|
| 1 | +FROM lablup/common-base:py38-cuda11.3 |
| 2 | + |
| 3 | +RUN python3 -m pip uninstall -y \ |
| 4 | + plaidml-keras==0.7.0 \ |
| 5 | + tensorflow==2.3.3 \ |
| 6 | + tensorflow-estimator==2.3.0 |
| 7 | + |
| 8 | +RUN python3 -m pip install --no-cache-dir --upgrade \ |
| 9 | + -q git+https://github.com/google-research/tensorflow_constrained_optimization |
| 10 | + |
| 11 | +RUN python3 -m pip install --no-cache-dir \ |
| 12 | + --use-feature=2020-resolver \ |
| 13 | + requests_oauthlib==1.3.1 \ |
| 14 | + neural-structured-learning==1.4.0 \ |
| 15 | + pydot==1.4.2 \ |
| 16 | + proto-plus==1.22.1 |
| 17 | +RUN python3 -m pip install --no-cache-dir \ |
| 18 | + --use-feature=2020-resolver \ |
| 19 | + tensorflow-transform==1.11.0 \ |
| 20 | + tensorflow_model_analysis==0.42.0 \ |
| 21 | + tensorflow-datasets==4.7.0 \ |
| 22 | + tensorflow-datasets-ko==0.2.3 \ |
| 23 | + tensorflow_probability==0.17.0 \ |
| 24 | + tensorflow-addons==0.18.0 \ |
| 25 | + tensorflow-model-optimization==0.7.3 \ |
| 26 | + tfx==1.11.0 \ |
| 27 | + tfx-bsl==1.11.0 \ |
| 28 | + tf-agents==0.14.0 |
| 29 | + |
| 30 | +RUN python3 -m pip uninstall -y google-api-core |
| 31 | +RUN python3 -m pip install --no-cache-dir \ |
| 32 | + --use-feature=2020-resolver \ |
| 33 | + google_cloud_spanner==1.19.3 \ |
| 34 | + googleapis-common-protos==1.56.4 \ |
| 35 | + google_cloud_videointelligence==1.15.0 \ |
| 36 | + google_cloud_vision==0.42.0 \ |
| 37 | + google_cloud_pubsub==2.13.10 \ |
| 38 | + google_cloud_dlp==3.9.2 \ |
| 39 | + google_cloud_datastore==1.15.5 \ |
| 40 | + google_cloud_bigtable==1.7.2 \ |
| 41 | + google-api-core[grpc,grpcgcp]==1.33.2 \ |
| 42 | + google_api_core==1.33.2 |
| 43 | + |
| 44 | +RUN python3 -m pip install --no-cache-dir \ |
| 45 | + mesh-tensorflow==0.1.21 \ |
| 46 | + tensorflow-cloud==0.1.16 \ |
| 47 | + tensorflow-hub==0.12.0 |
| 48 | + |
| 49 | +RUN python3 -m pip install --no-cache-dir \ |
| 50 | + -i https://pypi-nightly.tensorflow.org/simple tensorflow-data-validation==1.11.0 |
| 51 | +RUN python3 -m pip install --no-cache-dir \ |
| 52 | + tensorflow-graphics-gpu==1.0.0 \ |
| 53 | + tensorflow-io==0.28.0 \ |
| 54 | + tensorflow_ranking==0.5.1 \ |
| 55 | + tensorflow-plot==0.3.2 \ |
| 56 | + tensorflow_text==2.11.0 \ |
| 57 | + tensorflow-gan==2.1.0 |
| 58 | +RUN python3 -m pip install --no-cache-dir \ |
| 59 | + tensorflow-gpu==2.11.0 \ |
| 60 | + keras==2.11.0 \ |
| 61 | + Keras-Applications==1.0.8 \ |
| 62 | + Keras-Preprocessing==1.1.2 \ |
| 63 | + tensorboard-plugin-wit \ |
| 64 | + tensorboard numpy scipy |
| 65 | +RUN python3 -m pip install --no-cache-dir \ |
| 66 | + transformers \ |
| 67 | + accelerate \ |
| 68 | + xformers \ |
| 69 | + gradio |
| 70 | + |
| 71 | +WORKDIR /tmp |
| 72 | +RUN rm -f /usr/local/bin/code-server && \ |
| 73 | + rm -fr /usr/local/lib/code-server-* && \ |
| 74 | + curl -fL https://github.com/cdr/code-server/releases/download/v4.8.3/code-server-4.8.3-linux-amd64.tar.gz \ |
| 75 | + | tar -C /usr/local/lib -xz && \ |
| 76 | + mv /usr/local/lib/code-server-4.8.3-linux-amd64 /usr/local/lib/code-server-4.8.3 && \ |
| 77 | + ln -s /usr/local/lib/code-server-4.8.3/bin/code-server /usr/local/bin/code-server |
| 78 | + |
| 79 | +# Install Horovod, temporarily using CUDA stubs |
| 80 | +RUN ldconfig /usr/local/cuda/targets/x86_64-linux/lib/stubs && \ |
| 81 | + HOROVOD_GPU_ALLREDUCE=NCCL HOROVOD_GPU_BROADCAST=NCCL HOROVOD_NCCL_LINK=SHARED \ |
| 82 | + HOROVOD_WITH_TENSORFLOW=1 HOROVOD_WITHOUT_PYTORCH=1 HOROVOD_WITHOUT_MXNET=1\ |
| 83 | + HOROVOD_GPU=CUDA \ |
| 84 | + python3 -m pip install --no-cache-dir horovod==0.26.1 && \ |
| 85 | + ldconfig |
| 86 | + |
| 87 | +RUN python3 -m pip install --no-cache-dir \ |
| 88 | + mpi4py==3.1.4 \ |
| 89 | + nni==2.10 \ |
| 90 | + mlflow==2.0.1 \ |
| 91 | + scikit-nni==0.2.1 |
| 92 | + |
| 93 | +# Install ipython kernelspec |
| 94 | +Run python3 -m ipykernel install --display-name "TensorFlow 2.11 on Python 3.8 & CUDA 11.3" && \ |
| 95 | + cat /usr/local/share/jupyter/kernels/python3/kernel.json |
| 96 | + |
| 97 | +# Copy Backend.Ai multi-node support |
| 98 | +COPY ./service-defs /etc/backend.ai/service-defs |
| 99 | +COPY ./runner-scripts/bootstrap.sh runner-scripts/setup_multinode.py /opt/container/ |
| 100 | + |
| 101 | +# Backend.AI specifics |
| 102 | +LABEL ai.backend.kernelspec="1" \ |
| 103 | + ai.backend.envs.corecount="OPENBLAS_NUM_THREADS,OMP_NUM_THREADS,NPROC" \ |
| 104 | + ai.backend.features="batch query uid-match user-input" \ |
| 105 | + ai.backend.base-distro="ubuntu16.04" \ |
| 106 | + ai.backend.resource.min.cpu="1" \ |
| 107 | + ai.backend.resource.min.mem="1g" \ |
| 108 | + ai.backend.resource.min.cuda.device=0 \ |
| 109 | + ai.backend.resource.min.cuda.shares=0 \ |
| 110 | + ai.backend.runtime-type="python" \ |
| 111 | + ai.backend.runtime-path="/usr/bin/python3" \ |
| 112 | + ai.backend.service-ports="ipython:pty:3000,jupyter:http:8081,jupyterlab:http:8090,vscode:http:8180,tensorboard:http:6006,mlflow-ui:preopen:5000,nniboard:preopen:8080" |
| 113 | + |
| 114 | +WORKDIR /home/work |
0 commit comments