From 40faf44f7a1891dc84e5e798e8403e8ce5ee82c2 Mon Sep 17 00:00:00 2001
From: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
Date: Thu, 21 May 2026 07:34:52 +0100
Subject: [PATCH] [auto-detect] match Ring-2.6/Ling XML kv tool-call format via
 vocab signature (#25366)

---
 .../sglang/srt/managers/template_detection.py | 13 ++++++++++
 .../unit/managers/test_template_manager.py    | 24 +++++++++++++++++++
 2 files changed, 37 insertions(+)

diff --git a/python/sglang/srt/managers/template_detection.py b/python/sglang/srt/managers/template_detection.py
index 7efe8b6ff..0198e75f0 100644
--- a/python/sglang/srt/managers/template_detection.py
+++ b/python/sglang/srt/managers/template_detection.py
@@ -199,6 +199,16 @@ def _is_glm45(ctx):
     )
 
 
+def _is_xml_kv_tool_call(ctx):
+    # Structural signature for the GLM-4.5 / GLM-4.6 style tool-call format
+    # (`<tool_call>name<arg_key>k</arg_key>\n<arg_value>v</arg_value>...</tool_call>`).
+    # Matches any model whose tokenizer carries `<arg_key>` and `<arg_value>` as
+    # added tokens — e.g., inclusionAI/Ring-2.6, which borrows GLM's tool-call
+    # format but doesn't share the `[gMASK]<sop>` / `enable_thinking` family
+    # signature checked by `_is_glm45`.
+    return ctx.has_vocab("<arg_key>") and ctx.has_vocab("<arg_value>")
+
+
 def _is_mimo(ctx):
     return ctx.reasoning_config == ReasoningToggleConfig(
         toggle_param="enable_thinking", default_enabled=False
@@ -268,6 +278,9 @@ TOOL_CALL_PARSER_RULES = (
     DetectionRule(name="interns1", value="interns1", predicate=_is_interns1),
     DetectionRule(name="mistral", value="mistral", predicate=_is_mistral),
     DetectionRule(name="glm45", value="glm45", predicate=_is_glm45),
+    DetectionRule(
+        name="xml_kv_tool_call", value="glm45", predicate=_is_xml_kv_tool_call
+    ),
     DetectionRule(name="mimo", value="mimo", predicate=_is_mimo),
     DetectionRule(name="qwen", value="qwen", predicate=_is_qwen3),
     DetectionRule(name="deepseek_v3", value="deepseekv3", predicate=_is_deepseek_v3),
diff --git a/test/registered/unit/managers/test_template_manager.py b/test/registered/unit/managers/test_template_manager.py
index dd4db0c41..fe7d40c00 100644
--- a/test/registered/unit/managers/test_template_manager.py
+++ b/test/registered/unit/managers/test_template_manager.py
@@ -2,6 +2,7 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.managers.template_detection import (
+    TOOL_CALL_PARSER_RULES,
     ReasoningToggleConfig,
     detect_reasoning_parser,
     detect_reasoning_pattern,
@@ -265,6 +266,12 @@ class TestToolCallParserDetection(unittest.TestCase):
                 ["<|tool_calls_section_begin|>"],
                 "kimi_k2",
             ),
+            (
+                "xml_kv_tool_call_via_vocab",
+                "{% set reasoning_effort = reasoning_effort | default('high', true) %}\n<think>",
+                ["<tool_call>", "<arg_key>", "<arg_value>", "<|endoftext|>"],
+                "glm45",
+            ),
         ]
         for name, template, vocab, expected in cases:
             with self.subTest(name=name):
@@ -274,6 +281,23 @@ class TestToolCallParserDetection(unittest.TestCase):
                 )
                 self.assertEqual(result, expected)
 
+    def test_glm45_rule_precedes_xml_kv_fallback(self):
+        # The specific GLM-4.5 family check must run before the generic
+        # xml_kv_tool_call fallback. Both currently map to "glm45", so the
+        # value-based test above can't catch a swap — assert positions directly.
+        rule_index = {rule.name: i for i, rule in enumerate(TOOL_CALL_PARSER_RULES)}
+        self.assertLess(rule_index["glm45"], rule_index["xml_kv_tool_call"])
+
+    def test_xml_kv_requires_both_arg_tokens(self):
+        template = "Hello {{ user }}"
+        force, config = detect_reasoning_pattern(template)
+        for vocab in (["<arg_key>"], ["<arg_value>"], []):
+            with self.subTest(vocab=vocab):
+                result = detect_tool_call_parser(
+                    template, _DummyTokenizer(vocab), config, force
+                )
+                self.assertIsNone(result)
+
     def test_none_template_returns_none(self):
         self.assertIsNone(detect_tool_call_parser(None, None))