[Bugfix][Frontend] Emit kv/ec transfer params at the top level of render output (#61241)

Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
This commit is contained in:
Martin Hickey
2026-10-15 12:34:15 +00:00
committed by GitHub
parent e8e1e96196
commit c6b772cf6b
2 changed files with 79 additions and 3 deletions
@@ -12,6 +12,7 @@ import pytest
import pytest_asyncio
from tests.utils import RemoteLaunchRenderServer
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
from vllm.entrypoints.openai.completion.protocol import CompletionRequest
from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
from vllm.entrypoints.scale_out.render.api_router import router
@@ -115,6 +116,7 @@ async def test_render_responses_returns_generate_request_without_stored_state():
assert response.priority == 3
assert response.kv_transfer_params == {"do_remote_prefill": True}
assert response.ec_transfer_params == {"remote": True}
assert response.sampling_params.extra_args is None
serving.online_renderer.render_responses.assert_awaited_once_with(
request,
previous_messages=None,
@@ -124,6 +126,54 @@ async def test_render_responses_returns_generate_request_without_stored_state():
)
@pytest.mark.asyncio
@pytest.mark.skip_global_cleanup
async def test_render_chat_hoists_transfer_params_out_of_extra_args():
serving = _build_serving_render()
serving.online_renderer.render_chat = AsyncMock(
return_value=([], [{"prompt_token_ids": [7, 8, 9]}])
)
request = ChatCompletionRequest(
model=MODEL_NAME,
messages=[{"role": "user", "content": "Test prompt"}],
kv_transfer_params={"do_remote_prefill": True},
ec_transfer_params={"remote": True},
vllm_xargs={"foo": 1},
)
response = await serving.render_chat_request(request)
assert response.kv_transfer_params == {"do_remote_prefill": True}
assert response.ec_transfer_params == {"remote": True}
assert response.sampling_params.extra_args == {"foo": 1}
@pytest.mark.asyncio
@pytest.mark.skip_global_cleanup
async def test_render_completion_hoists_transfer_params_for_every_prompt():
serving = _build_serving_render()
serving.online_renderer.render_completion = AsyncMock(
return_value=[
{"prompt_token_ids": [7, 8, 9]},
{"prompt_token_ids": [4, 5]},
]
)
request = CompletionRequest(
model=MODEL_NAME,
prompt=["a", "b"],
kv_transfer_params={"do_remote_prefill": True},
ec_transfer_params={"remote": True},
)
responses = await serving.render_completion_request(request)
assert len(responses) == 2
for response in responses:
assert response.kv_transfer_params == {"do_remote_prefill": True}
assert response.ec_transfer_params == {"remote": True}
assert response.sampling_params.extra_args is None
@pytest.mark.asyncio
@pytest.mark.skip_global_cleanup
async def test_render_responses_rejects_previous_response_id():