Pylance errors

Connected

https://github.com/microsoft/pylance-release/issues/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

Start with the connected Pylance issue 7375, then reproduce the provided Python snippet in VS Code using the listed Python, PyTorch, TensorRT, CUDA, and Windows environment. The issue provides no repository file or test; completion would require an actionable diagnosis of why Pylance misses the functions and classes, or confirmation that the connected report addresses it.