Pylance errors

[SlimRG]

Connected

microsoft/pylance-release#7375

Description

PyLance doesn't see functions and classes

My code:

# CUDA
import pycuda.driver as cuda
import pycuda.autoinit  # Автоинициализация CUDA

# TensorRT
import tensorrt as trt

...

class TrtTextEncoder:
    """
    Универсальная обёртка для text encoder в TensorRT 10+.

    Аргументы:
        engine_path (str): путь к .trt файлу
        model_type (str): "clip" или "t5" — режим работы (для post-processing)
        dtype (torch.dtype): dtype PyTorch для вывода (обычно torch.float16/torch.float32)
        device (torch.device): устройство, куда возвращать тензоры (cpu или cuda)
    """

    def __init__(self,
                 engine_path: str,
                 model_type: str = "clip",
                 dtype: torch.dtype = torch.float16,
                 device: torch.device = torch.device("cpu")):

        assert model_type in ("clip", "t5"), "model_type must be 'clip' or 't5'"
        self.model_type = model_type
        self.dtype = dtype
        self.device = device

        # Инициализируем TensorRT runtime и десериализуем движок
        self.logger = trt.Logger(trt.Logger.WARNING)
        runtime = trt.Runtime(self.logger)
        with open(engine_path, "rb") as f:
            engine_bytes = f.read()
        self.engine = runtime.deserialize_cuda_engine(engine_bytes)
        if self.engine is None:
            raise RuntimeError(f"Failed to load TensorRT engine from {engine_path}")

        # Создаём execution context
        self.context = self.engine.create_execution_context()

        # Собираем все имена входных/выходных тензоров
        self.tensor_names = [
            self.engine.get_tensor_name(i)
            for i in range(self.engine.num_io_tensors())
        ]  # :contentReference[oaicite:4]{index=4}

        # Подготавливаем буферы и список указателей (bindings)
        self._allocate_buffers()

    def _allocate_buffers(self):
        """
        Создаёт host- и device-буферы для всех тензоров,
        а также список device-указателей в порядке engine.get_tensor_name(i).
        """
        self.h_buffers = {}
        self.d_buffers = {}
        self.bindings = []

        for name in self.tensor_names:
            # Получаем форму: кортеж int
            shape = tuple(self.engine.get_tensor_shape(name))
            # Выбираем numpy-dtype
            mode = self.engine.get_tensor_mode(name)
            if mode == trt.TensorIOMode.INPUT:
                dtype_np = np.int32
            else:
                dtype_np = np.float16 if self.dtype == torch.float16 else np.float32

            # Хост-буфер и его размер
            host_mem = np.zeros(shape, dtype=dtype_np)
            dev_mem = cuda.mem_alloc(host_mem.nbytes)

            self.h_buffers[name] = host_mem
            self.d_buffers[name] = dev_mem
            # сохраняем указатель на буфер
            self.bindings.append(int(dev_mem))

Environment

TensorRT Version: tensorrt-10.12.0.36-cp312-none-win_amd64.whl

NVIDIA GPU: RTX 4090

NVIDIA Driver Version: 576.80 Windows

CUDA Version: 12.8

CUDNN Version: 9.10.1.4_cuda12

Operating System: Windows 11 + VS Code

Python Version (if applicable): 3.12

PyTorch Version (if applicable): 2.7.0

Baremetal or Container (if so, version): Baremetal

Screens