-
Notifications
You must be signed in to change notification settings - Fork 0
/
inference.py
162 lines (135 loc) · 5.21 KB
/
inference.py
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
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
import argparse
import os
import numpy as np
import torch
import torch.nn.functional as F
import tqdm
import xarray as xr
from datamodule import ERA5DataModule
from gravity_wave_model import UNetWithTransformer
from torch.nn.parallel import DistributedDataParallel as DDP
import torch.distributed as dist
local_rank = int(os.environ["LOCAL_RANK"])
rank = int(os.environ["RANK"])
device = f"cuda:{local_rank}"
dtype = torch.float32
def setup():
dist.init_process_group("nccl")
torch.cuda.set_device(local_rank)
def cleanup():
dist.destroy_process_group()
def load_checkpoint(model,ckpt_singular):
print('Loading weights from', ckpt_singular)
state_dict = torch.load(f=ckpt_singular, map_location=device, weights_only=True)
ignore_layers = [
"input_scalers_mu",
"input_scalers_sigma",
"static_input_scalers_mu",
"static_input_scalers_sigma",
"patch_embedding.proj.weight",
"patch_embedding_static.proj.weight",
"unembed.weight",
"unembed.bias",
"output_scalers",
]
for layer in ignore_layers:
state_dict.pop(layer, None)
model.load_state_dict(state_dict)
print('Loaded weights')
return model
def get_model(cfg, vartype,ckpt_singular: str) -> torch.nn.Module:
model: torch.nn.Module = UNetWithTransformer(
lr=cfg.lr,
hidden_channels=cfg.hidden_channels,
in_channels={"uvtheta122": 366, "uvtp122": 488, "uvtp14": 56}[vartype],
out_channels={"uvtheta122": 244, "uvtp122": 366, "uvtp14": 42}[vartype],
n_lats_px=cfg.n_lats_px,
n_lons_px=cfg.n_lons_px,
in_channels_static=cfg.in_channels_static,
mask_unit_size_px=cfg.mask_unit_size_px,
patch_size_px=cfg.patch_size_px,
device=device,
)
model = DDP(model.to(local_rank, dtype=dtype), device_ids=[local_rank])
model = load_checkpoint(model,ckpt_singular)
return model
def get_data(data_path: str, file_glob_pattern: str) -> torch.utils.data.DataLoader:
datamodule = ERA5DataModule(
batch_size=8,
num_data_workers=8,
train_data_path=None,
valid_data_path=data_path,
file_glob_pattern=file_glob_pattern,
)
datamodule.setup(stage="predict")
dataloader = datamodule.predict_dataloader()
return dataloader
def main(cfg, vartype, ckpt_path: str, data_path: str, results_dir: str, file_glob_pattern: str='*.nc'):
setup()
model: torch.nn.Module = get_model(cfg, vartype, ckpt_singular=ckpt_path)
dataloader: torch.utils.data.DataLoader = get_data(
data_path=data_path, file_glob_pattern=file_glob_pattern
)
# Pre-allocate an xarray.DataArray to store results
da_results: xr.DataArray = xr.full_like(other=dataloader.dataset.ds.isel(odim=slice(0, model.module.decoder.final_conv.out_channels)).output,fill_value=np.NaN,)
# assert da_results.sizes == {"time": 744, "odim": 42, "lat": 64, "lon": 128}
# Main prediction loop
total: int = len(dataloader)
pbar = tqdm.tqdm(iterable=enumerate(dataloader), total=total)
for i, batch in pbar:
batch = {
k: v.to(device="cuda") for k, v in batch.items()
} # move data to the same device as the model
with torch.no_grad():
output: torch.Tensor = model(batch) # run inference
# assert output.shape == torch.Size([8, 366, 64, 128])
# Save input and output tensors to Pytorch format
# torch.save(output, os.path.join(results_dir, f"output_{i}.pt"))
# torch.save(batch, os.path.join(results_dir, f"input_batch_{i}.pt"))
# Save output to NetCDF
t_start: int = i * dataloader.batch_size
t_stop: int = t_start + dataloader.batch_size
t_slice = slice(t_start, t_stop)
da_results[dict(time=t_slice)] = xr.DataArray(
data=output.cpu(),
dims=da_results.dims,
coords=da_results.isel(time=t_slice).coords,
)
if i % 20 == 0 or i == total - 1: # Output to NetCDF every 20 steps
da_results.to_netcdf(
path=os.path.join(results_dir, "output.nc"),
mode="w", # always rewrite file
)
# Report loss
loss: torch.Tensor = F.mse_loss(input=output, target=batch["target"])
pbar.set_postfix(
ordered_dict={
"t_slice": f"{t_slice.start}-{t_slice.stop}", # time slice
"loss": loss.item(),
}
)
# %%
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--split",
default="uvtp122",
)
parser.add_argument(
"--ckpt_path",
default="checkpoints/uvtp122/magnet-flux-uvtp122-epoch-06-loss-0.2274.pt",
)
parser.add_argument(
"--data_path",
default="gravity_wave_flux/uvtp122",
)
parser.add_argument(
"--results_dir",
default="results/uvtp122",
)
args = parser.parse_args()
from config import get_cfg
cfg = get_cfg()
os.makedirs(name=args.results_dir, exist_ok=True)
main(cfg,args.split, args.ckpt_path, args.data_path, args.results_dir,)
cleanup()