Fix Qwen3 streaming content routing (#40820)

Signed-off-by: xy3 <120182408@qq.com>
Signed-off-by: sfeng33 <4florafeng@gmail.com>
Co-authored-by: sfeng33 <4florafeng@gmail.com>
This commit is contained in:
xy3
2026-05-06 17:22:01 -04:00
committed by GitHub
co-authored by sfeng33
parent deb737e323
commit 50acdc5b5c
4 changed files with 220 additions and 7 deletions
@@ -287,3 +287,47 @@ async def test_thinking_token_budget_qwen35_fp8_mtp_concurrent_mixed_budget_and_
f"index {i}: reasoning decode token ids ({n_reason}) != "
f"thinking_token_budget ({expected_budget})"
)
@pytest.mark.asyncio
@pytest.mark.parametrize("client", ["default", "auto_config"], indirect=True)
async def test_streaming_with_thinking_disabled_stays_in_content(
client: openai.AsyncOpenAI,
):
request_kwargs = {
"model": MODEL_NAME,
"messages": [
{
"role": "user",
"content": "Which is larger, 4 or 12?"
" Output exactly one token: 4 or 12.",
}
],
"max_tokens": 16,
"temperature": 0.0,
"extra_body": {"chat_template_kwargs": {"enable_thinking": False}},
}
response = await client.chat.completions.create(**request_kwargs)
message = response.choices[0].message
assert message.content is not None and message.content.strip() != ""
assert getattr(message, "reasoning", None) in (None, "")
stream = await client.chat.completions.create(
**request_kwargs,
stream=True,
)
content_chunks = []
reasoning_chunks = []
async for chunk in stream:
if not chunk.choices:
continue
delta = chunk.choices[0].delta
if getattr(delta, "content", None):
content_chunks.append(delta.content)
if getattr(delta, "reasoning", None):
reasoning_chunks.append(delta.reasoning)
assert "".join(content_chunks).strip() != ""
assert reasoning_chunks == []