[Frontend] Parallelize preprocessing within the same request for pooling models online serving. (#49153)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-07-22 10:56:23 +00:00
committed by GitHub
parent d6dbdb9b0d
commit 61c9ef986a
17 changed files with 538 additions and 820 deletions
+5 -8
View File
@@ -18,8 +18,11 @@ from .serving import ServingEmbedding
router = APIRouter()
def embedding(request: Request) -> ServingEmbedding | None:
return request.app.state.serving_embedding
def embedding(request: Request) -> ServingEmbedding:
handler = getattr(request.app.state, "serving_embedding", None)
if handler is None:
raise NotImplementedError("The model does not support Embeddings API")
return handler
@router.post(
@@ -37,9 +40,6 @@ async def create_embedding(
raw_request: Request,
):
handler = embedding(raw_request)
if handler is None:
raise NotImplementedError("The model does not support Embeddings API")
return await handler(request, raw_request)
@@ -58,7 +58,4 @@ async def create_cohere_embedding(
raw_request: Request,
):
handler = embedding(raw_request)
if handler is None:
raise NotImplementedError("The model does not support Embeddings API")
return await handler(request, raw_request)