From 6098c51bc229c360857d36832f4e6bdaa8c65dde Mon Sep 17 00:00:00 2001
From: shuwenn <47200617+alphabetc1@users.noreply.github.com>
Date: Thu, 2 Apr 2026 00:47:27 +0800
Subject: [PATCH] fix(MiMo-V2-Flash): add mimo reasoning parser (#21414)

---
 python/sglang/srt/entrypoints/openai/serving_chat.py | 6 ++++++
 python/sglang/srt/parser/reasoning_parser.py         | 1 +
 2 files changed, 7 insertions(+)

diff --git a/python/sglang/srt/entrypoints/openai/serving_chat.py b/python/sglang/srt/entrypoints/openai/serving_chat.py
index 67cfef0a9..f7dffd111 100644
--- a/python/sglang/srt/entrypoints/openai/serving_chat.py
+++ b/python/sglang/srt/entrypoints/openai/serving_chat.py
@@ -1268,6 +1268,12 @@ class OpenAIServingChat(OpenAIServingBase):
                 not request.chat_template_kwargs
                 or request.chat_template_kwargs.get("enable_thinking") is not False
             )
+        if self.reasoning_parser in ["mimo"]:
+            # Models that require explicit enable thinking (enable_thinking=True)
+            return (
+                request.chat_template_kwargs is not None
+                and request.chat_template_kwargs.get("enable_thinking") is True
+            )
         if self.reasoning_parser in ["mistral"]:
             # Mistral models only reason when reasoning_effort is explicitly
             # set to a value other than None/"none" (typically "high").
diff --git a/python/sglang/srt/parser/reasoning_parser.py b/python/sglang/srt/parser/reasoning_parser.py
index a6867c9f8..c3dbb3116 100644
--- a/python/sglang/srt/parser/reasoning_parser.py
+++ b/python/sglang/srt/parser/reasoning_parser.py
@@ -495,6 +495,7 @@ class ReasoningParser:
         "gpt-oss": GptOssDetector,
         "kimi": KimiDetector,
         "kimi_k2": KimiK2Detector,
+        "mimo": Qwen3Detector,
         "qwen3": Qwen3Detector,
         "qwen3-thinking": Qwen3Detector,
         "minimax": Qwen3Detector,