File size: 3,403 Bytes
0ba2894 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 | {
"model_name": "Prithvi WxC",
"model_type": "prithvi_wxc",
"architectures": [
"PrithviWxC"
],
"framework": "PyTorch",
"domain": "climate-and-atmosphere",
"task": "global-weather-forecasting",
"implementation": {
"entry_point": "model/prithvi_wxc.py",
"scope": "YAML-driven wrapper around the vendored official Hiera-MaxViT encoder-decoder, with identity model scalers for the repository's small connectivity configuration"
},
"architecture": {
"family": "Hiera-MaxViT encoder-decoder Vision Transformer",
"attention_mechanism": "alternating local block attention and global grid attention",
"input_format": "BTCHW atmospheric history plus BCHW static fields",
"output_format": "BCHW next atmospheric state",
"tokenization": "Conv2d patch embedding followed by mask-unit grouping and pixel-shuffle reconstruction",
"activation": "GELU",
"normalization": "LayerNorm",
"repository_default_config": {
"purpose": "small connectivity validation with synthetic or reduced ERA5 data",
"in_channels": 6,
"input_size_time": 2,
"in_channels_static": 4,
"grid_size": [
32,
64
],
"patch_size_px": [
2,
2
],
"mask_unit_size_px": [
8,
8
],
"mask_ratio_inputs": 0.0,
"embed_dim": 32,
"n_blocks_encoder": 1,
"n_blocks_decoder": 1,
"n_heads": 4,
"mlp_multiplier": 4.0,
"dropout": 0.0,
"drop_path": 0.0,
"parameter_dropout": 0.0,
"residual": "none",
"masking_mode": "global",
"positional_encoding": "absolute",
"encoder_shifting": false,
"decoder_shifting": false,
"normalization_scalers": "identity in the YAML wrapper"
},
"paper_reference_config": {
"grid_resolution_degrees": [
0.5,
0.625
],
"grid_size": "approximately 360/361 x 576",
"dynamic_channels": 160,
"embed_dim": 2560,
"encoder_blocks": "13 local plus 12 global blocks",
"decoder_blocks": "3 local plus 2 global blocks",
"n_heads": 16,
"pretraining_mask_ratio_inputs": 0.5,
"forecast_finetuning_mask_ratio_inputs": 0.0,
"parameter_count": "approximately 2.3 billion",
"compatibility_note": "official 2.3B checkpoints do not match the repository default small configuration without aligning channels, grid size, and architecture"
}
},
"data": {
"dataset": "ERA5",
"storage": "HDF5 fields with shape TCHW",
"temporal_interval_hours": 6,
"input_steps": 2,
"output_steps": 1,
"forecast_lead_time_hours": 6,
"dynamic_channels": [
"10m_u_component_of_wind",
"10m_v_component_of_wind",
"2m_temperature",
"mean_sea_level_pressure",
"geopotential_500",
"temperature_850"
],
"static_channels": 4,
"default_spatial_size": [
32,
64
],
"default_year_splits": {
"train": [
1951,
1952
],
"validation": [
1953
],
"test": [
1954
]
},
"synthetic_data_generator": "scripts/fake_data.py"
},
"configuration_sources": [
"conf/config.yaml",
"model/prithvi_wxc.py",
"model/prithvi_wxc_official.py",
"scripts/train.py",
"scripts/inference.py",
"scripts/fake_data.py",
"README.md",
"configuration.json"
]
}
|