Spaces:
Build error
Build error
Update app.py
Browse files
app.py
CHANGED
@@ -1,4 +1,331 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
st.write(x, 'squared is', x * x)
|
|
|
1 |
+
# import streamlit as st
|
2 |
+
|
3 |
+
# x = st.slider('Select a value')
|
4 |
+
# st.write(x, 'squared is', x * x)
|
5 |
+
|
6 |
+
|
7 |
import streamlit as st
|
8 |
+
import random
|
9 |
+
from pathlib import Path
|
10 |
+
import matplotlib.pyplot as plt
|
11 |
+
import numpy as np
|
12 |
+
import torch
|
13 |
+
from huggingface_hub import hf_hub_download, snapshot_download
|
14 |
+
import tarfile
|
15 |
+
import os
|
16 |
+
import sys
|
17 |
+
import yaml
|
18 |
+
|
19 |
+
st.title("PrithviWxC Model Inference")
|
20 |
+
|
21 |
+
st.write("Setting up environment...")
|
22 |
+
|
23 |
+
# Set up torch backends and seeds
|
24 |
+
torch.jit.enable_onednn_fusion(True)
|
25 |
+
if torch.cuda.is_available():
|
26 |
+
st.write(f"Using device: {torch.cuda.get_device_name()}")
|
27 |
+
torch.backends.cudnn.benchmark = True
|
28 |
+
torch.backends.cudnn.deterministic = True
|
29 |
+
|
30 |
+
random.seed(42)
|
31 |
+
if torch.cuda.is_available():
|
32 |
+
torch.cuda.manual_seed(42)
|
33 |
+
torch.manual_seed(42)
|
34 |
+
np.random.seed(42)
|
35 |
+
|
36 |
+
# Set device
|
37 |
+
if torch.cuda.is_available():
|
38 |
+
device = torch.device("cuda")
|
39 |
+
else:
|
40 |
+
device = torch.device("cpu")
|
41 |
+
|
42 |
+
st.write(f"Using device: {device}")
|
43 |
+
|
44 |
+
# Download and extract PrithviWxC module
|
45 |
+
st.write("Downloading and setting up PrithviWxC module...")
|
46 |
+
|
47 |
+
module_tar_path = hf_hub_download(
|
48 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
49 |
+
filename="PrithviWxC.tar.gz",
|
50 |
+
local_dir=".",
|
51 |
+
force_download=True
|
52 |
+
)
|
53 |
+
|
54 |
+
with tarfile.open(module_tar_path, "r:gz") as tar:
|
55 |
+
tar.extractall(path=".")
|
56 |
+
|
57 |
+
# Add the module path to sys.path
|
58 |
+
sys.path.append(os.path.abspath("./PrithviWxC"))
|
59 |
+
|
60 |
+
st.write("PrithviWxC module imported successfully.")
|
61 |
+
|
62 |
+
# Now import the module
|
63 |
+
from PrithviWxC.dataloaders.merra2 import Merra2Dataset, input_scalers, output_scalers, static_input_scalers, preproc
|
64 |
+
from PrithviWxC.model import PrithviWxC
|
65 |
+
|
66 |
+
# Variables and times
|
67 |
+
surface_vars = [
|
68 |
+
"EFLUX",
|
69 |
+
"GWETROOT",
|
70 |
+
"HFLUX",
|
71 |
+
"LAI",
|
72 |
+
"LWGAB",
|
73 |
+
"LWGEM",
|
74 |
+
"LWTUP",
|
75 |
+
"PS",
|
76 |
+
"QV2M",
|
77 |
+
"SLP",
|
78 |
+
"SWGNT",
|
79 |
+
"SWTNT",
|
80 |
+
"T2M",
|
81 |
+
"TQI",
|
82 |
+
"TQL",
|
83 |
+
"TQV",
|
84 |
+
"TS",
|
85 |
+
"U10M",
|
86 |
+
"V10M",
|
87 |
+
"Z0M",
|
88 |
+
]
|
89 |
+
static_surface_vars = ["FRACI", "FRLAND", "FROCEAN", "PHIS"]
|
90 |
+
vertical_vars = ["CLOUD", "H", "OMEGA", "PL", "QI", "QL", "QV", "T", "U", "V"]
|
91 |
+
levels = [
|
92 |
+
34.0,
|
93 |
+
39.0,
|
94 |
+
41.0,
|
95 |
+
43.0,
|
96 |
+
44.0,
|
97 |
+
45.0,
|
98 |
+
48.0,
|
99 |
+
51.0,
|
100 |
+
53.0,
|
101 |
+
56.0,
|
102 |
+
63.0,
|
103 |
+
68.0,
|
104 |
+
71.0,
|
105 |
+
72.0,
|
106 |
+
]
|
107 |
+
padding = {"level": [0, 0], "lat": [0, -1], "lon": [0, 0]}
|
108 |
+
|
109 |
+
st.write("Setting up dataset parameters...")
|
110 |
+
|
111 |
+
# User inputs for lead times and input times
|
112 |
+
lead_time = st.number_input("Lead Time (hours)", min_value=1, max_value=24, value=6)
|
113 |
+
input_time = st.number_input("Input Time Difference (hours)", min_value=-24, max_value=0, value=-6)
|
114 |
+
|
115 |
+
lead_times = [lead_time] # This variable can be changed to change the task
|
116 |
+
input_times = [input_time] # This variable can be changed to change the task
|
117 |
+
|
118 |
+
# Data file
|
119 |
+
time_range = ("2020-01-01T00:00:00", "2020-01-01T23:59:59")
|
120 |
+
|
121 |
+
st.write("Downloading data files...")
|
122 |
+
|
123 |
+
surf_dir = Path("./merra-2")
|
124 |
+
snapshot_download(
|
125 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
126 |
+
allow_patterns="merra-2/MERRA2_sfc_2020010[1].nc",
|
127 |
+
local_dir=".",
|
128 |
+
force_download=True,
|
129 |
+
)
|
130 |
+
|
131 |
+
vert_dir = Path("./merra-2")
|
132 |
+
snapshot_download(
|
133 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
134 |
+
allow_patterns="merra-2/MERRA_pres_2020010[1].nc",
|
135 |
+
local_dir=".",
|
136 |
+
force_download=True,
|
137 |
+
)
|
138 |
+
|
139 |
+
# Climatology
|
140 |
+
surf_clim_dir = Path("./climatology")
|
141 |
+
snapshot_download(
|
142 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
143 |
+
allow_patterns="climatology/climate_surface_doy00[1]*.nc",
|
144 |
+
local_dir=".",
|
145 |
+
force_download=True,
|
146 |
+
)
|
147 |
+
|
148 |
+
vert_clim_dir = Path("./climatology")
|
149 |
+
snapshot_download(
|
150 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
151 |
+
allow_patterns="climatology/climate_vertical_doy00[1]*.nc",
|
152 |
+
local_dir=".",
|
153 |
+
force_download=True,
|
154 |
+
)
|
155 |
+
|
156 |
+
st.write("Setting positional encoding...")
|
157 |
+
|
158 |
+
positional_encoding = "fourier"
|
159 |
+
|
160 |
+
st.write("Initializing dataset...")
|
161 |
+
|
162 |
+
dataset = Merra2Dataset(
|
163 |
+
time_range=time_range,
|
164 |
+
lead_times=lead_times,
|
165 |
+
input_times=input_times,
|
166 |
+
data_path_surface=surf_dir,
|
167 |
+
data_path_vertical=vert_dir,
|
168 |
+
climatology_path_surface=surf_clim_dir,
|
169 |
+
climatology_path_vertical=vert_clim_dir,
|
170 |
+
surface_vars=surface_vars,
|
171 |
+
static_surface_vars=static_surface_vars,
|
172 |
+
vertical_vars=vertical_vars,
|
173 |
+
levels=levels,
|
174 |
+
positional_encoding=positional_encoding,
|
175 |
+
)
|
176 |
+
|
177 |
+
assert len(dataset) > 0, "There doesn't seem to be any valid data."
|
178 |
+
|
179 |
+
st.write("Loading scalers...")
|
180 |
+
|
181 |
+
surf_in_scal_path = Path("./climatology/musigma_surface.nc")
|
182 |
+
hf_hub_download(
|
183 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
184 |
+
filename=f"climatology/{surf_in_scal_path.name}",
|
185 |
+
local_dir=".",
|
186 |
+
force_download=True,
|
187 |
+
)
|
188 |
+
|
189 |
+
vert_in_scal_path = Path("./climatology/musigma_vertical.nc")
|
190 |
+
hf_hub_download(
|
191 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
192 |
+
filename=f"climatology/{vert_in_scal_path.name}",
|
193 |
+
local_dir=".",
|
194 |
+
force_download=True,
|
195 |
+
)
|
196 |
+
|
197 |
+
surf_out_scal_path = Path("./climatology/anomaly_variance_surface.nc")
|
198 |
+
hf_hub_download(
|
199 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
200 |
+
filename=f"climatology/{surf_out_scal_path.name}",
|
201 |
+
local_dir=".",
|
202 |
+
force_download=True,
|
203 |
+
)
|
204 |
+
|
205 |
+
vert_out_scal_path = Path("./climatology/anomaly_variance_vertical.nc")
|
206 |
+
hf_hub_download(
|
207 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
208 |
+
filename=f"climatology/{vert_out_scal_path.name}",
|
209 |
+
local_dir=".",
|
210 |
+
force_download=True,
|
211 |
+
)
|
212 |
+
|
213 |
+
in_mu, in_sig = input_scalers(
|
214 |
+
surface_vars,
|
215 |
+
vertical_vars,
|
216 |
+
levels,
|
217 |
+
surf_in_scal_path,
|
218 |
+
vert_in_scal_path,
|
219 |
+
)
|
220 |
+
|
221 |
+
output_sig = output_scalers(
|
222 |
+
surface_vars,
|
223 |
+
vertical_vars,
|
224 |
+
levels,
|
225 |
+
surf_out_scal_path,
|
226 |
+
vert_out_scal_path,
|
227 |
+
)
|
228 |
+
|
229 |
+
static_mu, static_sig = static_input_scalers(
|
230 |
+
surf_in_scal_path,
|
231 |
+
static_surface_vars,
|
232 |
+
)
|
233 |
+
|
234 |
+
st.write("Setting up model...")
|
235 |
+
|
236 |
+
residual = "climate"
|
237 |
+
masking_mode = "local"
|
238 |
+
decoder_shifting = True
|
239 |
+
masking_ratio = 0.99
|
240 |
+
|
241 |
+
# Load model config
|
242 |
+
hf_hub_download(
|
243 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
244 |
+
filename="config.yaml",
|
245 |
+
local_dir=".",
|
246 |
+
force_download=True,
|
247 |
+
)
|
248 |
+
|
249 |
+
with open("./config.yaml", "r") as f:
|
250 |
+
config = yaml.safe_load(f)
|
251 |
+
|
252 |
+
model = PrithviWxC(
|
253 |
+
in_channels=config["params"]["in_channels"],
|
254 |
+
input_size_time=config["params"]["input_size_time"],
|
255 |
+
in_channels_static=config["params"]["in_channels_static"],
|
256 |
+
input_scalers_mu=in_mu,
|
257 |
+
input_scalers_sigma=in_sig,
|
258 |
+
input_scalers_epsilon=config["params"]["input_scalers_epsilon"],
|
259 |
+
static_input_scalers_mu=static_mu,
|
260 |
+
static_input_scalers_sigma=static_sig,
|
261 |
+
static_input_scalers_epsilon=config["params"]["static_input_scalers_epsilon"],
|
262 |
+
output_scalers=output_sig**0.5,
|
263 |
+
n_lats_px=config["params"]["n_lats_px"],
|
264 |
+
n_lons_px=config["params"]["n_lons_px"],
|
265 |
+
patch_size_px=config["params"]["patch_size_px"],
|
266 |
+
mask_unit_size_px=config["params"]["mask_unit_size_px"],
|
267 |
+
mask_ratio_inputs=masking_ratio,
|
268 |
+
embed_dim=config["params"]["embed_dim"],
|
269 |
+
n_blocks_encoder=config["params"]["n_blocks_encoder"],
|
270 |
+
n_blocks_decoder=config["params"]["n_blocks_decoder"],
|
271 |
+
mlp_multiplier=config["params"]["mlp_multiplier"],
|
272 |
+
n_heads=config["params"]["n_heads"],
|
273 |
+
dropout=config["params"]["dropout"],
|
274 |
+
drop_path=config["params"]["drop_path"],
|
275 |
+
parameter_dropout=config["params"]["parameter_dropout"],
|
276 |
+
residual=residual,
|
277 |
+
masking_mode=masking_mode,
|
278 |
+
decoder_shifting=decoder_shifting,
|
279 |
+
positional_encoding=positional_encoding,
|
280 |
+
checkpoint_encoder=[],
|
281 |
+
checkpoint_decoder=[],
|
282 |
+
)
|
283 |
+
|
284 |
+
st.write("Loading model weights...")
|
285 |
+
|
286 |
+
weights_path = Path("./weights/prithvi.wxc.2300m.v1.pt")
|
287 |
+
hf_hub_download(
|
288 |
+
repo_id="Prithvi-WxC/prithvi.wxc.2300m.v1",
|
289 |
+
filename=weights_path.name,
|
290 |
+
local_dir="./weights",
|
291 |
+
force_download=True,
|
292 |
+
)
|
293 |
+
|
294 |
+
state_dict = torch.load(weights_path, map_location=device)
|
295 |
+
if "model_state" in state_dict:
|
296 |
+
state_dict = state_dict["model_state"]
|
297 |
+
model.load_state_dict(state_dict, strict=True)
|
298 |
+
|
299 |
+
model = model.to(device)
|
300 |
+
|
301 |
+
st.write("Model loaded and ready.")
|
302 |
+
|
303 |
+
if st.button("Run Inference"):
|
304 |
+
st.write("Running inference...")
|
305 |
+
|
306 |
+
data = next(iter(dataset))
|
307 |
+
batch = preproc([data], padding)
|
308 |
+
|
309 |
+
for k, v in batch.items():
|
310 |
+
if isinstance(v, torch.Tensor):
|
311 |
+
batch[k] = v.to(device)
|
312 |
+
|
313 |
+
with torch.no_grad():
|
314 |
+
model.eval()
|
315 |
+
out = model(batch)
|
316 |
+
|
317 |
+
st.write("Inference completed. Generating plot...")
|
318 |
+
|
319 |
+
t2m = out[0, 12].cpu().numpy()
|
320 |
+
|
321 |
+
lat = np.linspace(-90, 90, out.shape[-2])
|
322 |
+
lon = np.linspace(-180, 180, out.shape[-1])
|
323 |
+
X, Y = np.meshgrid(lon, lat)
|
324 |
+
|
325 |
+
fig, ax = plt.subplots()
|
326 |
+
cs = ax.contourf(X, Y, t2m, 100)
|
327 |
+
ax.set_aspect("equal")
|
328 |
+
plt.colorbar(cs)
|
329 |
+
st.pyplot(fig)
|
330 |
|
331 |
+
st.write("Plot generated.")
|
|