ml-explore · snagnever · May 31, 2026
diff --git a/mlx_lm/tokenizer_utils.py b/mlx_lm/tokenizer_utils.py
@@ -549,6 +549,8 @@ def _infer_tool_parser(chat_template):
     """Attempt to auto-infer a tool parser from the chat template."""
     if not isinstance(chat_template, str):
         return None
+    elif "<｜DSML｜tool_calls>" in chat_template:
+        return "deepseek_dsml"
     elif "<minimax:tool_call>" in chat_template:
         return "minimax_m2"
     elif "<|tool_call>" in chat_template and "<tool_call|>" in chat_template:

diff --git a/mlx_lm/tool_parsers/deepseek_dsml.py b/mlx_lm/tool_parsers/deepseek_dsml.py
@@ -0,0 +1,79 @@
+# Tool parser for DeepSeek-V4's native DSML tool-call format. Deploy into
+# venv: mlx_lm/tool_parsers/deepseek_dsml.py, set tokenizer_config
+# "tool_parser_type": "deepseek_dsml", and use the official DSML chat template.
+#
+# Format (multiple invokes per tool_calls block => native parallel calls):
+#   <｜DSML｜tool_calls>
+#   <｜DSML｜invoke name="get_weather">
+#   <｜DSML｜parameter name="city" string="true">Paris</｜DSML｜parameter>
+#   <｜DSML｜parameter name="days" string="false">3</｜DSML｜parameter>
+#   </｜DSML｜invoke>
+#   ...
+#   </｜DSML｜tool_calls>
+#
+# string="true"  -> value is a literal string
+# string="false" -> value is JSON (numbers, booleans, arrays, objects)
+#
+# Start/end markers drop the trailing ">": mlx-lm matches markers by token-id
+# sequence and BPE merges ">" with the next byte ("...calls>\n" -> one token),
+# so the precomputed full marker never matches. "<｜DSML｜tool_calls" (the
+# <,｜DSML｜,tool,_c,alls tokens) is stable; the parser tolerates the leftover ">".
+import json
+
+import regex as re
+
+tool_call_start = "<｜DSML｜tool_calls"
+tool_call_end = "</｜DSML｜tool_calls"
+
+_INVOKE = re.compile(r"<｜DSML｜invoke\s+name=(.*?)</｜DSML｜invoke>", re.DOTALL)
+_PARAM = re.compile(
+    r'<｜DSML｜parameter\s+name=(?P<name>"[^"]*"|\'[^\']*\'|[^\s>]+)'
+    r'\s+string="(?P<is_str>true|false)"\s*>(?P<val>.*?)</｜DSML｜parameter>',
+    re.DOTALL,
+)
+_NAME_BODY = re.compile(r'\s*(?P<name>"[^"]*"|\'[^\']*\'|[^\s>]+)\s*>(?P<body>.*)', re.DOTALL)
+
+
+def _unquote(s):
+    s = s.strip()
+    if len(s) >= 2 and s[0] in "\"'" and s[-1] == s[0]:
+        return s[1:-1]
+    return s
+
+
+def _strip_one_newline(v):
+    if v.startswith("\n"):
+        v = v[1:]
+    if v.endswith("\n"):
+        v = v[:-1]
+    return v
+
+
+def parse_tool_call(text, tools=None):
+    calls = []
+    for invoke in _INVOKE.findall(text):
+        m = _NAME_BODY.match(invoke)
+        if not m:
+            continue
+        name = _unquote(m.group("name"))
+        args = {}
+        for pm in _PARAM.finditer(m.group("body")):
+            pname = _unquote(pm.group("name"))
+            val = _strip_one_newline(pm.group("val"))
+            if pm.group("is_str") == "true":
+                args[pname] = val
+            else:
+                try:
+                    args[pname] = json.loads(val)
+                except (json.JSONDecodeError, ValueError):
+                    args[pname] = val
+        # template fallback: whole arg blob passed as a single "arguments" param
+        if set(args) == {"arguments"} and isinstance(args["arguments"], str):
+            try:
+                args = json.loads(args["arguments"])
+            except (json.JSONDecodeError, ValueError):
+                pass
+        calls.append({"name": name, "arguments": args})
+    if not calls:
+        raise ValueError("no DSML tool call found")
+    return calls[0] if len(calls) == 1 else calls
diff --git a/tests/test_tool_parsing.py b/tests/test_tool_parsing.py
@@ -2,6 +2,7 @@
 from pathlib import Path
 
 from mlx_lm.tool_parsers import (
+    deepseek_dsml,
     function_gemma,
     gemma4,
     glm47,
@@ -33,6 +34,10 @@ def test_parsers(self):
                 '{"name": "multiply", "arguments": {"a": 12234585, "b": 48838483920}}',
                 json_tools,
             ),
+            (
+                '>\n<｜DSML｜invoke name="multiply">\n<｜DSML｜parameter name="a" string="false">12234585</｜DSML｜parameter>\n<｜DSML｜parameter name="b" string="false">48838483920</｜DSML｜parameter>\n</｜DSML｜invoke>\n',
+                deepseek_dsml,
+            ),
             (
                 '<invoke name="multiply">\n<parameter name="a">12234585</parameter>\n<parameter name="b">48838483920</parameter>\n</invoke>',
                 minimax_m2,
@@ -103,6 +108,10 @@ def test_parsers(self):
                 '{"name": "get_current_temperature", "arguments": {"location": "London"}}',
                 json_tools,
             ),
+            (
+                '>\n<｜DSML｜invoke name="get_current_temperature">\n<｜DSML｜parameter name="location" string="true">London</｜DSML｜parameter>\n</｜DSML｜invoke>\n',
+                deepseek_dsml,
+            ),
             (
                 '<invoke name="get_current_temperature">\n<parameter name="location">London</parameter>\n</invoke>',
                 minimax_m2,
@@ -329,6 +338,36 @@ def test_minimax_m2(self):
         tool_calls = minimax_m2.parse_tool_call(test_case, None)
         self.assertEqual(expected, tool_calls)
 
+    def test_deepseek_dsml_parallel(self):
+        # DSML puts multiple <｜DSML｜invoke> in one <｜DSML｜tool_calls> block = native
+        # parallel calls. The leading ">" is the leftover the server captures after the
+        # "<｜DSML｜tool_calls" prefix start marker (which drops the ">" that BPE merges).
+        test_case = (
+            ">\n"
+            '<｜DSML｜invoke name="get_weather">\n'
+            '<｜DSML｜parameter name="city" string="true">Tokyo</｜DSML｜parameter>\n'
+            "</｜DSML｜invoke>\n"
+            '<｜DSML｜invoke name="get_weather">\n'
+            '<｜DSML｜parameter name="city" string="true">Paris</｜DSML｜parameter>\n'
+            "</｜DSML｜invoke>\n"
+        )
+        expected = [
+            {"name": "get_weather", "arguments": {"city": "Tokyo"}},
+            {"name": "get_weather", "arguments": {"city": "Paris"}},
+        ]
+        self.assertEqual(deepseek_dsml.parse_tool_call(test_case, None), expected)
+
+        # string="true" -> literal string; string="false" -> JSON-decoded value
+        test_case = (
+            '<｜DSML｜invoke name="calc">\n'
+            '<｜DSML｜parameter name="expr" string="true">2+2</｜DSML｜parameter>\n'
+            '<｜DSML｜parameter name="n" string="false">3</｜DSML｜parameter>\n'
+            '<｜DSML｜parameter name="tags" string="false">["a", "b"]</｜DSML｜parameter>\n'
+            "</｜DSML｜invoke>"
+        )
+        tc = deepseek_dsml.parse_tool_call(test_case, None)
+        self.assertEqual(tc["arguments"], {"expr": "2+2", "n": 3, "tags": ["a", "b"]})
+
 
 if __name__ == "__main__":
     unittest.main()