Add CI stack for ML packages (#31592)

Basic stack of ML packages we would like to test and generate binaries for in CI. 

Spack now has a large CI framework in GitLab for PR testing and public binary generation.
We should take advantage of this to test and distribute optimized binaries for popular ML
frameworks.

This is a pretty extensive initial set, including CPU, ROCm, and CUDA versions of a core
`x96_64_v4` stack.

### Core ML frameworks

These are all popular core ML frameworks already available in Spack.

- [x] PyTorch
- [x] TensorFlow
- [x] Scikit-learn
- [x] MXNet
- [x] CNTK
- [x] Caffe
- [x] Chainer
- [x] XGBoost
- [x] Theano

### ML extensions

These are domain libraries and wrappers that build on top of core ML libraries

- [x] Keras
- [x] TensorBoard
- [x] torchvision
- [x] torchtext
- [x] torchaudio
- [x] TorchGeo
- [x] PyTorch Lightning
- [x] torchmetrics
- [x] GPyTorch
- [x] Horovod

### ML-adjacent libraries

These are libraries that aren't specific to ML but are still core libraries used in ML pipelines

- [x] numpy
- [x] scipy
- [x] pandas
- [x] ONNX
- [x] bazel

Co-authored-by: Jonathon Anderson <17242663+blue42u@users.noreply.github.com>

share/spack/gitlab/cloud_pipelines/.gitlab-ci.yml

@@ -749,3 +749,123 @@ tutorial-protected-build:
   needs:
     - artifacts: True
       job: tutorial-protected-generate
+
+########################################
+# Machine Learning (CPU)
+########################################
+.ml-cpu:
+  variables:
+    SPACK_CI_STACK_NAME: ml-cpu
+
+.ml-cpu-generate:
+  extends: .ml-cpu
+  image: ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21
+  tags: ["spack", "aws", "public", "medium", "x86_64_v4"]
+
+ml-cpu-pr-generate:
+  extends: [ ".ml-cpu-generate", ".pr-generate"]
+
+ml-cpu-protected-generate:
+  extends: [ ".ml-cpu-generate", ".protected-generate"]
+
+ml-cpu-pr-build:
+  extends: [ ".ml-cpu", ".pr-build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-cpu-pr-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-cpu-pr-generate
+
+ml-cpu-protected-build:
+  extends: [ ".ml-cpu", ".protected-build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-cpu-protected-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-cpu-protected-generate
+
+########################################
+# Machine Learning (CUDA)
+########################################
+.ml-cuda:
+  variables:
+    SPACK_CI_STACK_NAME: ml-cuda
+
+.ml-cuda-generate:
+  extends: .ml-cuda
+  image: ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21
+  tags: ["spack", "aws", "public", "medium", "x86_64_v4"]
+
+ml-cuda-pr-generate:
+  extends: [ ".ml-cuda-generate", ".pr-generate"]
+
+ml-cuda-protected-generate:
+  extends: [ ".ml-cuda-generate", ".protected-generate"]
+
+ml-cuda-pr-build:
+  extends: [ ".ml-cuda", ".pr-build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-cuda-pr-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-cuda-pr-generate
+
+ml-cuda-protected-build:
+  extends: [ ".ml-cuda", ".protected-build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-cuda-protected-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-cuda-protected-generate
+
+########################################
+# Machine Learning (ROCm)
+########################################
+.ml-rocm:
+  variables:
+    SPACK_CI_STACK_NAME: ml-rocm
+
+.ml-rocm-generate:
+  extends: .ml-rocm
+  image: ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21
+  tags: ["spack", "aws", "public", "medium", "x86_64_v4"]
+
+ml-rocm-pr-generate:
+  extends: [ ".ml-rocm-generate", ".pr-generate"]
+
+ml-rocm-protected-generate:
+  extends: [ ".ml-rocm-generate", ".protected-generate"]
+
+ml-rocm-pr-build:
+  extends: [ ".ml-rocm", ".pr-build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-rocm-pr-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-rocm-pr-generate
+
+ml-rocm-protected-build:
+  extends: [ ".ml-rocm", ".protected-build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-rocm-protected-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-rocm-protected-generate

share/spack/gitlab/cloud_pipelines/stacks/ml-cpu/spack.yaml

@@ -0,0 +1,142 @@
+spack:
+  view: false
+
+  concretizer:
+    reuse: false
+    unify: false
+
+  config:
+    concretizer: clingo
+    install_tree:
+      root: /home/software/spack
+      padded_length: 384
+      projections:
+        all: "{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}"
+
+  packages:
+    all:
+      compiler: [gcc@11.2.0]
+      target: [x86_64_v4]
+      variants: ~cuda~rocm
+
+  specs:
+    # Horovod
+    - py-horovod
+
+    # JAX
+    # https://github.com/google/jax/issues/12614
+    # - py-jax
+    # - py-jaxlib
+
+    # Keras
+    - py-keras
+    - py-keras-applications
+    - py-keras-preprocessing
+    - py-keras2onnx
+
+    # PyTorch
+    - py-botorch
+    - py-efficientnet-pytorch
+    - py-gpytorch
+    - py-kornia
+    - py-pytorch-gradual-warmup-lr
+    - py-pytorch-lightning
+    - py-segmentation-models-pytorch
+    - py-timm
+    - py-torch
+    - py-torch-cluster
+    - py-torch-geometric
+    # https://github.com/NVIDIA/apex/issues/1498
+    # - py-torch-nvidia-apex
+    - py-torch-scatter
+    - py-torch-sparse
+    - py-torch-spline-conv
+    - py-torchaudio
+    - py-torchdata
+    - py-torchfile
+    - py-torchgeo
+    - py-torchmeta
+    - py-torchmetrics
+    - py-torchtext
+    - py-torchvision
+    - py-vector-quantize-pytorch
+
+    # scikit-learn
+    - py-scikit-learn
+    - py-scikit-learn-extra
+
+    # TensorBoard
+    - py-tensorboard
+    - py-tensorboard-data-server
+    - py-tensorboard-plugin-wit
+    - py-tensorboardx
+
+    # TensorFlow
+    - py-tensorflow
+    - py-tensorflow-datasets
+    - py-tensorflow-estimator
+    - py-tensorflow-hub
+    - py-tensorflow-metadata
+    - py-tensorflow-probability
+
+    # XGBoost
+    - py-xgboost
+    # - r-xgboost
+    - xgboost
+
+  mirrors: { "mirror": "s3://spack-binaries/develop/ml-cpu" }
+
+  gitlab-ci:
+    script:
+      - . "./share/spack/setup-env.sh"
+      - spack --version
+      - cd ${SPACK_CONCRETE_ENV_DIR}
+      - spack env activate --without-view .
+      - spack config add "config:install_tree:projections:${SPACK_JOB_SPEC_PKG_NAME}:'morepadding/{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}'"
+      - mkdir -p ${SPACK_ARTIFACTS_ROOT}/user_data
+      - if [[ -r /mnt/key/intermediate_ci_signing_key.gpg ]]; then spack gpg trust /mnt/key/intermediate_ci_signing_key.gpg; fi
+      - if [[ -r /mnt/key/spack_public_key.gpg ]]; then spack gpg trust /mnt/key/spack_public_key.gpg; fi
+      - spack -d ci rebuild > >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_out.txt) 2> >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_err.txt >&2)
+
+    mappings:
+      - match:
+          - llvm
+        runner-attributes:
+          tags: [ "spack", "huge", "x86_64_v4" ]
+          variables:
+            CI_JOB_SIZE: huge
+            KUBERNETES_CPU_REQUEST: 11000m
+            KUBERNETES_MEMORY_REQUEST: 42G
+      - match:
+          - "@:"
+        runner-attributes:
+          tags: [ "spack", "large", "x86_64_v4" ]
+          variables:
+            CI_JOB_SIZE: large
+            KUBERNETES_CPU_REQUEST: 8000m
+            KUBERNETES_MEMORY_REQUEST: 12G
+
+    image: { "name": "ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21", "entrypoint": [""] }
+
+    broken-specs-url: "s3://spack-binaries/broken-specs"
+
+    service-job-attributes:
+      before_script:
+        - . "./share/spack/setup-env.sh"
+        - spack --version
+      image: { "name": "ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21", "entrypoint": [""] }
+      tags: ["spack", "public", "x86_64_v4"]
+
+    signing-job-attributes:
+      image: { "name": "ghcr.io/spack/notary:latest", "entrypoint": [""] }
+      tags: ["spack", "aws"]
+      script:
+        - aws s3 sync --exclude "*" --include "*spec.json*" ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache /tmp
+        - /sign.sh
+        - aws s3 sync --exclude "*" --include "*spec.json.sig*" /tmp ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache
+
+  cdash:
+    build-group: Machine Learning
+    url: https://cdash.spack.io
+    project: Spack Testing
+    site: Cloud Gitlab Infrastructure

share/spack/gitlab/cloud_pipelines/stacks/ml-cuda/spack.yaml

@@ -0,0 +1,144 @@
+spack:
+  view: false
+
+  concretizer:
+    reuse: false
+    unify: false
+
+  config:
+    concretizer: clingo
+    install_tree:
+      root: /home/software/spack
+      padded_length: 384
+      projections:
+        all: "{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}"
+
+  packages:
+    all:
+      compiler: [gcc@11.2.0]
+      target: [x86_64_v4]
+      variants: ~rocm+cuda cuda_arch=80
+    llvm:
+      # https://github.com/spack/spack/issues/27999
+      require: ~cuda
+
+  specs:
+    # Horovod
+    - py-horovod
+
+    # JAX
+    # https://github.com/google/jax/issues/12614
+    # - py-jax
+    # - py-jaxlib
+
+    # Keras
+    - py-keras
+    - py-keras-applications
+    - py-keras-preprocessing
+    - py-keras2onnx
+
+    # PyTorch
+    - py-botorch
+    - py-efficientnet-pytorch
+    - py-gpytorch
+    - py-kornia
+    - py-pytorch-gradual-warmup-lr
+    - py-pytorch-lightning
+    - py-segmentation-models-pytorch
+    - py-timm
+    - py-torch
+    - py-torch-cluster
+    - py-torch-geometric
+    - py-torch-nvidia-apex
+    - py-torch-scatter
+    - py-torch-sparse
+    - py-torch-spline-conv
+    - py-torchaudio
+    - py-torchdata
+    - py-torchfile
+    - py-torchgeo
+    - py-torchmeta
+    - py-torchmetrics
+    - py-torchtext
+    - py-torchvision
+    - py-vector-quantize-pytorch
+
+    # scikit-learn
+    - py-scikit-learn
+    - py-scikit-learn-extra
+
+    # TensorBoard
+    - py-tensorboard
+    - py-tensorboard-data-server
+    - py-tensorboard-plugin-wit
+    - py-tensorboardx
+
+    # TensorFlow
+    - py-tensorflow
+    - py-tensorflow-datasets
+    - py-tensorflow-estimator
+    - py-tensorflow-hub
+    - py-tensorflow-metadata
+    - py-tensorflow-probability
+
+    # XGBoost
+    - py-xgboost
+    # - r-xgboost
+    - xgboost
+
+  mirrors: { "mirror": "s3://spack-binaries/develop/ml-cuda" }
+
+  gitlab-ci:
+    script:
+      - . "./share/spack/setup-env.sh"
+      - spack --version
+      - cd ${SPACK_CONCRETE_ENV_DIR}
+      - spack env activate --without-view .
+      - spack config add "config:install_tree:projections:${SPACK_JOB_SPEC_PKG_NAME}:'morepadding/{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}'"
+      - mkdir -p ${SPACK_ARTIFACTS_ROOT}/user_data
+      - if [[ -r /mnt/key/intermediate_ci_signing_key.gpg ]]; then spack gpg trust /mnt/key/intermediate_ci_signing_key.gpg; fi
+      - if [[ -r /mnt/key/spack_public_key.gpg ]]; then spack gpg trust /mnt/key/spack_public_key.gpg; fi
+      - spack -d ci rebuild > >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_out.txt) 2> >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_err.txt >&2)
+
+    mappings:
+      - match:
+          - llvm
+        runner-attributes:
+          tags: [ "spack", "huge", "x86_64_v4" ]
+          variables:
+            CI_JOB_SIZE: huge
+            KUBERNETES_CPU_REQUEST: 11000m
+            KUBERNETES_MEMORY_REQUEST: 42G
+      - match:
+          - "@:"
+        runner-attributes:
+          tags: [ "spack", "large", "x86_64_v4" ]
+          variables:
+            CI_JOB_SIZE: large
+            KUBERNETES_CPU_REQUEST: 8000m
+            KUBERNETES_MEMORY_REQUEST: 12G
+
+    image: { "name": "ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21", "entrypoint": [""] }
+
+    broken-specs-url: "s3://spack-binaries/broken-specs"
+
+    service-job-attributes:
+      before_script:
+        - . "./share/spack/setup-env.sh"
+        - spack --version
+      image: { "name": "ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21", "entrypoint": [""] }
+      tags: ["spack", "public", "x86_64_v4"]
+
+    signing-job-attributes:
+      image: { "name": "ghcr.io/spack/notary:latest", "entrypoint": [""] }
+      tags: ["spack", "aws"]
+      script:
+        - aws s3 sync --exclude "*" --include "*spec.json*" ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache /tmp
+        - /sign.sh
+        - aws s3 sync --exclude "*" --include "*spec.json.sig*" /tmp ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache
+
+  cdash:
+    build-group: Machine Learning
+    url: https://cdash.spack.io
+    project: Spack Testing
+    site: Cloud Gitlab Infrastructure

share/spack/gitlab/cloud_pipelines/stacks/ml-rocm/spack.yaml

@@ -0,0 +1,147 @@
+spack:
+  view: false
+
+  concretizer:
+    reuse: false
+    unify: false
+
+  config:
+    concretizer: clingo
+    install_tree:
+      root: /home/software/spack
+      padded_length: 384
+      projections:
+        all: "{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}"
+
+  packages:
+    all:
+      compiler: [gcc@11.2.0]
+      target: [x86_64_v4]
+      variants: ~cuda+rocm amdgpu_target=gfx90a
+    gl:
+      require: "osmesa"
+    py-torch:
+      # Does not yet support Spack-installed ROCm
+      require: ~rocm
+
+  specs:
+    # Horovod
+    - py-horovod
+
+    # JAX
+    # https://github.com/google/jax/issues/12614
+    # - py-jax
+    # - py-jaxlib
+
+    # Keras
+    - py-keras
+    - py-keras-applications
+    - py-keras-preprocessing
+    - py-keras2onnx
+
+    # PyTorch
+    # Does not yet support Spack-install ROCm
+    # - py-botorch
+    # - py-efficientnet-pytorch
+    # - py-gpytorch
+    # - py-kornia
+    # - py-pytorch-gradual-warmup-lr
+    # - py-pytorch-lightning
+    # - py-segmentation-models-pytorch
+    # - py-timm
+    # - py-torch
+    # - py-torch-cluster
+    # - py-torch-geometric
+    # - py-torch-nvidia-apex
+    # - py-torch-scatter
+    # - py-torch-sparse
+    # - py-torch-spline-conv
+    # - py-torchaudio
+    # - py-torchdata
+    # - py-torchfile
+    # - py-torchgeo
+    # - py-torchmeta
+    # - py-torchmetrics
+    # - py-torchtext
+    # - py-torchvision
+    # - py-vector-quantize-pytorch
+
+    # scikit-learn
+    - py-scikit-learn
+    - py-scikit-learn-extra
+
+    # TensorBoard
+    - py-tensorboard
+    - py-tensorboard-data-server
+    - py-tensorboard-plugin-wit
+    - py-tensorboardx
+
+    # TensorFlow
+    - py-tensorflow
+    - py-tensorflow-datasets
+    - py-tensorflow-estimator
+    - py-tensorflow-hub
+    - py-tensorflow-metadata
+    - py-tensorflow-probability
+
+    # XGBoost
+    - py-xgboost
+    # - r-xgboost
+    - xgboost
+
+  mirrors: { "mirror": "s3://spack-binaries/develop/ml-rocm" }
+
+  gitlab-ci:
+    script:
+      - . "./share/spack/setup-env.sh"
+      - spack --version
+      - cd ${SPACK_CONCRETE_ENV_DIR}
+      - spack env activate --without-view .
+      - spack config add "config:install_tree:projections:${SPACK_JOB_SPEC_PKG_NAME}:'morepadding/{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}'"
+      - mkdir -p ${SPACK_ARTIFACTS_ROOT}/user_data
+      - if [[ -r /mnt/key/intermediate_ci_signing_key.gpg ]]; then spack gpg trust /mnt/key/intermediate_ci_signing_key.gpg; fi
+      - if [[ -r /mnt/key/spack_public_key.gpg ]]; then spack gpg trust /mnt/key/spack_public_key.gpg; fi
+      - spack -d ci rebuild > >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_out.txt) 2> >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_err.txt >&2)
+
+    mappings:
+      - match:
+          - llvm
+        runner-attributes:
+          tags: [ "spack", "huge", "x86_64_v4" ]
+          variables:
+            CI_JOB_SIZE: huge
+            KUBERNETES_CPU_REQUEST: 11000m
+            KUBERNETES_MEMORY_REQUEST: 42G
+      - match:
+          - "@:"
+        runner-attributes:
+          tags: [ "spack", "large", "x86_64_v4" ]
+          variables:
+            CI_JOB_SIZE: large
+            KUBERNETES_CPU_REQUEST: 8000m
+            KUBERNETES_MEMORY_REQUEST: 12G
+
+    image: { "name": "ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21", "entrypoint": [""] }
+
+    broken-specs-url: "s3://spack-binaries/broken-specs"
+
+    service-job-attributes:
+      before_script:
+        - . "./share/spack/setup-env.sh"
+        - spack --version
+      image: { "name": "ghcr.io/spack/e4s-amazonlinux-2:v2022-03-21", "entrypoint": [""] }
+      tags: ["spack", "public", "x86_64_v4"]
+
+    signing-job-attributes:
+      image: { "name": "ghcr.io/spack/notary:latest", "entrypoint": [""] }
+      tags: ["spack", "aws"]
+      script:
+        - aws s3 sync --exclude "*" --include "*spec.json*" ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache /tmp
+        - /sign.sh
+        - aws s3 sync --exclude "*" --include "*spec.json.sig*" /tmp ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache
+
+  cdash:
+    build-group: Machine Learning
+    url: https://cdash.spack.io
+    project: Spack Testing
+    site: Cloud Gitlab Infrastructure