Update Docker base image to CUDA 12.4.1 for TensorRT compatibility

[huyyxy]

Is your feature request related to a problem? Please describe.

Currently, the Dockerfile uses nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 which only supports CUDA 11.8. When attempting to use TensorRT with newer GPU hardware/drivers requiring CUDA 12.x, the existing Docker environment fails to provide compatible dependencies, leading to runtime errors and compatibility issues with modern AI frameworks.

Describe the solution you'd like

Update the Docker base image to:

FROM nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04

This change will:

Provide native CUDA 12.4.1 and cuDNN support

Enable seamless TensorRT integration

Align with modern GPU driver requirements (>= 545.x)

Maintain Ubuntu 22.04 LTS compatibility

Describe alternatives you've considered

1. Manually installing CUDA 12.4 in existing container:

Would require complex Dockerfile modifications

Risks version conflicts with base CUDA 11.8 libraries

2. Using CUDA forward-compatibility:

May introduce unexpected behavior

Doesn't resolve underlying dependency mismatches

3. Maintaining separate Dockerfiles:

Increases maintenance overhead

Fragments development/production environments

Additional context

1. Validation Plan:

Verify CUDA version post-build:

nvidia-smi | grep CUDA

Confirm TensorRT availability in Python environment:

import tensorrt
print(tensorrt.__version__)  # Should return >= 10.0.1

2. Potential Impact:

May require updating related dependency versions (e.g., PyTorch, TorchAudio)

Need to validate all existing pipeline functionality with CUDA 12.x

3. Reference Documentation:

NVIDIA CUDA 12.4 Release Notes

TensorRT CUDA Compatibility Matrix