From ced693e845508a318e8f8c1d7395d8f2c8d4620b Mon Sep 17 00:00:00 2001
From: Xu Wenqing <121550081+Xu-Wenqing@users.noreply.github.com>
Date: Fri, 26 Sep 2025 17:25:39 +0800
Subject: [PATCH] Support LongCat-Flash-Chat tool call (#24083)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
---
 docs/features/tool_calling.md                 |  9 +++++
 .../openai/tool_parsers/__init__.py           |  2 +
 .../tool_parsers/longcat_tool_parser.py       | 39 +++++++++++++++++++
 3 files changed, 50 insertions(+)
 create mode 100644 vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py

diff --git a/docs/features/tool_calling.md b/docs/features/tool_calling.md
index 291c313cd57af..16693db7255ca 100644
--- a/docs/features/tool_calling.md
+++ b/docs/features/tool_calling.md
@@ -310,6 +310,15 @@ Flags:
 * For non-reasoning: `--tool-call-parser hunyuan_a13b`
 * For reasoning: `--tool-call-parser hunyuan_a13b --reasoning-parser hunyuan_a13b --enable_reasoning`
 
+### LongCat-Flash-Chat Models (`longcat`)
+
+Supported models:
+
+* `meituan-longcat/LongCat-Flash-Chat`
+* `meituan-longcat/LongCat-Flash-Chat-FP8`
+
+Flags: `--tool-call-parser longcat`
+
 ### GLM-4.5 Models (`glm45`)
 
 Supported models:
diff --git a/vllm/entrypoints/openai/tool_parsers/__init__.py b/vllm/entrypoints/openai/tool_parsers/__init__.py
index 5e77c406b8d92..2c5a0a6af23f0 100644
--- a/vllm/entrypoints/openai/tool_parsers/__init__.py
+++ b/vllm/entrypoints/openai/tool_parsers/__init__.py
@@ -14,6 +14,7 @@ from .jamba_tool_parser import JambaToolParser
 from .kimi_k2_tool_parser import KimiK2ToolParser
 from .llama4_pythonic_tool_parser import Llama4PythonicToolParser
 from .llama_tool_parser import Llama3JsonToolParser
+from .longcat_tool_parser import LongcatFlashToolParser
 from .minimax_tool_parser import MinimaxToolParser
 from .mistral_tool_parser import MistralToolParser
 from .openai_tool_parser import OpenAIToolParser
@@ -36,6 +37,7 @@ __all__ = [
     "Llama3JsonToolParser",
     "JambaToolParser",
     "Llama4PythonicToolParser",
+    "LongcatFlashToolParser",
     "PythonicToolParser",
     "Phi4MiniJsonToolParser",
     "DeepSeekV3ToolParser",
diff --git a/vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py b/vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py
new file mode 100644
index 0000000000000..87a3fdc44397e
--- /dev/null
+++ b/vllm/entrypoints/openai/tool_parsers/longcat_tool_parser.py
@@ -0,0 +1,39 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import regex as re
+
+from vllm.entrypoints.openai.tool_parsers.abstract_tool_parser import (
+    ToolParserManager)
+from vllm.entrypoints.openai.tool_parsers.hermes_tool_parser import (
+    Hermes2ProToolParser)
+from vllm.transformers_utils.tokenizer import AnyTokenizer
+
+
+@ToolParserManager.register_module("longcat")
+class LongcatFlashToolParser(Hermes2ProToolParser):
+
+    def __init__(self, tokenizer: AnyTokenizer):
+        super().__init__(tokenizer)
+
+        self.tool_call_start_token: str = "<longcat_tool_call>"
+        self.tool_call_end_token: str = "</longcat_tool_call>"
+
+        self.tool_call_regex = re.compile(
+            r"<longcat_tool_call>(.*?)</longcat_tool_call>|<longcat_tool_call>(.*)",
+            re.DOTALL)
+
+        self.tool_call_start_token_ids = self.model_tokenizer.encode(
+            self.tool_call_start_token, add_special_tokens=False)
+        self.tool_call_end_token_ids = self.model_tokenizer.encode(
+            self.tool_call_end_token, add_special_tokens=False)
+
+        self.tool_call_start_token_array = [
+            self.model_tokenizer.decode([token_id])
+            for token_id in self.tool_call_start_token_ids
+        ]
+
+        self.tool_call_end_token_array = [
+            self.model_tokenizer.decode([token_id])
+            for token_id in self.tool_call_end_token_ids
+        ]