test_pytorch_sampling_backend.py

import json
import unittest
from types import SimpleNamespace

import requests

from sglang.srt.utils import kill_child_process
from sglang.test.run_eval import run_eval
from sglang.test.test_utils import (
    DEFAULT_MODEL_NAME_FOR_TEST,
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
    DEFAULT_URL_FOR_TEST,
    popen_launch_server,
)


class TestPyTorchSamplingBackend(unittest.TestCase):
    @classmethod
    def setUpClass(cls):
        cls.model = DEFAULT_MODEL_NAME_FOR_TEST
        cls.base_url = DEFAULT_URL_FOR_TEST
        cls.process = popen_launch_server(
            cls.model,
            cls.base_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=["--sampling-backend", "pytorch"],
        )

    @classmethod
    def tearDownClass(cls):
        kill_child_process(cls.process.pid)

    def test_mmlu(self):
        args = SimpleNamespace(
            base_url=self.base_url,
            model=self.model,
            eval_name="mmlu",
            num_examples=64,
            num_threads=32,
        )

        metrics = run_eval(args)
        assert metrics["score"] >= 0.65

    def test_greedy(self):
        response_single = requests.post(
            self.base_url + "/generate",
            json={
                "text": "The capital of France is",
                "sampling_params": {
                    "temperature": 0,
                    "max_new_tokens": 32,
                },
            },
        ).json()
        response_batch = requests.post(
            self.base_url + "/generate",
            json={
                "text": ["The capital of France is"] * 10,
                "sampling_params": {
                    "temperature": 0,
                    "max_new_tokens": 32,
                },
            },
        ).json()
        text = response_single["text"]
        print(text)
        for i in range(10):
            assert response_batch[i]["text"] == text


if __name__ == "__main__":
    unittest.main()