Optimize Qwen-Image 2.1 attention and decode graph reuse (#7790)

Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn>
This commit is contained in:
Tianyu Guo authored and GitHub committed 2026-09-18 19:37:19 +08:00
1 parent f930348e79
commit ab4cf24b3e
24 files changed
+1147 -414

No files matched your search

+5 -3
View File
@@ -1605,8 +1605,8 @@ def _build_image_generation_response(
peak_memory_mb=peak_memory_mb,
),
}
if request.size is not None:
response_kwargs["size"] = request.size
if images:
response_kwargs["size"] = f"{images[0].width}x{images[0].height}"
response = ImageGenerationResponse(**response_kwargs)
if request.response_format == ResponseFormat.FILE:
return response.stream_response()
@@ -1839,6 +1839,7 @@ async def generate_images(
@router.post(
"/v1/images/edits",
response_model=None,
responses={
HTTPStatus.OK.value: {"model": ImageGenerationResponse},
HTTPStatus.BAD_REQUEST.value: {"model": ErrorResponse},
@@ -1846,6 +1847,7 @@ async def generate_images(
HTTPStatus.INTERNAL_SERVER_ERROR.value: {"model": ErrorResponse},
},
)
@with_cancellation
async def edit_images(
raw_request: Request,
image: list[UploadFile] | None = File(None),
@@ -2181,7 +2183,7 @@ async def edit_images(
created=int(time.time()),
data=image_data,
output_format=output_format,
size=size_str,
size=f"{images[0].width}x{images[0].height}" if images else size_str,
cot_output=cot_output,
metrics=_build_image_response_metrics(
response_metrics=response_metrics,