biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix] Fix async scheduling + reasoning with struct output (#31332)

Browse Source 
Signed-off-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
Nick Hill
2025-12-25 15:01:02 -08:00
committed by GitHub
parent f1531d9f2a
commit 81786c8774
3 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/v1/entrypoints/llm/test_struct_output_generate.py
View File

@@ -608,7 +608,7 @@ Make the response as short as possible.
@pytest.mark.parametrize(
"model_name, backend, tokenizer_mode, reasoning_parser, speculative_config", # noqa: E501
"model_name, backend, tokenizer_mode, reasoning_parser, speculative_config, async_scheduling", # noqa: E501
[
(
"deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
@@ -616,8 +616,10 @@ Make the response as short as possible.
"auto",
"deepseek_r1",
NGRAM_SPEC_CONFIG,
False,
),
("Qwen/Qwen3-1.7B", "xgrammar", "auto", "deepseek_r1", None),
("Qwen/Qwen3-1.7B", "xgrammar", "auto", "deepseek_r1", None, False),
("Qwen/Qwen3-1.7B", "xgrammar", "auto", "deepseek_r1", None, True),
],
)
def test_structured_output_with_reasoning_matrices(
@@ -626,6 +628,7 @@ def test_structured_output_with_reasoning_matrices(
reasoning_parser: str,
model_name: str,
speculative_config: dict[str, Any] | None,
async_scheduling: bool,
):
if current_platform.is_tpu() and speculative_config:
pytest.skip("TPU does not support speculative decoding")
@@ -646,6 +649,7 @@ def test_structured_output_with_reasoning_matrices(
),
tokenizer_mode=tokenizer_mode,
speculative_config=speculative_config,
async_scheduling=async_scheduling,
)
tokenizer = llm.get_tokenizer()
reasoner = ReasoningParserManager.get_reasoning_parser(reasoning_parser)(