinking"] = False
-            chat_body["reasoning_effort"] = "none"
+        if is_mimo:
+            if not REASONING_ENABLED or REASONING_EFFORT == "none":
+                chat_body["thinking"] = {"type": "disabled"}
+            else:
+                mimo_effort = {"minimal": "low", "max": "high"}.get(REASONING_EFFORT, REASONING_EFFORT)
+                chat_body["thinking"] = {"type": "enabled"}
+                chat_body["reasoning_effort"] = mimo_effort
         else:
-            chat_body["reasoning_effort"] = REASONING_EFFORT
+            if not REASONING_ENABLED or REASONING_EFFORT == "none":
+                chat_body["enable_thinking"] = False
+                chat_body["reasoning_effort"] = "none"
+            else:
+                chat_body["reasoning_effort"] = REASONING_EFFORT
         return chat_body
 
     def _handle_antigravity_v2(self, body, model, stream, tracker=None):
@@ -8572,42 +8589,67 @@ def _handle_shutdown_signal(sig, frame):
          SERVER.shutdown()
  
 def _anti_stall_cleanup():
+    import subprocess as _sp
     my_pid = os.getpid()
-    my_ppid = os.getppid()
-    my_pgid = os.getpgid(0)
     killed = []
     try:
-        import subprocess as _sp
-        out = _sp.run(["pgrep", "-f", "translate-proxy"], capture_output=True, text=True, timeout=5).stdout.strip()
-        for pid_str in out.splitlines():
-            pid_str = pid_str.strip()
-            if not pid_str or not pid_str.isdigit():
-                continue
-            pid = int(pid_str)
-            if pid == my_pid or pid == my_ppid:
-                continue
-            try:
-                pgid = os.getpgid(pid)
-                if pgid == my_pgid:
+        if sys.platform == "win32":
+            out = _sp.run(
+                ["tasklist", "/FI", "IMAGENAME eq python.exe", "/FO", "CSV", "/NH"],
+                capture_output=True, text=True, timeout=5,
+            ).stdout.strip()
+            for line in out.splitlines():
+                parts = line.split(",")
+                if len(parts) >= 2:
+                    pid_str = parts[1].strip('"')
+                    if not pid_str.isdigit():
+                        continue
+                    pid = int(pid_str)
+                    if pid == my_pid:
+                        continue
+                    cmd = _sp.run(
+                        ["wmic", "process", "where", f"ProcessId={pid}", "/FORMAT:CommandLine", "/VALUE"],
+                        capture_output=True, text=True, timeout=5,
+                    ).stdout.strip()
+                    if "translate-proxy" not in cmd:
+                        continue
+                    try:
+                        _sp.run(["taskkill", "/PID", str(pid), "/F"], capture_output=True, timeout=5)
+                        killed.append(pid)
+                    except Exception:
+                        pass
+        else:
+            my_ppid = os.getppid()
+            my_pgid = os.getpgid(0)
+            out = _sp.run(["pgrep", "-f", "translate-proxy"], capture_output=True, text=True, timeout=5).stdout.strip()
+            for pid_str in out.splitlines():
+                pid_str = pid_str.strip()
+                if not pid_str or not pid_str.isdigit():
                     continue
-            except OSError:
-                pass
-            try:
-                stat = open(f"/proc/{pid}/stat").read().split()
-                start_ticks = int(stat[21])
-                import time as _t
-                ticks_per_sec = os.sysconf('SC_CLK_TCK')
-                start_time = start_ticks / ticks_per_sec
-                age = _t.time() - start_time
-                if age < 60:
+                pid = int(pid_str)
+                if pid == my_pid or pid == my_ppid:
                     continue
-            except Exception:
-                continue
-            try:
-                os.kill(pid, signal.SIGTERM)
-                killed.append(pid)
-            except (ProcessLookupError, PermissionError):
-                pass
+                try:
+                    pgid = os.getpgid(pid)
+                    if pgid == my_pgid:
+                        continue
+                except OSError:
+                    pass
+                try:
+                    stat = open(f"/proc/{pid}/stat").read().split()
+                    start_ticks = int(stat[21])
+                    ticks_per_sec = os.sysconf('SC_CLK_TCK')
+                    start_time = start_ticks / ticks_per_sec
+                    age = time.time() - start_time
+                    if age < 60:
+                        continue
+                except Exception:
+                    continue
+                try:
+                    os.kill(pid, signal.SIGTERM)
+                    killed.append(pid)
+                except (ProcessLookupError, PermissionError):
+                    pass
     except Exception:
         pass
     try:
@@ -8621,6 +8663,7 @@ def _anti_stall_cleanup():
         print(f"[anti-stall] killed {len(killed)} stale proxy process(es): {killed}", flush=True)
         time.sleep(1)
 
+
 def main():
     global SERVER, _START_TIME
     _START_TIME = time.time()