Method setUp

tests/trainer/trainer_test.py:1130–1162

(self)

class TestServeWithTrainer(RayUnittestBaseAsync):
    def setUp(self):
        if multiprocessing.get_start_method(allow_none=True) != "spawn":
            multiprocessing.set_start_method("spawn", force=True)
        checkpoint_path = get_checkpoint_path()
        shutil.rmtree(os.path.join(checkpoint_path, "unittest"), ignore_errors=True)

        config = get_template_config()
        config.project = "unittest"
        config.name = f"serve_with_trainer_{datetime.now().strftime('%Y%m%d%H%M%S')}"
        config.checkpoint_root_dir = get_checkpoint_path()
        config.model.model_path = get_model_path()
        config.buffer.batch_size = 4
        config.buffer.train_batch_size = 4
        config.algorithm.algorithm_type = "ppo"
        config.algorithm.repeat_times = 1
        config.cluster.gpu_per_node = 2
        config.cluster.node_num = 1
        config.buffer.trainer_input.experience_buffer = ExperienceBufferConfig(
            name="exp_buffer",
            storage_type=StorageType.SQL.value,
            schema_type="experience",
        )
        config.buffer.explorer_input.taskset = get_unittest_dataset_config("gsm8k")
        config.buffer.total_steps = 3
        config.trainer.save_interval = 1
        config.synchronizer.sync_interval = 1
        config.synchronizer.sync_method = SyncMethod.CHECKPOINT
        config.explorer.rollout_model.engine_num = 2
        config.explorer.rollout_model.enable_openai_api = True
        config.explorer.rollout_model.tensor_parallel_size = 1
        config.explorer.service_status_check_interval = 5
        self.config = config
        self.process_list = []

    async def test_serve_with_trainer(self):  # noqa: C901
        trainer_config = deepcopy(self.config)