A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[fix] fix seed-oss-parser (#23560)
Signed-off-by: jiabin.00 <jiabin.00@bytedance.com>
Bin Jia committed
Aug 26, 2025 at 14:16 UTC
959783fb996d0d15598f45ca12ffcbee4b681424
2 files changed
+5
-7
tests/tool_use/test_seed_oss_tool_parser.py
+2
-7
@@ -102,9 +102,7 @@ def test_extract_tool_calls_no_tools(seed_oss_tool_parser):
102
],
103
argnames=["model_output", "expected_tool_calls", "expected_content"],
104
argvalues=[
105
- ("""<seed:think>\n</seed:cot_budget_reflect>\n</seed:cot_budget_reflect>\n"""
106
- """The current thinking budget is 0, so I will directly start answering the question.\n</seed:think>\n"""
107
- """<seed:tool_call>\n<function=get_weather>\n"""
105
+ ("""<seed:tool_call>\n<function=get_weather>\n"""
106
"""<parameter=location>Barcelona, Spain</parameter>\n</function>\n</seed:tool_call>""",
107
[
108
ToolCall(function=FunctionCall(
@@ -114,10 +112,7 @@ def test_extract_tool_calls_no_tools(seed_oss_tool_parser):
112
}, ),
113
),
114
type='function')
117
- ],
118
- """<seed:think>\n</seed:cot_budget_reflect>\n</seed:cot_budget_reflect>\n"""
119
- """The current thinking budget is 0, so I will directly start answering the question.\n</seed:think>\n"""
120
- ),
115
+ ], None),
116
(
117
"""<seed:think>The user\'s current thinking budget is 512.</seed:cot_budget_reflect>\nLet me analyze the """
118
"""question. The user wants to know the weather in Barcelona, Spain. Looking at the functions available, """
vllm/entrypoints/openai/tool_parsers/seed_oss_tool_parser.py
+3
@@ -271,6 +271,9 @@ class SeedOssToolParser(ToolParser):
271
# Extract content after think end token
272
result_content = model_output[think_end_index:]
273
thinking_content = model_output[:think_end_index]
274
+ else:
275
+ thinking_content = ""
276
+ result_content = model_output
277
278
try:
279
function_calls = self._get_function_calls(result_content)