onescience_run_manifest.yaml

onescience_manifest_version: "0.1"
resource_type: dataset
resource:
  id: OneScience/CMEMS
  name: CMEMS
  domain: earth
  domain_tags: [earth, ocean, cmems]
  task: ocean_forecasting
  task_tags: [ocean_forecast, train_data, eval_data]
  modalities: [gridded_field]
  input_formats: [hdf5]
  output_formats: []
  summary: CMEMS 是适配 OneScience/XiHe 的 1993-1999 年全球海洋预报 HDF5 数据集。

platform_resource:
  primary:
    platform: modelscope
    repo_id: OneScience/CMEMS
    repo_type: dataset
    url: https://modelscope.cn/datasets/OneScience/CMEMS
    revision: main
    readme_path: README.md
    manifest_path: manifest.yaml
  mirrors: []
  access:
    visibility: public
    license: null

website_integration:
  enabled: true
  click_target:
    platform: modelscope
    resource_url: https://modelscope.cn/datasets/OneScience/CMEMS
  llm_handoff:
    readme_required: true
    manifest_required: true
    download_readme_first: true
    resolve_related_models: true
    default_run_goal: dataset_validation

runtime:
  enabled: true
  onescience_domain: earth
  min_onescience_version: null
  supported_execution: [local, local_slurm, remote_slurm, scnet_mcp]

onescience:
  repo: https://gitee.com/onescience-ai/onescience
  official_links:
    gitee:
      doc: https://gitee.com/onescience-ai/onescience-doc
      onescience: https://gitee.com/onescience-ai/onescience
      skills: https://gitee.com/onescience-ai/oneskills
    github:
      doc: https://github.com/onescience-ai/OneScience-doc
      onescience: https://github.com/onescience-ai/OneScience
      skills: https://github.com/onescience-ai/oneskills
  install:
    required_by_default: false
    command: bash install.sh earth
  source_paths: []
  compatibility:
    examples_path: null
    status: standard_first

environment:
  install_mode: onescience_domain
  python: "3.11"
  hardware:
    cpu: supported
    gpu: optional
    dcu: optional
  extra_dependencies: [h5py, numpy, pyyaml]

runtime_package:
  kind: dataset_package
  package_root: .
  standard_layout:
    workdir: .
    config_dir: metadata
    checkpoint_dir: null
    data_dir: data
    output_dir: null
  apply_policy:
    mode: direct_use
    target: session_workdir
    overwrite: true
    protect_installed_onescience: true
  entry_files:
    - metadata/schema.yaml
    - metadata/file_checksums.sha256
    - scripts/validate_dataset.py
  entrypoints:
    preflight: scripts/validate_dataset.py
    inference: null
    train: null
    finetune: null
    evaluate: null
    visualize: null
    deploy: null

files:
  model_files: []
  dataset_files:
    - id: cmems_1993_h5
      path: data/1993.h5
      role: train_data
      description_zh: 1993 年 CMEMS HDF5 年份文件，包含 fields、global_means 和 global_stds。
      required: true
      required_for: [preflight, train]
      format: hdf5
      source:
        platform: modelscope
        repo_id: OneScience/CMEMS
        repo_type: dataset
        path: data/1993.h5
        revision: main
        url: null
        download_method: command_ref
        command_ref: commands.download.download_dataset
      mirrors: []
      local_path: CMEMS/data/1993.h5
      size: 10157335296
      checksum: sha256:57c694c0faf46cb5762751fffca473aeba783fb029288c2c1dea62c50a4b16e8
      large_file: true
    - id: cmems_1994_h5
      path: data/1994.h5
      role: train_data
      description_zh: 1994 年 CMEMS HDF5 年份文件。
      required: true
      required_for: [preflight, train]
      format: hdf5
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: data/1994.h5, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/data/1994.h5
      size: 10157335296
      checksum: sha256:d31d57c292518b799f2dafac6e57663ee5bdb2128c5f7286ae9b4afdc4a51859
      large_file: true
    - id: cmems_1995_h5
      path: data/1995.h5
      role: train_data
      description_zh: 1995 年 CMEMS HDF5 年份文件。
      required: true
      required_for: [preflight, train]
      format: hdf5
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: data/1995.h5, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/data/1995.h5
      size: 10157335296
      checksum: sha256:f86f6a02ebaa852b7ed88ceafc0dd4d4def09ec53269d1aba71031bd3ac0f7fe
      large_file: true
    - id: cmems_1996_h5
      path: data/1996.h5
      role: train_data
      description_zh: 1996 年 CMEMS HDF5 年份文件。
      required: true
      required_for: [preflight, train]
      format: hdf5
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: data/1996.h5, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/data/1996.h5
      size: 10157335296
      checksum: sha256:ffe95335363b97d97744b310af22f1fc9b79a77446bf0c8981dd27e77e3b0098
      large_file: true
    - id: cmems_1997_h5
      path: data/1997.h5
      role: train_data
      description_zh: 1997 年 CMEMS HDF5 年份文件。
      required: true
      required_for: [preflight, train]
      format: hdf5
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: data/1997.h5, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/data/1997.h5
      size: 10157335296
      checksum: sha256:2d9c427a4fe6e9ee621edff3a641c088e506b7c5ee0d9ab0928d98d2aacfb5fc
      large_file: true
    - id: cmems_1998_h5
      path: data/1998.h5
      role: eval_data
      description_zh: 1998 年 CMEMS HDF5 年份文件，XiHe 标准配置用于验证。
      required: true
      required_for: [preflight, train, evaluate]
      format: hdf5
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: data/1998.h5, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/data/1998.h5
      size: 10157335296
      checksum: sha256:c25a62f81c720663fddec49a081568e7a5a46e767204e67ab9945e7c121d6e43
      large_file: true
    - id: cmems_1999_h5
      path: data/1999.h5
      role: inference_input
      description_zh: 1999 年 CMEMS HDF5 年份文件，XiHe 标准配置用于测试、推理和评测。
      required: true
      required_for: [preflight, inference, evaluate]
      format: hdf5
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: data/1999.h5, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/data/1999.h5
      size: 10157335296
      checksum: sha256:a3a8040afb56ad2bee79d19fb8bd6b44d47411c8417ada95b063ef91cfbb08de
      large_file: true
    - id: cmems_schema
      path: metadata/schema.yaml
      role: schema
      description_zh: HDF5 结构、shape、dtype、attrs、变量列表和划分说明。
      required: true
      required_for: [preflight]
      format: yaml
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: metadata/schema.yaml, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/metadata/schema.yaml
      size: null
      checksum: null
      large_file: false
    - id: cmems_checksums
      path: metadata/file_checksums.sha256
      role: checksum
      description_zh: 数据文件 SHA256 校验清单。
      required: true
      required_for: [preflight]
      format: text
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: metadata/file_checksums.sha256, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/metadata/file_checksums.sha256
      size: null
      checksum: null
      large_file: false
  config_files:
    - id: validate_dataset_script
      path: scripts/validate_dataset.py
      role: preflight_script
      description_zh: 数据集读取、schema、attrs、shape、dtype 和可选 SHA256 校验脚本。
      required: true
      required_for: [preflight]
      format: python
      source: {platform: modelscope, repo_id: OneScience/CMEMS, repo_type: dataset, path: scripts/validate_dataset.py, revision: main, url: null, download_method: command_ref, command_ref: commands.download.download_dataset}
      mirrors: []
      local_path: CMEMS/scripts/validate_dataset.py
      size: null
      checksum: null
      large_file: false
  sample_files: []

dataset:
  format: [hdf5]
  sample_unit: one daily ocean state transition sample per time index
  schema:
    path: metadata/schema.yaml
  checksums:
    path: metadata/file_checksums.sha256
    algorithm: sha256
  splits:
    train: [1993, 1994, 1995, 1996, 1997]
    validation: [1998]
    test: [1999]
  structure:
    hdf5_datasets: [fields, global_means, global_stds]
    fields_shape: [3, 96, 2041, 4320]
    stats_shape: [1, 96, 1, 1]
    dtype: float32
    time_step_hours: 24

relations:
  required_datasets: []
  optional_datasets: []
  compatible_models:
    - id: OneScience/XiHe
      role: train_data
      required_for: [preflight, train, inference, evaluate, visualize]
      resource_ref:
        platform: modelscope
        repo_id: OneScience/XiHe
        repo_type: model
        url: https://modelscope.cn/models/OneScience/XiHe
        revision: main
        readme_path: README.md
        manifest_path: manifest.yaml
      expected_local_path: XiHe/data/data
      adapter: XiHe/scripts/prepare_runtime_data.py

run_matrix:
  scenarios:
    - name: dataset_validation
      capability: preflight
      default: true
      description_zh: 验证 CMEMS 数据集目录结构、HDF5 schema、shape、dtype、attrs 和统计量。
      required_datasets:
        - id: OneScience/CMEMS
          role: self
          local_path: .
      required_model_files: []
      required_data_files: [data/1993.h5, data/1994.h5, data/1995.h5, data/1996.h5, data/1997.h5, data/1998.h5, data/1999.h5]
      preconditions:
        - cwd 为下载后的 CMEMS 数据集仓库根目录。
      command_refs:
        - commands.preflight.validate_structure
      outputs: []
    - name: dataset_checksum_validation
      capability: preflight
      description_zh: 在结构验证基础上读取全部 7 个大文件并校验 SHA256，耗时较长。
      required_datasets:
        - id: OneScience/CMEMS
          role: self
          local_path: .
      required_model_files: []
      required_data_files: [data/*.h5, metadata/file_checksums.sha256]
      preconditions:
        - cwd 为下载后的 CMEMS 数据集仓库根目录。
      command_refs:
        - commands.preflight.validate_with_checksums
      outputs: []
    - name: prepare_for_xihe
      capability: prepare
      description_zh: 与 OneScience/XiHe 模型配合时，由模型包脚本将本数据集整理到 XiHe/data/data。
      required_datasets:
        - id: OneScience/CMEMS
          role: train_data
          local_path: CMEMS
      required_model_files: [XiHe/scripts/prepare_runtime_data.py, XiHe/conf/config.yaml]
      required_data_files: [CMEMS/data/*.h5]
      preconditions:
        - OneScience/XiHe 下载到 ./XiHe。
        - OneScience/CMEMS 下载到 ./CMEMS。
      command_refs:
        - commands.prepare.prepare_for_xihe
      outputs:
        - XiHe/data/data/*.h5
        - XiHe/data/stats/global_means.npy
        - XiHe/data/land_mask.npy

capabilities:
  inference: false
  train: false
  finetune: false
  evaluate: false
  visualize: false
  deploy: false

commands:
  download:
    - name: download_dataset
      description: 下载 CMEMS 数据集标准仓库。
      target: dataset
      repo_id: OneScience/CMEMS
      source: .
      local_path: ./CMEMS
      required: true
      required_for: [preflight, train, inference, evaluate]
      command: modelscope download --dataset OneScience/CMEMS --local_dir ./CMEMS
    - name: download_compatible_model
      description: 下载适配模型 XiHe。
      target: model
      repo_id: OneScience/XiHe
      source: .
      local_path: ./XiHe
      required: false
      required_for: [train, inference, evaluate]
      command: modelscope download --model OneScience/XiHe --local_dir ./XiHe
  prepare:
    - name: prepare_for_xihe
      description: 调用 XiHe 模型包脚本，把 CMEMS 数据组织成 XiHe 运行布局。
      cwd: ./XiHe
      command: python scripts/prepare_runtime_data.py --dataset-root ../CMEMS --runtime-data-dir ./data
      required_files: [../CMEMS/data/1993.h5, scripts/prepare_runtime_data.py]
      required_env: [python, h5py, numpy]
      expected_outputs:
        - path: data/data/1993.h5
          type: file
      success_criteria: [exit_code == 0, output_path_exists]
  preflight:
    - name: validate_structure
      description: 快速读取每个 HDF5 文件的结构、变量、shape、dtype 和统计量。
      cwd: .
      command: python scripts/validate_dataset.py --dataset-root .
      required_files: [metadata/schema.yaml, metadata/file_checksums.sha256, data/1993.h5, data/1999.h5]
      required_env: [python, h5py, numpy, pyyaml]
      expected_outputs: []
      success_criteria: [exit_code == 0]
    - name: validate_with_checksums
      description: 完整读取所有 HDF5 文件并校验 SHA256。
      cwd: .
      command: python scripts/validate_dataset.py --dataset-root . --verify-checksums
      required_files: [metadata/file_checksums.sha256, data/1993.h5, data/1999.h5]
      required_env: [python, h5py, numpy, pyyaml]
      expected_outputs: []
      success_criteria: [exit_code == 0]
  inference: []
  train: []
  finetune: []
  evaluate: []
  visualize: []
  deploy: []

expected_outputs:
  - path: stdout
    type: text
    produced_by: commands.preflight.validate_structure
    contains: CMEMS dataset validation passed.
  - path: XiHe/data/stats/global_means.npy
    type: file
    produced_by: commands.prepare.prepare_for_xihe

diagnostics:
  - name: missing_hdf5
    match: "missing file"
    explanation_zh: 数据集未完整下载或当前工作目录不是 CMEMS 仓库根目录。
    fix_zh: 执行 modelscope download --dataset OneScience/CMEMS --local_dir ./CMEMS，并在 ./CMEMS 中运行验证命令。
  - name: schema_mismatch
    match: "variables attr does not match schema"
    explanation_zh: HDF5 文件变量顺序或变量名与 schema.yaml 不一致。
    fix_zh: 重新校验上传文件是否来自同一批 CMEMS 数据，必要时重新生成 schema 和 checksum。
  - name: checksum_mismatch
    match: "sha256 mismatch"
    explanation_zh: 数据文件内容与上传前校验清单不一致。
    fix_zh: 重新下载或重新上传对应年份文件。

domain_extension:
  earth:
    subdomain: ocean
    source_family: CMEMS
    grid:
      lat_size: 2041
      lon_size: 4320
      spatial_resolution: [0.0833333333, 0.0833333333]
    time:
      years: [1993, 1994, 1995, 1996, 1997, 1998, 1999]
      time_step_hours: 24
      samples_per_year_file: 3
    variables_count: 96