minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

config: ServerArgs holds the raw input (#36255)

Browse Source
This commit is contained in:
Cheng Wan
2026-08-26 05:14:05 -07:00
committed by GitHub
parent 27c36368b6
commit 413df1f8db
34 changed files with 1201 additions and 383 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/unit/parser/test_template_manager.py
+38 -26
View File
@@ -765,6 +765,18 @@ class TestToolCallParserDetection(unittest.TestCase):
self.assertEqual(result, "minicpm5")
def _declared(server_args, field):
"""What late resolution decided for `field` on this record.
`resolve_auto_parsers` declares; the field keeps what the operator passed,
so the decision is read through the resolution result -- the same surface
the config bags are projected from.
"""
from sglang.srt.arg_groups.overrides import resolution_result
return resolution_result(server_args, field)
class TestResolveAutoParsers(unittest.TestCase):
"""Tests for resolve_auto_parsers()."""
@@ -790,8 +802,8 @@ class TestResolveAutoParsers(unittest.TestCase):
with _patch_hf_transformers_utils(Mock(return_value=tokenizer)):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "qwen3")
self.assertEqual(args.tool_call_parser, "qwen")
self.assertEqual(_declared(args, "reasoning_parser"), "qwen3")
self.assertEqual(_declared(args, "tool_call_parser"), "qwen")
def test_resolves_reasoning_parser_only(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser=None)
@@ -800,8 +812,8 @@ class TestResolveAutoParsers(unittest.TestCase):
with _patch_hf_transformers_utils(Mock(return_value=tokenizer)):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "qwen3")
self.assertIsNone(args.tool_call_parser)
self.assertEqual(_declared(args, "reasoning_parser"), "qwen3")
self.assertIsNone(_declared(args, "tool_call_parser"))
def test_resolves_tool_call_parser_only(self):
args = self._make_server_args(reasoning_parser="qwen3", tool_call_parser="auto")
@@ -810,14 +822,14 @@ class TestResolveAutoParsers(unittest.TestCase):
with _patch_hf_transformers_utils(Mock(return_value=tokenizer)):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "qwen3")
self.assertEqual(args.tool_call_parser, "qwen")
self.assertEqual(_declared(args, "reasoning_parser"), "qwen3")
self.assertEqual(_declared(args, "tool_call_parser"), "qwen")
def test_neither_auto_is_noop(self):
args = self._make_server_args(reasoning_parser="qwen3", tool_call_parser="qwen")
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "qwen3")
self.assertEqual(args.tool_call_parser, "qwen")
self.assertEqual(_declared(args, "reasoning_parser"), "qwen3")
self.assertEqual(_declared(args, "tool_call_parser"), "qwen")
def test_nonexistent_model_disables_both_parsers(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -832,8 +844,8 @@ class TestResolveAutoParsers(unittest.TestCase):
):
resolve_auto_parsers(args)
self.assertIsNone(args.reasoning_parser)
self.assertIsNone(args.tool_call_parser)
self.assertIsNone(_declared(args, "reasoning_parser"))
self.assertIsNone(_declared(args, "tool_call_parser"))
def test_none_chat_template_disables_both_parsers(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -842,8 +854,8 @@ class TestResolveAutoParsers(unittest.TestCase):
with _patch_hf_transformers_utils(Mock(return_value=tokenizer)):
resolve_auto_parsers(args)
self.assertIsNone(args.reasoning_parser)
self.assertIsNone(args.tool_call_parser)
self.assertIsNone(_declared(args, "reasoning_parser"))
self.assertIsNone(_declared(args, "tool_call_parser"))
def test_deepseek_v32_arch_without_chat_template_uses_custom_encoder(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -855,8 +867,8 @@ class TestResolveAutoParsers(unittest.TestCase):
):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "deepseek-v3")
self.assertEqual(args.tool_call_parser, "deepseekv32")
self.assertEqual(_declared(args, "reasoning_parser"), "deepseek-v3")
self.assertEqual(_declared(args, "tool_call_parser"), "deepseekv32")
def test_deepseek_v4_arch_without_chat_template_uses_custom_encoder(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -868,8 +880,8 @@ class TestResolveAutoParsers(unittest.TestCase):
):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "deepseek-v4")
self.assertEqual(args.tool_call_parser, "deepseekv4")
self.assertEqual(_declared(args, "reasoning_parser"), "deepseek-v4")
self.assertEqual(_declared(args, "tool_call_parser"), "deepseekv4")
def test_kimi_k3_arch_without_chat_template_uses_custom_encoder(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -883,8 +895,8 @@ class TestResolveAutoParsers(unittest.TestCase):
):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "kimi_k3")
self.assertEqual(args.tool_call_parser, "kimi_k3")
self.assertEqual(_declared(args, "reasoning_parser"), "kimi_k3")
self.assertEqual(_declared(args, "tool_call_parser"), "kimi_k3")
def test_kimi_k3_model_type_without_architecture_uses_custom_encoder(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -896,8 +908,8 @@ class TestResolveAutoParsers(unittest.TestCase):
):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "kimi_k3")
self.assertEqual(args.tool_call_parser, "kimi_k3")
self.assertEqual(_declared(args, "reasoning_parser"), "kimi_k3")
self.assertEqual(_declared(args, "tool_call_parser"), "kimi_k3")
def test_deepseek_arch_fallback_runs_when_tokenizer_load_fails(self):
args = self._make_server_args(reasoning_parser="auto", tool_call_parser="auto")
@@ -909,8 +921,8 @@ class TestResolveAutoParsers(unittest.TestCase):
):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "deepseek-v3")
self.assertEqual(args.tool_call_parser, "deepseekv32")
self.assertEqual(_declared(args, "reasoning_parser"), "deepseek-v3")
self.assertEqual(_declared(args, "tool_call_parser"), "deepseekv32")
def test_explicit_non_jinja_template_skips_architecture_fallback(self):
args = self._make_server_args(
@@ -926,8 +938,8 @@ class TestResolveAutoParsers(unittest.TestCase):
resolve_auto_parsers(args)
get_config.assert_not_called()
self.assertIsNone(args.reasoning_parser)
self.assertIsNone(args.tool_call_parser)
self.assertIsNone(_declared(args, "reasoning_parser"))
self.assertIsNone(_declared(args, "tool_call_parser"))
def test_explicit_jinja_template_takes_precedence(self):
tokenizer = _DummyTokenizer([], chat_template=None)
@@ -947,8 +959,8 @@ class TestResolveAutoParsers(unittest.TestCase):
with _patch_hf_transformers_utils(Mock(return_value=tokenizer)):
resolve_auto_parsers(args)
self.assertEqual(args.reasoning_parser, "deepseek-v3")
self.assertEqual(args.tool_call_parser, "deepseekv32")
self.assertEqual(_declared(args, "reasoning_parser"), "deepseek-v3")
self.assertEqual(_declared(args, "tool_call_parser"), "deepseekv32")
if __name__ == "__main__":