TensorRT-Model-Connect/Dockerfile.dev.x86-gpu at main · NVIDIA/TensorRT-Model-Connect · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

# x86_64 GPU development image for the community GPU CI pipeline (L4 / L40 / L40S,
# Ada Lovelace, compute capability 8.9). The internal root Dockerfile targets
# GB300 (aarch64, compute capability 10.0) only and cannot build on x86_64 hosts;
# this file is the x86_64 counterpart used for CUDA-enabled testing on
# externally reserved (Brev) GPU instances.
ARG TENSORRT_IMAGE=nvcr.io/nvidia/tensorrt:26.07-py3@sha256:b82db1abc23750ab0069abc99bbe4ea29138dbdc23ea39861199e2346638b48a
FROM ${TENSORRT_IMAGE}

ARG TORCH_CUDA_INDEX=https://download.pytorch.org/whl/cu130
ARG TORCH_VERSION=2.12.0+cu130
ARG TORCHVISION_VERSION=0.27.0+cu130
ARG TORCHAUDIO_VERSION=2.11.0+cu130

ENV DEBIAN_FRONTEND=noninteractive
ENV PIP_NO_CACHE_DIR=1
ENV TORCH_CUDA_ARCH_LIST=8.9
ENV NVIDIA_VISIBLE_DEVICES=all
ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility

RUN apt-get update && \
    apt-get install -y --no-install-recommends \
      build-essential \
      cmake \
      git \
      ninja-build \
      nlohmann-json3-dev \
      patchelf \
      pkg-config \
      python3.12-dev \
      python3.12-venv && \
    rm -rf /var/lib/apt/lists/*

ENV VIRTUAL_ENV=/opt/venv
RUN python3.12 -m venv --system-site-packages ${VIRTUAL_ENV} && \
    ln -s /usr/bin/cmake ${VIRTUAL_ENV}/bin/cmake
ENV PATH="${VIRTUAL_ENV}/bin:${PATH}"

COPY community-ci.txt /tmp/trtmc-community-ci.txt

RUN pip install -U pip && \
    pip install \
      "cuda-python>=13.0.3,<14" \
      "huggingface_hub>=0.23" \
      "ml_dtypes>=0.4" \
      "numpy>=1.24,<2.5" \
      "onnx>=1.16" \
      "PyYAML>=6.0" \
      "safetensors>=0.4" \
      "sentencepiece>=0.1.99" \
      "transformers==5.2.0" && \
    pip install \
      "torch==${TORCH_VERSION}" \
      "torchvision==${TORCHVISION_VERSION}" \
      "torchaudio==${TORCHAUDIO_VERSION}" \
      --index-url "${TORCH_CUDA_INDEX}" && \
    pip install "setuptools>=80,<82" && \
    pip install --requirement /tmp/trtmc-community-ci.txt && \
    rm -f /tmp/trtmc-community-ci.txt

ENV TRT_LIB_DIR=/usr/lib/x86_64-linux-gnu
ENV TRT_INC_DIR=/usr/include/x86_64-linux-gnu
ENV LD_LIBRARY_PATH="${TRT_LIB_DIR}:/usr/local/cuda/lib64"

RUN cmake --version && \
    python3.12 -c \
      "import importlib.metadata as metadata; import onnx, tensorrt, torch, tvm_ffi; assert tensorrt.__version__ == '11.1.0.106'; assert hasattr(tensorrt, 'CausalMaskKind'); assert torch.__version__ == '${TORCH_VERSION}'; assert torch.version.cuda is not None; assert metadata.version('apache-tvm-ffi') == '0.1.12'" && \
    test -f "${TRT_INC_DIR}/NvInferVersion.h" && \
    test -f "${TRT_LIB_DIR}/libnvinfer.so.11"

WORKDIR /workspace/tensorrt-model-connect

CMD ["bash"]