Hugging Face's logo

Datasets:
montefiore-sail
/
appa

Languages:
English
Size:
n>1T
License:
Dataset card Data Studio Files
xet
Community
iSach commited on
Commit
9c9c7c9
·
verified ·
1 Parent(s): 97d26b3

Upload folder using huggingface_hub

Browse files
Files changed (17) hide show
  1. README.md +5 -3
  2. autoencoders/workshop/0/config.yaml +86 -0
  3. autoencoders/workshop/0/latents/workshop/ae/config.yaml +86 -0
  4. autoencoders/workshop/0/latents/workshop/ae/model.pth +3 -0
  5. autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/config.yaml +56 -0
  6. autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/metadata.yaml +4 -0
  7. autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/model_best.pth +3 -0
  8. autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/model_last.pth +3 -0
  9. autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/optimizer_last.pth +3 -0
  10. autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/scheduler_last.pth +3 -0
  11. autoencoders/workshop/0/latents/workshop/era5_1993-2021-1h-1440x721.h5 +3 -0
  12. autoencoders/workshop/0/latents/workshop/stats.pth +3 -0
  13. autoencoders/workshop/0/metadata.yaml +4 -0
  14. autoencoders/workshop/0/model_best.pth +3 -0
  15. autoencoders/workshop/0/model_last.pth +3 -0
  16. autoencoders/workshop/0/optimizer_last.pth +3 -0
  17. autoencoders/workshop/0/scheduler_last.pth +3 -0
README.md CHANGED
@@ -1,3 +1,5 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
1
+ # Appa
2
+
3
+ This repository contains data and model checkpoints for Appa: Bending Weather Dynamics with Latent Diffusion Models for Global Data Assimilation by Gérôme Andry, Sacha Lewin, François Rozet, Omer Rochman, Victor Mangeleer, Matthias Pirlet, Elise Faulx, Marilaure Grégoire, and Gilles Louppe.
4
+
5
+ Please check out the [Wiki](https://github.com/montefiore-sail/appa/wiki) for more information.
autoencoders/workshop/0/config.yaml ADDED
@@ -0,0 +1,86 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ae:
2
+ name: cae_f32c128
3
+ shape:
4
+ - 721
5
+ - 1440
6
+ resize: 32
7
+ saturation: softclip2
8
+ noise_level: 0.01
9
+ lat_channels: 128
10
+ hid_channels:
11
+ - 128
12
+ - 256
13
+ - 384
14
+ - 512
15
+ - 768
16
+ - 1024
17
+ hid_blocks:
18
+ - 3
19
+ - 3
20
+ - 3
21
+ - 3
22
+ - 3
23
+ - 3
24
+ pixel_shuffle: true
25
+ ffn_factor: 1
26
+ dropout: 0.0
27
+ patch_size:
28
+ - 1
29
+ - 1
30
+ checkpointing: true
31
+ identity_init: true
32
+ in_channels: 71
33
+ context_channels: 1
34
+ latent_shape:
35
+ - 23
36
+ - 47
37
+ - 128
38
+ optim:
39
+ name: soap_3e-05_linear
40
+ optimizer: soap
41
+ betas:
42
+ - 0.9
43
+ - 0.99
44
+ - 0.99
45
+ precondition_frequency: 16
46
+ precondition_size: 4096
47
+ merge_dims: false
48
+ learning_rate: 3.0e-05
49
+ weight_decay: 0.0
50
+ warmup: 0
51
+ scheduler: linear
52
+ grad_clip: 1
53
+ train:
54
+ update_steps: 200000
55
+ batch_element_per_gpu: 1
56
+ batch_size_per_step: 64
57
+ precision: float16
58
+ log_interval: 1024
59
+ save_interval: 8192
60
+ shuffle_training: true
61
+ trajectory_size: 1
62
+ sub_pressure_levels: true
63
+ valid:
64
+ log_interval: 16384
65
+ n_valid_samples: 64
66
+ n_valid_snapshots: 6
67
+ loss:
68
+ error: mse
69
+ latitude_weighting: true
70
+ level_weighting: true
71
+ l2_weight: 0.0
72
+ num_model_params: 341131207
73
+ path: !!python/object/apply:pathlib.PosixPath
74
+ - /
75
+ - gpfs
76
+ - scratch
77
+ - acad
78
+ - sail
79
+ - appa
80
+ - autoencoders
81
+ - fvjfblu7
82
+ - '0'
83
+ hardware:
84
+ num_nodes: 16
85
+ num_gpus: 64
86
+ num_cpus: 512
autoencoders/workshop/0/latents/workshop/ae/config.yaml ADDED
@@ -0,0 +1,86 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ae:
2
+ name: cae_f32c128
3
+ shape:
4
+ - 721
5
+ - 1440
6
+ resize: 32
7
+ saturation: softclip2
8
+ noise_level: 0.01
9
+ lat_channels: 128
10
+ hid_channels:
11
+ - 128
12
+ - 256
13
+ - 384
14
+ - 512
15
+ - 768
16
+ - 1024
17
+ hid_blocks:
18
+ - 3
19
+ - 3
20
+ - 3
21
+ - 3
22
+ - 3
23
+ - 3
24
+ pixel_shuffle: true
25
+ ffn_factor: 1
26
+ dropout: 0.0
27
+ patch_size:
28
+ - 1
29
+ - 1
30
+ checkpointing: true
31
+ identity_init: true
32
+ in_channels: 71
33
+ context_channels: 1
34
+ latent_shape:
35
+ - 23
36
+ - 47
37
+ - 128
38
+ optim:
39
+ name: soap_3e-05_linear
40
+ optimizer: soap
41
+ betas:
42
+ - 0.9
43
+ - 0.99
44
+ - 0.99
45
+ precondition_frequency: 16
46
+ precondition_size: 4096
47
+ merge_dims: false
48
+ learning_rate: 3.0e-05
49
+ weight_decay: 0.0
50
+ warmup: 0
51
+ scheduler: linear
52
+ grad_clip: 1
53
+ train:
54
+ update_steps: 200000
55
+ batch_element_per_gpu: 1
56
+ batch_size_per_step: 64
57
+ precision: float16
58
+ log_interval: 1024
59
+ save_interval: 8192
60
+ shuffle_training: true
61
+ trajectory_size: 1
62
+ sub_pressure_levels: true
63
+ valid:
64
+ log_interval: 16384
65
+ n_valid_samples: 64
66
+ n_valid_snapshots: 6
67
+ loss:
68
+ error: mse
69
+ latitude_weighting: true
70
+ level_weighting: true
71
+ l2_weight: 0.0
72
+ num_model_params: 341131207
73
+ path: !!python/object/apply:pathlib.PosixPath
74
+ - /
75
+ - gpfs
76
+ - scratch
77
+ - acad
78
+ - sail
79
+ - appa
80
+ - autoencoders
81
+ - fvjfblu7
82
+ - '0'
83
+ hardware:
84
+ num_nodes: 16
85
+ num_gpus: 64
86
+ num_cpus: 512
autoencoders/workshop/0/latents/workshop/ae/model.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c8418a17cd3d7a54480d19ab6c627d54f30818504b48acb1b2ffea66792e0ca9
3
+ size 1364606478
autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/config.yaml ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ backbone:
2
+ name: vit_large
3
+ arch: vit
4
+ mod_features: 256
5
+ hid_channels: 1024
6
+ hid_blocks: 16
7
+ ffn_factor: 4
8
+ attention_heads: 8
9
+ qk_norm: true
10
+ rope: true
11
+ patch_size:
12
+ - 2
13
+ - 1
14
+ - 1
15
+ window_size: null
16
+ dropout: 0.0
17
+ checkpointing: true
18
+ blanket_size: 24
19
+ optim:
20
+ name: adamw_0.0001_None
21
+ optimizer: adamw
22
+ betas:
23
+ - 0.9
24
+ - 0.99
25
+ learning_rate: 0.0001
26
+ weight_decay: 0.0
27
+ warmup: 0
28
+ scheduler: null
29
+ grad_clip: 1
30
+ train:
31
+ update_steps: 500000
32
+ batch_element_per_gpu: 4
33
+ batch_size_per_step: 256
34
+ precision: float16
35
+ log_interval: 1024
36
+ save_interval: 8192
37
+ shuffle_training: true
38
+ standardize: true
39
+ blanket_dt: 1
40
+ blanket_size: 24
41
+ noise_schedule: rectified
42
+ sigma_min: 0.001
43
+ sigma_max: 1000.0
44
+ inject_ae_noise: true
45
+ loss:
46
+ sigma_min_penalty: 0
47
+ valid:
48
+ log_interval: 16384
49
+ n_valid_samples: 32
50
+ n_valid_dates: 2
51
+ n_valid_samples_per_date: 1
52
+ n_valid_channels: 3
53
+ denoising_steps: 64
54
+ ae_run: fvjfblu7/0
55
+ latent_dump: full
56
+ latent_data: era5_1993-2021-1h-1440x721-gencast
autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/metadata.yaml ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ last_step_done: 123199
2
+ best_val_loss: 0.2735000550746918
3
+ lap: 2
4
+ runid: 2gnjx06l
autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/model_best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a21edd8ae444dc02cf799ad4c804424a877d3b1babd81874c5573901fec8d6e
3
+ size 892932934
autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/model_last.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:93b397268695f8b1991a33c147eb9d622a46e07f93f93fd7492c13ecb8cb8c96
3
+ size 892932934
autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/optimizer_last.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:553a2995649b9b9be0e560d89b9b637a34c3d015dd421dfd983e706e04a4ac76
3
+ size 1785882408
autoencoders/workshop/0/latents/workshop/denoisers/workshop/0/scheduler_last.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54c170a66d16d3c4e5a7d767bbe6b0a86f63898c45ef0fad586e74bd28d7d66a
3
+ size 1020
autoencoders/workshop/0/latents/workshop/era5_1993-2021-1h-1440x721.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5aa9ebb778ed07dc474135ce8ccebc30fe7e3189e5fdc1c3b00ba52ede22a4d8
3
+ size 140701079552
autoencoders/workshop/0/latents/workshop/stats.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7dccfb4f0918b314566742d8be366bceaff73207489d909427149c60b969398
3
+ size 2316
autoencoders/workshop/0/metadata.yaml ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ last_step_done: 93055
2
+ best_val_loss: 0.0004049637063872069
3
+ lap: 0
4
+ runid: fvjfblu7
autoencoders/workshop/0/model_best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0eeddb8e4c0fd3d1865392e7dab05a0e2dbd3844fd5f395f54f37f357b79387d
3
+ size 1364606478
autoencoders/workshop/0/model_last.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c8418a17cd3d7a54480d19ab6c627d54f30818504b48acb1b2ffea66792e0ca9
3
+ size 1364606478
autoencoders/workshop/0/optimizer_last.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8c6196107f32dae3ce3373e2d1a4cd9bb317cd69339bf83e88956108066da9a4
3
+ size 3466330134
autoencoders/workshop/0/scheduler_last.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:413cabda06067d5702172eb0ef49ae6b26e3850cc8ec373ed9b28140d4a4ca03
3
+ size 1020