def test_nvt_tf_rossmann_inference_triton(asv_db, bench_info, n_rows, err_tol):
with test_utils.run_triton_server(
os.path.expanduser(MODEL_DIR),
"rossmann",
TRITON_SERVER_PATH,
TRITON_DEVICE_ID,
"tensorflow",
) as client:
diff, run_time = _run_rossmann_query(client, n_rows)
assert (diff < err_tol).all()
benchmark_results = []
result = create_bench_result("test_nvt_tf_rossmann_inference_triton",
[("n_rows", n_rows)], run_time,
"datetime")
benchmark_results.append(result)
def test_nvt_tf_movielens_inference_triton_mt(asv_db, bench_info, n_rows,
err_tol):
futures = []
with test_utils.run_triton_server(
os.path.expanduser(MODEL_DIR),
"movielens",
TRITON_SERVER_PATH,
TRITON_DEVICE_ID,
"tensorflow",
) as client:
with concurrent.futures.ThreadPoolExecutor() as executor:
for n_row in n_rows:
futures.append(
executor.submit(_run_movielens_query, client, n_row))
for future in concurrent.futures.as_completed(futures):
diff, run_time = future.result()
assert (diff < err_tol).all()
benchmark_results = []
result = create_bench_result(
"test_nvt_tf_movielens_inference_triton_mt", [("n_rows", n_rows)],
run_time, "datetime")
benchmark_results.append(result)
def test_inference(n_rows, err_tol):
warnings.simplefilter("ignore")
data_path = DATA_DIR + "test/data.csv"
output_path = DATA_DIR + "test/output.csv"
ps_file = TRAIN_DIR + "ps.json"
workflow_path = MODEL_DIR + MODEL_NAME + "_nvt/1/workflow"
_write_ps_hugectr(ps_file, MODEL_NAME, SPARSE_FILES, DENSE_FILE, NETWORK_FILE)
with test_utils.run_triton_server(
os.path.expanduser(MODEL_DIR),
MODEL_NAME + "_ens",
TRITON_SERVER_PATH,
TRITON_DEVICE_ID,
"hugectr",
ps_file,
) as client:
diff, run_time = _run_query(
client,
n_rows,
MODEL_NAME + "_ens",
workflow_path,
data_path,
output_path,
"OUTPUT0",
CATEGORICAL_COLUMNS,
"hugectr",
)
assert (diff < err_tol).all()
benchmark_results = []
result = create_bench_result(
"test_nvt_hugectr_inference", [("n_rows", n_rows)], run_time, "datetime"
)
benchmark_results.append(result)