20 lines
544 B
Python
20 lines
544 B
Python
|
# SPDX-License-Identifier: Apache-2.0
|
||
|
|
import subprocess
|
||
|
|
|
||
|
|
import pytest
|
||
|
|
|
||
|
|
MODEL_NAME = "meta-llama/Llama-3.2-1B-Instruct"
|
||
|
|
|
||
|
|
|
||
|
|
@pytest.mark.benchmark
|
||
|
|
def test_bench_throughput():
|
||
|
|
command = [
|
||
|
|
"vllm", "bench", "throughput", "--model", MODEL_NAME, "--input-len",
|
||
|
|
"32", "--output-len", "1", "--enforce-eager", "--load-format", "dummy"
|
||
|
|
]
|
||
|
|
result = subprocess.run(command, capture_output=True, text=True)
|
||
|
|
print(result.stdout)
|
||
|
|
print(result.stderr)
|
||
|
|
|
||
|
|
assert result.returncode == 0, f"Benchmark failed: {result.stderr}"
|