[Frontend] Re-enable running MaxSim on GPU (#38620)
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
@@ -26,13 +26,18 @@ TEXTS_2 = [
|
||||
]
|
||||
|
||||
|
||||
@pytest.fixture(scope="module")
|
||||
def server():
|
||||
@pytest.fixture(scope="module", params=[True, False])
|
||||
def server(request):
|
||||
args = [
|
||||
"--max-model-len",
|
||||
str(MAX_MODEL_LEN),
|
||||
]
|
||||
|
||||
# Test run pooling score MaxSim on worker side (GPU)
|
||||
# aka flash-late-interaction
|
||||
if not request.param:
|
||||
args += ["--no-enable-flash-late-interaction"]
|
||||
|
||||
with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
|
||||
yield remote_server
|
||||
|
||||
|
||||
@@ -4,12 +4,12 @@
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.entrypoints.pooling.scoring.utils import compute_maxsim_score
|
||||
from vllm.pooling_params import LateInteractionParams, PoolingParams
|
||||
from vllm.v1.pool.late_interaction import (
|
||||
LATE_INTERACTION_MODE_CACHE_QUERY,
|
||||
build_late_interaction_doc_params,
|
||||
build_late_interaction_query_params,
|
||||
compute_maxsim_score,
|
||||
)
|
||||
from vllm.v1.worker.gpu.pool.late_interaction_runner import LateInteractionRunner
|
||||
|
||||
|
||||
Reference in New Issue
Block a user