Upload config_finetuning.yaml

config_finetuning.yaml

@@ -0,0 +1,483 @@
+data:
+  format: zarr
+  resolution: n320
+  frequency: 6h
+  timestep: 6h
+  forcing:
+    - cos_latitude
+    - cos_longitude 
+    - sin_latitude
+    - sin_longitude
+    - cos_julian_day
+    - cos_local_time
+    - sin_julian_day
+    - sin_local_time
+    - insolation
+    - lsm
+    - sdor
+    - slor
+    - z
+  diagnostic:
+    - tp
+    - cp
+    - sf
+    - tcc
+    - hcc
+    - lcc
+    - mcc
+    - ro
+    - ssrd
+    - strd
+    - 100u
+    - 100v
+  remapped: null
+  normalizer:
+    default: mean-std
+    remap:
+      cp: tp
+      sf: tp
+    std:
+      - tp
+      - cp
+      - sf
+      - ro
+      - tcw
+      - ssrd
+      - q_50
+      - q_100
+      - q_150
+      - q_200
+      - q_250
+      - q_300
+      - q_400
+      - q_500
+      - q_600
+      - q_700
+      - q_850
+      - q_925
+      - q_1000
+    min-max: null
+    max:
+      - sdor
+      - slor
+      - z
+    none:
+      - cos_latitude
+      - cos_longitude
+      - sin_latitude
+      - sin_longitude
+      - cos_julian_day
+      - cos_local_time
+      - sin_julian_day
+      - sin_local_time
+      - insolation
+      - lsm
+      - tcc
+      - mcc
+      - hcc
+      - lcc
+      - swvl1
+      - swvl2
+  imputer:
+    default: none
+  remapper:
+    default: none
+  processors:
+    normalizer:
+      _target_: anemoi.models.preprocessing.normalizer.InputNormalizer
+      _convert_: all
+      config:
+        default: mean-std
+        remap:
+          cp: tp
+          sf: tp
+        std:
+          - tp
+          - cp
+          - sf
+          - ro
+          - tcw
+          - ssrd
+          - q_50
+          - q_100
+          - q_150
+          - q_200
+          - q_250
+          - q_300
+          - q_400
+          - q_500
+          - q_600
+          - q_700
+          - q_850
+          - q_925
+          - q_1000
+        min-max: null
+        max:
+          - sdor
+          - slor
+          - z
+        none:
+          - cos_latitude
+          - cos_longitude
+          - sin_latitude
+          - sin_longitude
+          - cos_julian_day
+          - cos_local_time
+          - sin_julian_day
+          - sin_local_time
+          - insolation
+          - lsm
+          - tcc
+          - mcc
+          - hcc
+          - lcc
+          - swvl1
+          - swvl2
+  num_features: 115
+
+dataloader:
+  prefetch_factor: 2
+  pin_memory: True
+  read_group_size: 4
+  num_workers:
+    training: 8
+    validation: 8
+    test: 8
+    predict: 8
+  batch_size:
+    training: 1
+    validation: 1
+    test: 4
+    predict: 4
+  limit_batches:
+    training: 1000
+    validation: 10
+    test: 20
+    predict: 20
+  dataset: ${hardware.paths.data}/${hardware.files.dataset}
+  land_dataset: ${hardware.paths.data}/${hardware.files.dataset_land}
+  land_variables: [100u, 100v, swvl1, swvl2, stl1, stl2, tcc, lcc, mcc, hcc, sf, ro, strd, ssrd]
+  training:
+    dataset:
+      - dataset: ${dataloader.dataset}
+        start: null
+        end: 2022
+        frequency: ${data.frequency}
+        drop: []
+      - dataset: ${dataloader.land_dataset}
+        start: null
+        end: 2022
+        frequency: ${data.frequency}
+        select: ${dataloader.land_variables}
+    start: null
+    end: 2022
+    drop: []
+  validation:
+    dataset:
+      - dataset: ${dataloader.dataset}
+        start: 2022
+        end: 2022
+        frequency: ${data.frequency}
+        drop: []
+      - dataset: ${dataloader.land_dataset}
+        start: 2022
+        end: 2022
+        frequency: ${data.frequency}
+        select: ${dataloader.land_variables}
+    start: 2022
+    end: 2022
+    drop: []
+  validation_rollout: 1
+
+diagnostics:
+  plot:
+    asynchronous: False
+    datashader: True
+    frequency:
+      batch: 750
+      epoch: 10
+    parameters: [tp]
+    sample_idx: 0
+    precip_and_related_fields: [tp, cp]
+    callbacks: []
+    enabled: True
+    scatter: False
+    mode: asyncio
+  callbacks: {}
+  benchmark_profiler:
+    memory:
+      enabled: True
+      steps: 5
+      warmup: 2
+      extra_plots: False
+      trace_rank0_only: False
+    time:
+      enabled: True
+      verbose: False
+    speed:
+      enabled: True
+    system:
+      enabled: True
+    model_summary:
+      enabled: True
+    snapshot:
+      enabled: True
+      steps: 4
+      warmup: 0
+  debug:
+    anomaly_detection: False
+  profiler: False
+  enable_checkpointing: True
+  checkpoint:
+    every_n_minutes:
+      save_frequency: 30
+      num_models_saved: 3
+    every_n_epochs:
+      save_frequency: 1
+      num_models_saved: 3
+    every_n_train_steps:
+      save_frequency: null
+      num_models_saved: 0
+  log:
+    wandb:
+      enabled: False
+    tensorboard:
+      enabled: False
+    mlflow:
+      enabled: False
+    interval: 100
+  enable_progress_bar: True
+  print_memory_summary: False
+
+hardware:
+  paths:
+    data: ${oc.decode:${oc.env:DATASETS_PATH}}
+    output: ${oc.decode:${oc.env:OUTPUT_DIR}}
+    logs:
+      base: ${hardware.paths.output}/logs
+      wandb: ${hardware.paths.output}/logs/wandb
+      mlflow: ${hardware.paths.output}/logs/mlflow
+      tensorboard: ${hardware.paths.output}/logs/tensorboard
+    checkpoints: ${hardware.paths.output}/checkpoint/
+    plots: ${hardware.paths.output}/plots/
+    profiler: ${hardware.paths.output}/profiler/
+    graph: ${hardware.paths.output}/graphs/
+  files:
+    dataset: aifs-od-an-oper-0001-mars-n320-2016-2023-6h-v6.zarr
+    dataset_land: aifs-od-an-oper-0001-mars-n320-2016-2023-6h-v1-land.zarr
+    graph: graph_enc_proc_dec_n320.pt
+    checkpoint:
+      every_n_epochs: aifs-by_epoch-epoch_{epoch:03d}-val_wmse_{val_wmse:.3e}
+      every_n_train_steps: aifs-by_step-epoch_{epoch:03d}-step_{step:06d}
+      every_n_minutes: aifs-by_time-epoch_{epoch:03d}-step_{step:06d}
+    warm_start: null
+  accelerator: auto
+  num_gpus_per_node: 4
+  num_nodes: 16
+  num_gpus_per_model: 4
+
+graph:
+  overwrite: True
+  data: data
+  hidden: hidden
+  nodes:
+    data:
+      node_builder:
+        _target_: anemoi.graphs.nodes.ZarrDatasetNodes
+        dataset: ${dataloader.dataset}
+      attributes:
+        area_weight:
+          _target_: anemoi.graphs.nodes.attributes.AreaWeights
+          norm: unit-max
+    hidden:
+      node_builder:
+        _target_: anemoi.graphs.nodes.ReducedGaussianGridNodes
+        grid: o96
+  edges:
+    - source_name: data
+      target_name: hidden
+      edge_builder:
+        _target_: anemoi.graphs.edges.CutOffEdges
+        cutoff_factor: 0.6
+      attributes:
+        edge_length:
+          _target_: anemoi.graphs.edges.attributes.EdgeLength
+          norm: unit-std
+        edge_dirs:
+          _target_: anemoi.graphs.edges.attributes.EdgeDirection
+          norm: unit-std
+    - source_name: hidden
+      target_name: data
+      edge_builder:
+        _target_: anemoi.graphs.edges.KNNEdges
+        num_nearest_neighbours: 3
+      attributes:
+        edge_length:
+          _target_: anemoi.graphs.edges.attributes.EdgeLength
+          norm: unit-std
+        edge_dirs:
+          _target_: anemoi.graphs.edges.attributes.EdgeDirection
+          norm: unit-std
+  attributes:
+    nodes:
+      area_weight:
+        _target_: anemoi.graphs.nodes.attributes.AreaWeights
+        norm: unit-max
+    edges:
+      edge_length:
+        _target_: anemoi.graphs.edges.attributes.EdgeLength
+        norm: unit-std
+      edge_dirs:
+        _target_: anemoi.graphs.edges.attributes.EdgeDirection
+        norm: unit-std
+
+model:
+  activation: GELU
+  num_channels: 1024
+  model:
+    _target_: anemoi.models.models.encoder_processor_decoder.AnemoiModelEncProcDec
+  processor:
+    _target_: anemoi.models.layers.processor.TransformerProcessor
+    _convert_: all
+    activation: GELU
+    num_layers: 16
+    num_chunks: 2
+    mlp_hidden_ratio: 4
+    num_heads: 16
+    window_size: 1120
+    dropout_p: 0.0
+  encoder:
+    _target_: anemoi.models.layers.mapper.GraphTransformerForwardMapper
+    _convert_: all
+    trainable_size: 8
+    sub_graph_edge_attributes: [edge_length, edge_dirs]
+    activation: GELU
+    num_chunks: 1
+    mlp_hidden_ratio: 4
+    num_heads: 16
+  decoder:
+    _target_: anemoi.models.layers.mapper.GraphTransformerBackwardMapper
+    _convert_: all
+    trainable_size: 8
+    sub_graph_edge_attributes: [edge_length, edge_dirs]
+    activation: GELU
+    num_chunks: 1
+    mlp_hidden_ratio: 4
+    num_heads: 16
+  trainable_parameters:
+    data: 8
+    hidden: 8
+    data2hidden: 8
+    hidden2data: 8
+  attributes:
+    edges: [edge_length, edge_dirs]
+    nodes: []
+  node_loss_weight: area_weight
+  bounding:
+    - _target_: anemoi.models.layers.bounding.ReluBounding
+      variables:
+        - tp
+        - ro
+        - tcw
+        - ssrd
+        - q_50
+        - q_100
+        - q_150
+        - q_200
+        - q_250
+        - q_300
+        - q_400
+        - q_500
+        - q_600
+        - q_700
+        - q_850
+        - q_925
+        - q_1000
+    - _target_: anemoi.models.layers.bounding.HardtanhBounding
+      variables: [tcc, swvl1, swvl2]
+      min_val: 0
+      max_val: 1
+    - _target_: anemoi.models.layers.bounding.FractionBounding
+      variables: [cp, sf]
+      min_val: 0
+      max_val: 1
+      total_var: tp
+    - _target_: anemoi.models.layers.bounding.FractionBounding
+      variables: [lcc, mcc, hcc]
+      min_val: 0
+      max_val: 1
+      total_var: tcc
+
+training:
+  run_id: ${oc.decode:${oc.env:PRETRAINING_RUN_ID}}
+  fork_run_id: ${oc.decode:${oc.env:PRETRAINING_RUN_ID}}
+  load_weights_only: True
+  deterministic: False
+  precision: 16-mixed
+  multistep_input: 2
+  accum_grad_batches: 1
+  num_sanity_val_steps: 6
+  gradient_clip:
+    val: 32.0
+    algorithm: value
+  swa:
+    enabled: False
+    lr: 0.0001
+  zero_optimizer: False
+  training_loss:
+    _target_: anemoi.training.losses.mse.WeightedMSELoss
+    scalars:
+      - variable
+      - loss_weights_mask
+    ignore_nans: False
+  loss_gradient_scaling: False
+  validation_metrics:
+    - _target_: anemoi.training.losses.mse.WeightedMSELoss
+      scalars: []
+      ignore_nans: True
+  rollout:
+    start: 1
+    epoch_increment: 1
+    max: 12
+  max_epochs: 13
+  max_steps: 150000
+  lr:
+    rate: 8.0e-7
+    iterations: 7900
+    min: 3.0e-7
+    warmup_t: 100
+  variable_loss_scaling:
+    default: 1
+    pl:
+      q: 0.6
+      t: 6
+      u: 0.8
+      v: 0.5
+      w: 0.001
+      z: 12
+    sfc:
+      sp: 10
+      10u: 0.5
+      10v: 0.5
+      100u: 0.1
+      100v: 0.1
+      2d: 0.5
+      tp: 0.025
+      cp: 0.0025
+      ro: 0.005
+      sf: 0.025
+      tcc: 0.1
+      mcc: 0.1
+      lcc: 0.1
+      hcc: 0.1
+      swvl2: 200
+      swvl1: 100
+      stl2: 10
+      stl1: 1
+      ssrd: 0.05
+      strd: 0.1
+  metrics: [z_500, t_850, u_850, v_850]
+  pressure_level_scaler:
+    _target_: anemoi.training.data.scaling.ReluPressureLevelScaler
+    minimum: 0.2
+    slope: 0.001