particleflow_spec.yaml

# ParticleFlow Project Unified Configuration Specification
# This file serves as the single source of truth for dataset generation,
# processing pipeline, and model definitions.

project:
  name: "particleflow"
  description: "Machine Learning for Particle Flow Reconstruction"

  # Site-specific configurations (Anchors)
  # To switch sites, update the merge key (<<: *lxplus
  sites:
    #Tallinn has been tested extensively
    tallinn: &tallinn
      site_name: "tallinn"
      executor: "slurm"
      slurm_account: "hepusers"
      container: "/scratch/persistent/joosep/singularity/pytorch-20260305-08d6950.sif"
      bind_mounts:
        - "/local"
        - "/cvmfs"
        - "/scratch/local"
        - "/scratch/persistent"
      paths:
        storage_root: "/local/joosep/mlpf"
        scratch_root: "/scratch/local/joosep"
        project_root: "/home/joosep/particleflow"
      cmssw_dir: "/scratch/persistent/joosep/CMSSW_15_0_5"
      gpu_partition: "gpu"
      cpu_partition: "main"

    #lxplus has been tested to some extent, but not fully due to htcondor being quite overloaded
    lxplus: &lxplus
      site_name: "lxplus"
      executor: "condor"
      container: "/eos/user/j/jpata/www/pytorch-20260305-08d6950.sif"
      bind_mounts:
        - "/afs"
        - "/cvmfs"
        - "/eos/user/j/jpata"
        - "/etc/pki/ca-trust"
        - "/etc/tnsnames.ora"
        - "/run"
      paths:
        storage_root: "/eos/user/j/jpata/mlpf"
        scratch_root: "/tmp"
        project_root: "/afs/cern.ch/work/j/jpata/particleflow"
      cmssw_dir: "/cvmfs/cms.cern.ch"
      gpu_partition: "tomorrow"
      cpu_partition: "tomorrow"

    #Running on a local machine is possible, change the paths below as needed
    local: &local
      site_name: "local"
      executor: "local"
      container: "https://jpata.web.cern.ch/jpata/pytorch-20260305-08d6950.sif"
      bind_mounts:
        - "/tmp"
        - "/mnt/work"
      paths:
        storage_root: "/mnt/work/mlpf/"
        scratch_root: "/tmp/particleflow"
        project_root: "/home/joosep/particleflow"
      cmssw_dir: "/home/joosep/particleflow/CMSSW_15_0_5"
      gpu_partition: "local"
      cpu_partition: "local"

  # Selected site defaults (Merged into 'project'), choose 'tallinn' or 'lxplus'
  <<: *local

# -----------------------------------------------------------------------------
# Dataset Productions
# Defines how raw data is generated (or found), processed, and registered in TFDS.
# Each production follows a unified directory structure:
#   $workspace_dir/gen  - Raw ROOT/EDM files from simulation
#   $workspace_dir/post - Postprocessed intermediate files (parquet/pkl)
#   $workspace_dir/tfds - Final TensorFlow Datasets
# -----------------------------------------------------------------------------
productions:
  # CMS Production Campaign
  cms_run3:
    type: "cms"
    model: "pyg-cms-v1"
    comment: "Main production with CMSSW_15_0_5"
    workspace_dir: "${project.paths.storage_root}/cms/20260204_cmssw_15_0_5_117d32"
    config_dir: "${project.paths.project_root}"
    gen_container: "/cvmfs/singularity.opensciencegrid.org/cmssw/cms:rhel8-x86_64"

    # Common execution environment for this campaign
    environment:
      cmssw_release: "CMSSW_15_0_5"
      cmssw_dir: "${project.cmssw_dir}"
      scram_arch: "el8_amd64_gcc12"

    memory:
      gen: 6000
      post: 4000
      tfds: 8000
      val: 8000 #fairly high mem limit due to CPU execution of the attention model and 8 threads
      train: 16000

    validation:
      job_types: ["pf", "mlpf"]
      use_cuda: False
      threads: 8

    validation_data:
      job_types: ["pf", "mlpf"]
      use_cuda: False
      threads: 8
      samples:
        jetmet0:
          input_filelist: "data/cms/run3/${project.site_name}/jetmet0.txt"
          seed_range: [1, 100]
          output_subdir: "Run2024C"

    slurm_partition: "${project.cpu_partition}"
    slurm_runtime: "48h"

    # Step 1: Generation
    # Defines the MC samples to be generated or processed
    samples:
      ttbar_pu:
        process_name: "TTbar_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [100000, 100210]
        events_per_job: 50
        output_subdir: "pu55to75"
        pu_type: "pu55to75"

      ztt_pu:
        process_name: "ZTT_All_hadronic_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [200000, 200210]
        events_per_job: 100
        output_subdir: "pu55to75"
        pu_type: "pu55to75"

      qcd_pu:
        process_name: "QCDForPF_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [300000, 300210]
        events_per_job: 100
        output_subdir: "pu55to75"
        pu_type: "pu55to75"

      qcd_val:
        process_name: "QCDForPF_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [400000, 400210]
        events_per_job: 100
        output_subdir: "pu55to75_val"
        pu_type: "pu55to75"
        copy_step2: True

      ttbar_nopu:
        process_name: "TTbar_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [800000, 800210]
        events_per_job: 100
        output_subdir: "nopu"
        pu_type: "nopu"

      qcd_nopu:
        process_name: "QCDForPF_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [1000000, 1000210]
        events_per_job: 100
        output_subdir: "nopu"
        pu_type: "nopu"

      ztt_nopu:
        process_name: "ZTT_All_hadronic_13p6TeV_TuneCUETP8M1_cfi"
        gen_script: "mlpf/data/cms/genjob_pu.sh"
        seed_range: [1100000, 1100210]
        events_per_job: 100
        output_subdir: "nopu"
        pu_type: "nopu"

    # Step 2: Postprocessing
    # Converts ROOT/EDM output to intermediate format (pkl/parquet) for ML
    postprocessing:
      script: "mlpf/data/cms/postprocessing2.py"
      args:
        save_full_graph: false
        num_events: -1 # process all

    # Step 3: TFDS Conversion
    # Mapping samples to TensorFlow Dataset Builders
    tfds_mapping:
      ttbar_pu:
        builder_path: "mlpf/heptfds/cms_pf/ttbar"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      qcd_pu:
        builder_path: "mlpf/heptfds/cms_pf/qcd"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      ztt_pu:
        builder_path: "mlpf/heptfds/cms_pf/ztt"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      ttbar_nopu:
        builder_path: "mlpf/heptfds/cms_pf/ttbar_nopu"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      qcd_nopu:
        builder_path: "mlpf/heptfds/cms_pf/qcd_nopu"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      ztt_nopu:
        builder_path: "mlpf/heptfds/cms_pf/ztt_nopu"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

  # CLD Production
  cld:
    type: "key4hep"
    model: "pyg-cld-v1"
    comment: "CLD production with Key4Hep"
    workspace_dir: "${project.paths.storage_root}/cld/v1.2.3_key4hep_2025-05-29_CLD_f1e8f9"
    config_dir: "${project.paths.project_root}/mlpf/data/key4hep/gen/cld/CLDConfig"
    gen_container: "/cvmfs/unpacked.cern.ch/gitlab-registry.cern.ch/key4hep/k4-deploy/alma9:latest"

    memory:
      gen: 4000
      post: 2000
      tfds: 4000

    slurm_partition: "main"
    slurm_runtime: "120m"

    samples:
      zz:
        process_name: "p8_ee_ZZ_ecm365"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [100000, 101010]
        events_per_job: 100
      zz_tautau_365:
        process_name: "p8_ee_ZZ_tautau_ecm365"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [100000, 101010]
        events_per_job: 100
      zz_tautau_240:
        process_name: "p8_ee_ZZ_tautau_ecm240"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [100000, 101010]
        events_per_job: 100
      zh_tautau_365:
        process_name: "p8_ee_ZH_Htautau_ecm365"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [200000, 201010]
        events_per_job: 100
      zh_tautau_240:
        process_name: "p8_ee_ZH_Htautau_ecm240"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [200000, 201010]
        events_per_job: 100
      ttbar:
        process_name: "p8_ee_ttbar_ecm365"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [300000, 300110]
        events_per_job: 100
      ww_fullhad:
        process_name: "p8_ee_WW_fullhad_ecm365"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [400000, 401010]
        events_per_job: 100
      qq:
        process_name: "p8_ee_qq_ecm365"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [500000, 501010]
        events_per_job: 100
      z_qq:
        process_name: "p8_ee_Z_qq_ecm91"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [600000, 601010]
        events_per_job: 100
      z_tautau:
        process_name: "p8_ee_Z_tautau_ecm91"
        gen_script: "mlpf/data/key4hep/gen/cld/run_sim.sh"
        seed_range: [700000, 701010]
        events_per_job: 100

    postprocessing:
      script: "mlpf/data/key4hep/postprocessing.py"
      args: {}

    tfds_mapping:
      ttbar:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep/ttbar"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      ww_fullhad:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep/ww_fullhad"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      zz:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep/zz"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      qq:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep/qq"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

    tfds_hit_mapping:
      ttbar:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep_hits/ttbar"
        version: "3.1.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      ww_fullhad:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep_hits/ww_fullhad"
        version: "3.1.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      zz:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep_hits/zz"
        version: "3.1.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      qq:
        builder_path: "mlpf/heptfds/cld_pf_edm4hep_hits/qq"
        version: "3.1.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

  # CLIC Production
  clic:
    type: "key4hep"
    model: "pyg-clic-v1"
    comment: "CLIC production with Key4Hep"
    workspace_dir: "${project.paths.storage_root}/clic/v1.2.4_key4hep_2025-05-29_CLIC_819e4e"
    config_dir: "${project.paths.project_root}/mlpf/data/key4hep/gen/clic"
    gen_container: "/cvmfs/unpacked.cern.ch/gitlab-registry.cern.ch/key4hep/k4-deploy/alma9:latest"

    memory:
      gen: 4000
      post: 1000
      tfds: 8000

    slurm_partition: "main"
    slurm_runtime: "240m"

    samples:
      ttbar:
        process_name: "p8_ee_ttbar_ecm380"
        gen_script: "mlpf/data/key4hep/gen/clic/run_sim.sh"
        seed_range: [300000, 301010]
        events_per_job: 100
      ww_fullhad:
        process_name: "p8_ee_WW_fullhad_ecm380"
        gen_script: "mlpf/data/key4hep/gen/clic/run_sim.sh"
        seed_range: [400000, 401010]
        events_per_job: 100
      qq:
        process_name: "p8_ee_qq_ecm380"
        gen_script: "mlpf/data/key4hep/gen/clic/run_sim.sh"
        seed_range: [500000, 501010]
        events_per_job: 100

    postprocessing:
      script: "mlpf/data/key4hep/postprocessing.py"
      args: {}

    tfds_mapping:
      ttbar:
        builder_path: "mlpf/heptfds/clic_pf_edm4hep/ttbar"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      ww_fullhad:
        builder_path: "mlpf/heptfds/clic_pf_edm4hep/ww_fullhad"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
      qq:
        builder_path: "mlpf/heptfds/clic_pf_edm4hep/qq"
        version: "3.0.0"
        config_ids: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

# -----------------------------------------------------------------------------
# Model Configurations
# Defines training recipes. Can inherit from defaults or other models.
# The multipliers are tuned for an 80GB A100 for training.
# -----------------------------------------------------------------------------
models:
  defaults: &defaults
    backend: pytorch
    gpus: 1
    threads: 16
    gpu_type: a100
    mem_per_gpu_mb: 80000
    slurm_partition: "${project.gpu_partition}"
    slurm_runtime: "48h"
    optimizer: adamw
    lr_schedule: cosinedecay
    dtype: bfloat16
    load: null
    num_steps: 100000
    comet: false
    comet_step_freq: 10000
    ntrain: null #use full train dataset for training
    nvalid: null #use full valid dataset for validation
    ntest: 100 #do test on a small subset
    sort_data: false
    num_workers: 8
    prefetch_factor: 2
    patience: 10000
    checkpoint_freq: 10000
    val_freq: 10000

  pyg-cms-v1:
    <<: *defaults
    dataset: cms
    gpu_batch_multiplier: 4

    # Dataset Selection
    train_datasets:
      physical_pu:
        batch_size: 1
        samples:
          - name: "cms_pf_ttbar"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_qcd"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_ztt"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      physical_nopu:
        batch_size: 16
        samples:
          - name: "cms_pf_ttbar_nopu"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_qcd_nopu"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_ztt_nopu"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    validation_datasets:
      physical_pu:
        batch_size: 1
        samples:
          - name: "cms_pf_ttbar"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_qcd"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_ztt"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      physical_nopu:
        batch_size: 16
        samples:
          - name: "cms_pf_ttbar_nopu"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_qcd_nopu"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cms_pf_ztt_nopu"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    test_datasets:
      - name: "cms_pf_ttbar"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cms_pf_qcd"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cms_pf_ztt"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    # Hyperparameters
    hyperparameters:
      batch_size: 1
      lr: 0.0004

    # Architecture Definition
    architecture:
      type: "attention"
      input_encoding: "split"
      attention:
        num_convs: 3
        head_dim: 16
        num_heads: 16

  # Example: CLD Model
  pyg-cld-v1:
    <<: *defaults
    dataset: cld
    gpu_batch_multiplier: 1024

    hyperparameters:
      batch_size: 1
      lr: 0.0005

    architecture:
      type: "attention"
      input_encoding: "split"
      attention:
        num_convs: 3
        head_dim: 16
        num_heads: 16

    # Dataset Selection
    train_datasets:
      physical:
        batch_size: 1
        samples:
          - name: "cld_edm_ttbar_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_ww_fullhad_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_qq_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    validation_datasets:
      physical:
        batch_size: 1
        samples:
          - name: "cld_edm_ttbar_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_ww_fullhad_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_qq_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    test_datasets:
      - name: "cld_edm_ttbar_pf"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cld_edm_ww_fullhad_pf"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cld_edm_qq_pf"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

  pyg-cld-hits-v1:
    <<: *defaults
    dataset: cld_hits
    gpu_batch_multiplier: 16

    hyperparameters:
      batch_size: 1
      lr: 0.0001

    architecture:
      type: "attention"
      input_encoding: "split"
      attention:
        num_convs: 3
        head_dim: 16
        num_heads: 16

    # Dataset Selection
    train_datasets:
      physical:
        batch_size: 1
        samples:
          - name: "cld_edm_ttbar_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_ww_fullhad_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_qq_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_zz_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    validation_datasets:
      physical:
        batch_size: 1
        samples:
          - name: "cld_edm_ttbar_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_ww_fullhad_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_qq_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "cld_edm_zz_hits"
            version: "3.1.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    test_datasets:
      - name: "cld_edm_ttbar_hits"
        version: "3.1.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cld_edm_ww_fullhad_hits"
        version: "3.1.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cld_edm_qq_hits"
        version: "3.1.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "cld_edm_zz_hits"
        version: "3.1.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

  # CLIC Model
  pyg-clic-v1:
    <<: *defaults
    dataset: clic
    gpu_batch_multiplier: 1024

    hyperparameters:
      batch_size: 1
      lr: 0.0005

    architecture:
      type: "attention"
      input_encoding: "split"
      attention:
        num_convs: 3
        head_dim: 16
        num_heads: 16

    # Dataset Selection
    train_datasets:
      physical:
        batch_size: 1
        samples:
          - name: "clic_edm_ttbar_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "clic_edm_ww_fullhad_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "clic_edm_qq_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    validation_datasets:
      physical:
        batch_size: 1
        samples:
          - name: "clic_edm_ttbar_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "clic_edm_ww_fullhad_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
          - name: "clic_edm_qq_pf"
            version: "3.0.0"
            splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]

    test_datasets:
      - name: "clic_edm_ttbar_pf"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "clic_edm_ww_fullhad_pf"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]
      - name: "clic_edm_qq_pf"
        version: "3.0.0"
        splits: ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"]