minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Upgrade xgrammar to 0.2.1 (#25676)

Browse Source
This commit is contained in:
Xinyuan Tong
2026-05-29 11:40:07 +08:00
committed by GitHub
parent 272066566f
commit 79c844527c
13 changed files with 407 additions and 172 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/unit/entrypoints/openai/test_serving_chat.py
+284
View File
@@ -147,6 +147,241 @@ class ServingChatTestCase(unittest.TestCase):
self.assertFalse(adapted.stream)
self.assertEqual(processed, self.basic_req)
def test_kimi_tool_call_keeps_default_reasoning(self):
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {"a": {"type": "integer"}},
},
},
}
],
tool_choice="required",
)
with patch.object(self.chat, "_process_messages") as proc_mock:
proc_mock.return_value = MessageProcessingResult(
"",
[1, 2, 3],
None,
None,
[],
[],
None,
)
adapted, _ = self.chat._convert_to_internal_request(req)
self.assertTrue(adapted.require_reasoning)
def test_kimi_tool_call_keeps_explicit_reasoning(self):
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {"a": {"type": "integer"}},
},
},
}
],
tool_choice="required",
chat_template_kwargs={"thinking": True},
)
with patch.object(self.chat, "_process_messages") as proc_mock:
proc_mock.return_value = MessageProcessingResult(
"",
[1, 2, 3],
None,
None,
[],
[],
None,
)
adapted, _ = self.chat._convert_to_internal_request(req)
self.assertTrue(adapted.require_reasoning)
def test_kimi_tool_call_respects_explicit_reasoning_disable(self):
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {"a": {"type": "integer"}},
},
},
}
],
tool_choice="required",
chat_template_kwargs={"thinking": False},
)
with patch.object(self.chat, "_process_messages") as proc_mock:
proc_mock.return_value = MessageProcessingResult(
"",
[1, 2, 3],
None,
None,
[],
[],
None,
)
adapted, _ = self.chat._convert_to_internal_request(req)
self.assertFalse(adapted.require_reasoning)
def test_kimi_tool_call_keeps_template_default_thinking(self):
self.template_manager.chat_template_name = None
self.template_manager.jinja_template_content_format = "string"
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {"a": {"type": "integer"}},
},
},
}
],
tool_choice="required",
)
self.chat._process_messages(req, is_multimodal=False)
kwargs = self.tm.tokenizer.apply_chat_template.call_args.kwargs
self.assertNotIn("thinking", kwargs)
def test_kimi_tool_call_keeps_explicit_template_thinking(self):
self.template_manager.chat_template_name = None
self.template_manager.jinja_template_content_format = "string"
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {"a": {"type": "integer"}},
},
},
}
],
tool_choice="required",
chat_template_kwargs={"thinking": True},
)
self.chat._process_messages(req, is_multimodal=False)
kwargs = self.tm.tokenizer.apply_chat_template.call_args.kwargs
self.assertTrue(kwargs["thinking"])
def test_kimi_tool_call_keeps_explicit_template_thinking_false(self):
self.template_manager.chat_template_name = None
self.template_manager.jinja_template_content_format = "string"
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {"a": {"type": "integer"}},
},
},
}
],
tool_choice="required",
chat_template_kwargs={"thinking": False},
)
self.chat._process_messages(req, is_multimodal=False)
kwargs = self.tm.tokenizer.apply_chat_template.call_args.kwargs
self.assertFalse(kwargs["thinking"])
def test_jinja_uses_openai_tool_schema_first(self):
"""Ensure Jinja chat templates receive OpenAI-shaped tools by default."""
self.template_manager.chat_template_name = None
@@ -225,6 +460,55 @@ class ServingChatTestCase(unittest.TestCase):
second_tools, [tool.function.model_dump() for tool in req.tools]
)
def test_xgrammar_tag_omits_reasoning_when_parser_owns_it(self):
"""ReasonerGrammarBackend owns the thinking prefix when a parser is set."""
self.template_manager.chat_template_name = None
self.template_manager.jinja_template_content_format = "string"
self.template_manager.reasoning_config = ReasoningToggleConfig(
toggle_param="thinking", default_enabled=True
)
self.tm.server_args.reasoning_parser = "kimi_k2"
self.tm.server_args.tool_call_parser = "kimi_k2"
self.chat.reasoning_parser = "kimi_k2"
self.chat.tool_call_parser = "kimi_k2"
self.tm.tokenizer.apply_chat_template.return_value = [1, 2, 3]
req = ChatCompletionRequest(
model="x",
messages=[{"role": "user", "content": "What is 2+2?"}],
tools=[
{
"type": "function",
"function": {
"name": "add",
"parameters": {
"type": "object",
"properties": {
"a": {"type": "integer"},
"b": {"type": "integer"},
},
"required": ["a", "b"],
},
"strict": True,
},
}
],
tool_choice="required",
)
with patch(
"sglang.srt.entrypoints.openai.serving_chat.FunctionCallParser"
) as parser_cls:
parser = parser_cls.return_value
parser.get_structure_constraint.return_value = ("structural_tag", "tag")
self.chat._process_messages(req, is_multimodal=False)
parser.get_structure_constraint.assert_called_once()
self.assertFalse(
parser.get_structure_constraint.call_args.kwargs["thinking_mode"]
)
def test_stop_str_isolation_between_requests(self):
"""Test that stop strings from one request don't affect subsequent requests.