README.md CHANGED
@@ -13,6 +13,11 @@ individual trained checkpoints for the atmosphere and ocean components, which
13
  can be used to improve model variability in ensembles. The model architecture
14
  is a U-Net with padding operations modified to support using the HEALPix grid.
15
 
 
 
 
 
 
16
  For training recipies see [PhysicsNeMo](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/dlwp_healpix), for inference [Earth2Studio](https://nvidia.github.io/earth2studio/examples/14_dlesym_example.html#sphx-glr-examples-14-dlesym-example-py).
17
 
18
  This model is ready for commercial/non-commercial use.
@@ -178,4 +183,4 @@ corresponds to an approximate 1 degree lat/lon grid. <br>
178
 
179
  NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
180
 
181
- Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
 
13
  can be used to improve model variability in ensembles. The model architecture
14
  is a U-Net with padding operations modified to support using the HEALPix grid.
15
 
16
+ The default package provided for this model contains the checkpoints used
17
+ in the ECMWF AI Weather Quest S2S competition. These checkpoints are trained
18
+ with CRPS loss and use sampled random noise to produce ensemble variability
19
+ each forward pass, seeded with the `set_rng` method.
20
+
21
  For training recipies see [PhysicsNeMo](https://github.com/NVIDIA/physicsnemo/tree/main/examples/weather/dlwp_healpix), for inference [Earth2Studio](https://nvidia.github.io/earth2studio/examples/14_dlesym_example.html#sphx-glr-examples-14-dlesym-example-py).
22
 
23
  This model is ready for commercial/non-commercial use.
 
183
 
184
  NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
185
 
186
+ Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
atmos_model_0.mdlus → checkpoints/v1.0/atmos/atmos_model_0.mdlus RENAMED
File without changes
atmos_model_1.mdlus → checkpoints/v1.0/atmos/atmos_model_1.mdlus RENAMED
File without changes
atmos_model_2.mdlus → checkpoints/v1.0/atmos/atmos_model_2.mdlus RENAMED
File without changes
atmos_model_3.mdlus → checkpoints/v1.0/atmos/atmos_model_3.mdlus RENAMED
File without changes
checkpoints/v1.0/config.yaml ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ io:
2
+ atmos_variables: ["z500", "tau300-700", "z1000", "t2m", "tcwv", "t850", "z250", "ws10m"]
3
+ ocean_variables: ["sst"]
4
+ atmos_coupling_variables: ["sst"]
5
+ ocean_coupling_variables: ["z1000", "ws10m"]
6
+ atmos_diagnostic_variables: []
7
+ ocean_diagnostic_variables: []
8
+ atmos_input_times: [-18, -12, -6, 0]
9
+ ocean_input_times: [-48, 0]
10
+ atmos_output_times: [6, 12, 18, 24, 30, 36, 42, 48, 54, 60, 66, 72, 78, 84, 90, 96]
11
+ ocean_output_times: [48, 96]
12
+ atmos_constants:
13
+ - land_sea_mask
14
+ - topography
15
+ ocean_constants:
16
+ - land_sea_mask
17
+ models:
18
+ atmos_model_checkpoints:
19
+ - atmos_model_0.mdlus
20
+ - atmos_model_1.mdlus
21
+ - atmos_model_2.mdlus
22
+ - atmos_model_3.mdlus
23
+ ocean_model_checkpoints:
24
+ - ocean_model_0.mdlus
25
+ - ocean_model_1.mdlus
26
+ - ocean_model_2.mdlus
27
+ - ocean_model_3.mdlus
28
+ data:
29
+ nside: 64
30
+ scaling:
31
+ tcwv:
32
+ mean: 24.147533416748047
33
+ std: 16.7200984954834
34
+ tau300-700:
35
+ mean: 61808.91796875
36
+ std: 2609.878662109375
37
+ sst:
38
+ mean: 291.37762451171875
39
+ std: 10.202408790588379
40
+ t2m:
41
+ mean: 287.40704345703125
42
+ std: 15.39192008972168
43
+ t850:
44
+ mean: 281.1323547363281
45
+ std: 12.35011100769043
46
+ ws10m:
47
+ mean: 6.144942283630371
48
+ std: 3.6602368354797363
49
+ z1000:
50
+ mean: 937.166748046875
51
+ std: 901.9874877929688
52
+ z500:
53
+ mean: 55511.10546875
54
+ std: 2737.783447265625
55
+ z250:
56
+ mean: 103580.8984375
57
+ std: 4610.26611328125
58
+ cln:
59
+ enabled: false
ocean_model_0.mdlus → checkpoints/v1.0/ocean/ocean_model_0.mdlus RENAMED
File without changes
ocean_model_1.mdlus → checkpoints/v1.0/ocean/ocean_model_1.mdlus RENAMED
File without changes
ocean_model_2.mdlus → checkpoints/v1.0/ocean/ocean_model_2.mdlus RENAMED
File without changes
ocean_model_3.mdlus → checkpoints/v1.0/ocean/ocean_model_3.mdlus RENAMED
File without changes
checkpoints/v1.1/atmos/atmos_model_0.mdlus ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d3752ccd163c0157263f1dcdb722824cb4a2cca7368b07050172a8ba632a0950
3
+ size 20650834
checkpoints/v1.1/config.yaml ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ io:
2
+ atmos_variables: ["z500", "tau300-700", "z1000", "t2m", "tcwv", "t850", "z250", "ws10m", "ttr-3h"]
3
+ ocean_variables: ["sst"]
4
+ atmos_coupling_variables: ["sst"]
5
+ ocean_coupling_variables: ["z1000", "ws10m"]
6
+ atmos_input_times: [-18, -12, -6, 0]
7
+ ocean_input_times: [-48, 0]
8
+ atmos_output_times: [6, 12, 18, 24, 30, 36, 42, 48, 54, 60, 66, 72, 78, 84, 90, 96]
9
+ ocean_output_times: [48, 96]
10
+ atmos_constants:
11
+ - land_sea_mask
12
+ - topography
13
+ ocean_constants:
14
+ - land_sea_mask
15
+ atmos_diagnostic_variables: ["tp6", "msl"]
16
+ ocean_diagnostic_variables: []
17
+ models:
18
+ atmos_model_checkpoints:
19
+ - atmos_model_0.mdlus
20
+ ocean_model_checkpoints:
21
+ - ocean_model_0.mdlus
22
+ data:
23
+ nside: 64
24
+ scaling:
25
+ z500:
26
+ mean: 55482.71875
27
+ std: 2740.59130859375
28
+ tau300-700:
29
+ mean: 61743.29296875
30
+ std: 2610.57080078125
31
+ z1000:
32
+ mean: 936.1904907226562
33
+ std: 903.8828125
34
+ t2m:
35
+ mean: 287.40704345703125
36
+ std: 15.39192008972168
37
+ tcwv:
38
+ mean: 24.147533416748047
39
+ std: 16.7200984954834
40
+ t850:
41
+ mean: 280.5975036621094
42
+ std: 12.360532760620117
43
+ z250:
44
+ mean: 103530.8125
45
+ std: 4612.06005859375
46
+ ws10m:
47
+ mean: 6.1527934074401855
48
+ std: 3.650933265686035
49
+ ttr-3h:
50
+ mean: -2617476.9631782947
51
+ std: 494939.84
52
+ sst:
53
+ mean: 290.95892333984375
54
+ std: 10.623672485351562
55
+ tp6:
56
+ mean: 7.1346015
57
+ std: 4.5873095
58
+ msl:
59
+ mean: 101142.414
60
+ std: 1112.5691
61
+ cln:
62
+ enabled: true
63
+ condition_shape: 32
checkpoints/v1.1/ocean/ocean_model_0.mdlus ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f88299adef56d7fbfba500c2656ebba3bc13da70b150a6dbb00fed82f8715e4
3
+ size 19555332
config.yaml CHANGED
@@ -1,55 +1,18 @@
1
- io:
2
- atmos_variables: ["z500", "tau300-700", "z1000", "t2m", "tcwv", "t850", "z250", "ws10m"]
3
- ocean_variables: ["sst"]
4
- atmos_coupling_variables: ["sst"]
5
- ocean_coupling_variables: ["z1000", "ws10m"]
6
- atmos_input_times: [-18, -12, -6, 0]
7
- ocean_input_times: [-48, 0]
8
- atmos_output_times: [6, 12, 18, 24, 30, 36, 42, 48, 54, 60, 66, 72, 78, 84, 90, 96]
9
- ocean_output_times: [48, 96]
10
- atmos_constants:
11
- - land_sea_mask
12
- - topography
13
- ocean_constants:
14
- - land_sea_mask
15
- models:
16
- atmos_model_checkpoints:
17
- - atmos_model_0.mdlus
18
- - atmos_model_1.mdlus
19
- - atmos_model_2.mdlus
20
- - atmos_model_3.mdlus
21
- ocean_model_checkpoints:
22
- - ocean_model_0.mdlus
23
- - ocean_model_1.mdlus
24
- - ocean_model_2.mdlus
25
- - ocean_model_3.mdlus
26
- data:
27
- nside: 64
28
- scaling:
29
- tcwv:
30
- mean: 24.147533416748047
31
- std: 16.7200984954834
32
- tau300-700:
33
- mean: 61808.91796875
34
- std: 2609.878662109375
35
- sst:
36
- mean: 291.37762451171875
37
- std: 10.202408790588379
38
- t2m:
39
- mean: 287.40704345703125
40
- std: 15.39192008972168
41
- t850:
42
- mean: 281.1323547363281
43
- std: 12.35011100769043
44
- ws10m:
45
- mean: 6.144942283630371
46
- std: 3.6602368354797363
47
- z1000:
48
- mean: 937.166748046875
49
- std: 901.9874877929688
50
- z500:
51
- mean: 55511.10546875
52
- std: 2737.783447265625
53
- z250:
54
- mean: 103580.8984375
55
- std: 4610.26611328125
 
1
+ # default_version is the checkpoint submitted to the ECMWF AI Weather Quest
2
+ # competition (https://aiweatherquest.ecmwf.int/): conditional layer norm
3
+ # (single set of weights, sampled per forecast) + diagnostic outputs
4
+ # (tp6, msl). v1.0 is the prior ERA5 checkpoint (multiple
5
+ # checkpoint pairs for ensembling, prognostic-only output), kept for
6
+ # reproducibility and marked deprecated -- pass version="v1.0"
7
+ # explicitly to load it.
8
+ default_version: v1.1
9
+ versions:
10
+ v1.0:
11
+ path: checkpoints/v1.0
12
+ description: "Prior ERA5 checkpoint; multiple checkpoint pairs for ensembling; prognostic-only output."
13
+ deprecated: true
14
+ deprecation_message: "DLESyM checkpoint version 'v1.0' will be removed in a future release; use 'v1.1' instead."
15
+ v1.1:
16
+ path: checkpoints/v1.1
17
+ description: "AI Weather Quest submission (https://aiweatherquest.ecmwf.int/); conditional layer norm ensembling + diagnostic outputs (tp6, msl)."
18
+ deprecated: false