Conditional forecasting model checkpoints

README.md

@@ -1,3 +1,60 @@
 ---
 license: cc-by-4.0
+tags:
+  - physics
+  - probabilistic-forecasting
+  - generative-models
+  - pytorch
 ---
+
+# Conditional Forecasting Models
+
+This repository contains the trained model checkpoints that accompany an anonymous research submission on conditional generative forecasting from coarse-grained observations.
+
+## Models
+
+The five benchmark datasets each have Direct SI, Cascaded SI and EDM checkpoints, plus the Cascaded SI models of the depth sweep:
+
+| Dataset | Direct SI | Cascaded SI (selected depth) | Cascaded SI (other depths) | EDM |
+|---|---|---|---|---|
+| `euler` | `direct-si` | `cascaded-si-S10` | `cascaded-si-S02`, `S05`, `S20`, `S40` | `edm` |
+| `sqg` | `direct-si` | `cascaded-si-S10` | `cascaded-si-S02`, `S05`, `S20`, `S40` | `edm` |
+| `het-waves` | `direct-si` | `cascaded-si-S05` | `cascaded-si-S02`, `S10`, `S20`, `S40` | `edm` |
+| `nonlinear-kg` | `direct-si` | `cascaded-si-S20` | `cascaded-si-S02`, `S05`, `S10`, `S40` | `edm` |
+| `passive-scalar` | `direct-si` | `direct-si` (S = 1) | `cascaded-si-S02`, `S05`, `S10`, `S20`, `S40` | `edm` |
+
+S is the cascade depth. For the passive scalar the selected depth is S = 1, so its Cascaded SI model is the Direct SI model.
+
+Each directory contains:
+
+- `model.pt` — PyTorch checkpoint with the model and EMA weights, the training step and the training configuration
+- `config.yaml` — the run configuration used for sampling
+
+Sampling uses the EMA weights. The checkpoints contain only tensors and plain Python values and load with `torch.load(..., weights_only=True)`.
+
+## Datasets
+
+The corresponding datasets are `coarse-grained-forecasting-benchmark/euler`, `sqg`, `het-waves`, `nonlinear-kg` and `passive-scalar`. Each model uses the normalization of its dataset (`data/normalization.npz`).
+
+## Files and usage
+
+Download using:
+
+```bash
+hf download coarse-grained-forecasting-benchmark/forecasting-models --local-dir forecasting-models
+```
+
+With the accompanying reproduction code, pass a checkpoint with `--checkpoint`, for example:
+
+```bash
+python -m reproduce.table2 generate --system euler --method cascaded_si \
+    --data euler/data --truth-bank euler/truth_bank \
+    --checkpoint forecasting-models/euler/cascaded-si-S10/model.pt \
+    --out runs/example --gids 36 --n-samples 16 --device cpu
+```
+
+The code names the systems `euler`, `sqg`, `waves`, `nlkg` and `passive_scalar`, and the methods `direct_si`, `cascaded_si` and `edm`; add `--S <depth>` for a depth-sweep model. `SHA256SUMS` lists the file checksums.
+
+## License
+
+Creative Commons Attribution 4.0 International (CC BY 4.0).

SHA256SUMS

@@ -0,0 +1,71 @@
+9ba9550ad48438d0836ddab3da480b3b69ffa0aac7b7878b5a0039e7ab429411  LICENSE
+2ce0320c1846906673c723e840a7fec5e88b81b625789a929033e7c66b6b67ca  euler/cascaded-si-S02/config.yaml
+3b6e0173fabe9528bf63a1bb13b05454630f046f05d17a94b9fbc3467272b4a0  euler/cascaded-si-S02/model.pt
+f8dbeb6c1dcd251c87356e2cd7eee447a8808ba3737363de37f93d2ed4185049  euler/cascaded-si-S05/config.yaml
+9833a124ffaa155d569140630eb8bd91f7854c4965eb0042a437462a4f0c8407  euler/cascaded-si-S05/model.pt
+0c3b5f92d6679817b32250278b894fb8d691b1a41ed9913c99e2d88e72653f9c  euler/cascaded-si-S10/config.yaml
+d467e250984f326c17ab3daf4b6a956e47976ddd9948dc56ab87c00b96308dbb  euler/cascaded-si-S10/model.pt
+b5baeb18922a6864552626a769f89e29836974844f97ad171a30984000755f64  euler/cascaded-si-S20/config.yaml
+64edb1d25448691ecae8edb69a1090300bb30f56504ebe4ea8b60ee25103b466  euler/cascaded-si-S20/model.pt
+c58ccdcd2f14261396e7d21fb72e51a216259c7a6a915458db5aac392fc89746  euler/cascaded-si-S40/config.yaml
+3c7d254bda7e5f9d528666be727a1c2825c7bede04500212e12b560f75e842d6  euler/cascaded-si-S40/model.pt
+ef1b0ecbbd5b32869fc8ce5ab54504aa4297578d3366aaf55b9e814c74761eca  euler/direct-si/config.yaml
+d670d439ea0e36979cef8bcee46bc78fd0b47b4fcc4cc517ffe0219b2d255c32  euler/direct-si/model.pt
+dd1f369e43f8387602f29d593524299246f0b07f27b92c0a027bfc55661952d5  euler/edm/config.yaml
+3d1d855977e423cc7929d7c2c2be749b2651142bb7d0d8a1a479cee3d0aad715  euler/edm/model.pt
+b31a33fcf2e1d4b27d043fc5adad6a82f54f0be3561cf6ea1fb6eb60dddf615b  het-waves/cascaded-si-S02/config.yaml
+81ad0a0ffd6813fdf0d29049755205130592d292168db6eb405815b6752d9843  het-waves/cascaded-si-S02/model.pt
+4e26fa6a2f2f90370a864d1622d8ab2570a0fbef90ef946c8fadaadad8b49413  het-waves/cascaded-si-S05/config.yaml
+55d7268475ea02239b848aea1f0e848303bed5fce1b942db4d39d2acaafdd1a3  het-waves/cascaded-si-S05/model.pt
+64f1f29d3b7a4f81b39b637da13b0ea94348b1d10241d90847ef14e460fedb2e  het-waves/cascaded-si-S10/config.yaml
+19ae726f34a4633f2b0cff8c7e18bf77838798647ab4100d786f9c13f9cc62f0  het-waves/cascaded-si-S10/model.pt
+601b237425b7ac123889b28f24de4e3cc05c53a3aaccceb403c96c29baec6b1d  het-waves/cascaded-si-S20/config.yaml
+8af387e8de03c0137dc29786fad6b7838e3a870fa31d3051725211c56981204f  het-waves/cascaded-si-S20/model.pt
+2664ad9e84845dce34b4a67d7979656fc3edce55546f626a97f36104bc7ba0bf  het-waves/cascaded-si-S40/config.yaml
+6a01579cc242c5c0af7bff9f23263e47177e0db5a9b6dd211c99d0230899faf7  het-waves/cascaded-si-S40/model.pt
+6ade9b2c799943cb9a58d08dccec909bc2b2f213c2beb0e1fb52ef24f0b5efda  het-waves/direct-si/config.yaml
+f13c4ecaef486f3cc746dc4d1ac8c17af474e88863de88b99f2adbca34c0f449  het-waves/direct-si/model.pt
+97cc7454772646c2e26ee221d4a276d66f8be568db7246d1e46f2aeb9213c654  het-waves/edm/config.yaml
+239c3b5a212dc128d49607c7d3dd701116b5419ba8c0e58f085dace1b6b24327  het-waves/edm/model.pt
+06300f94e0cd0795580580bfd0799a1a51468d8f7583ff121e1a43443fb66f46  nonlinear-kg/cascaded-si-S02/config.yaml
+30c313ffaf84bec673566d35268f152e88dccd32686890f70bee271b79727b10  nonlinear-kg/cascaded-si-S02/model.pt
+014dfc103283625b3b6db1434e6a068ca7f93a8ea3669ca1c314c760b28546f0  nonlinear-kg/cascaded-si-S05/config.yaml
+b01eeee29892c90a77ccb156738e1d4024167f2cbd30b2f624124eb0d19342c8  nonlinear-kg/cascaded-si-S05/model.pt
+9c6f02bdca7324e23bcfc107c7b0a694a2410cd93732e28bfcfbc5248d283cd0  nonlinear-kg/cascaded-si-S10/config.yaml
+4c21f902a8a64fa6e7e03bf8aa2a5d0d1d647fd02bf0cf0d9520651bede97185  nonlinear-kg/cascaded-si-S10/model.pt
+bb61d4672b5a2e88b90ff133d04b7df328f5c9d386505606eed72d6e35d20fa2  nonlinear-kg/cascaded-si-S20/config.yaml
+546ee87205aaf7ac0b7f688abf0de10309b212781d5f4c6c5f945323fb243c4d  nonlinear-kg/cascaded-si-S20/model.pt
+b8199d9874219cd81fd64ed52a01dc3204250bf66a5b73a69aa3d30e60dafcfa  nonlinear-kg/cascaded-si-S40/config.yaml
+356ffe38e95681e7c84dec6927e5fb6da2032816ec7ab77118e4223640439bd9  nonlinear-kg/cascaded-si-S40/model.pt
+154decd6aab58df65f978156bfbba15402e4259dad2a857e1b4d008d773e1e85  nonlinear-kg/direct-si/config.yaml
+1c4ee6e6261d46f193cf328a3e5198df3e7e1bcce29ab31a91ca567737dd9646  nonlinear-kg/direct-si/model.pt
+9e437a7e03029aa12030df2ed8cf6936907c950670e6aa2bd8b5984894222278  nonlinear-kg/edm/config.yaml
+633bbcce5f6d8f39ccb29e6c674bd497cdc4337e05d51de9d2a6db90a64e5843  nonlinear-kg/edm/model.pt
+2c8f47af9c62d68dd5955f86642f94ddb706850e2f4b42c51489b685e064f3a6  passive-scalar/cascaded-si-S02/config.yaml
+7faa6a6466ddcae430d77e501e0f76f3b92836baa415dfd606425ed863493b9f  passive-scalar/cascaded-si-S02/model.pt
+7fcf8f14126729844243c2a41c4031dc99abf331eaf89d585bb2b2ef1a5b5584  passive-scalar/cascaded-si-S05/config.yaml
+41be255ea7bb0beaef16caa37d2bc408c6c5b525a6c26958916f5e27618e801b  passive-scalar/cascaded-si-S05/model.pt
+c0c64954cca1c00c34c94ac3895a2e4e894559b439fc7769a68694725ee9d6ff  passive-scalar/cascaded-si-S10/config.yaml
+f730fbbc1bfd042b771cee1500b863c191bfb3e41722a19859987683fb12180b  passive-scalar/cascaded-si-S10/model.pt
+ca7f75609ff7320757e0de21b16ef0c2a0653f8a441ea2589b97145886890263  passive-scalar/cascaded-si-S20/config.yaml
+1cbcf0ed47cc2423f20d97480c8ecdba416bab1f155c3d0bdf4a9ca2f34cdbb9  passive-scalar/cascaded-si-S20/model.pt
+1dfb63bc27e14d3f5ff5aff0de26c121ba35f469824f1a350b5000096f746937  passive-scalar/cascaded-si-S40/config.yaml
+5d3c6ff0b138298dc2c0af3c116316909d6aa10d15ae9ee85c71906aa6d5e21c  passive-scalar/cascaded-si-S40/model.pt
+f75427771f81cc67f85de633614744dbcc320887b33564f439a108ecaa8d60f5  passive-scalar/direct-si/config.yaml
+7ac8fbd0edb3658952fcafe1c8e10feec069bae40df67bdb73ec3b2c5fd27f3f  passive-scalar/direct-si/model.pt
+ca38b0ccc694a2cdf4082c1fcd897ec3f34be97408b5841455a74b93211dd34a  passive-scalar/edm/config.yaml
+c74afc28febc20af1221ed8cc580b8214e6372a9d9f67d909b94cb8365a5c7bb  passive-scalar/edm/model.pt
+b39159bbdf7a61cf2ab88899a531cc4d5786a11b97a07adde1371d7824358a8c  sqg/cascaded-si-S02/config.yaml
+7897199322040cd8e416c47d52f9c627a897ab3f61315c159c3fc21524dfbf95  sqg/cascaded-si-S02/model.pt
+e9626e04eff65f5d775a4bf68add0b9c205cc932e80bd165cc72731061067ca3  sqg/cascaded-si-S05/config.yaml
+a37023ac4e537ad368207189137a505301f460b1e745288752686cd48837be3b  sqg/cascaded-si-S05/model.pt
+bb7045d83f80a9615941e93df391d3d7c5013f8b5362a8d651add2b459021ed8  sqg/cascaded-si-S10/config.yaml
+aa94d2608a7aa21ec81cb421692ce4d8cd456d9925f428692003238f55d4cb8d  sqg/cascaded-si-S10/model.pt
+acd13a60af3238371303b8f82d2b00886de7d0f9a17f70e453a9259ef9847b37  sqg/cascaded-si-S20/config.yaml
+52cb13be41c51528ca28ba9b2e1ea00c5f8fab9e86c68812d94432a009df4883  sqg/cascaded-si-S20/model.pt
+8a76c1eda2e0387096cb5aff2dc0aad8ae32f40b6ee9fe6eb62c9a7a131487a7  sqg/cascaded-si-S40/config.yaml
+9add06517f1ff2a6f0f0a22bc52553323926f7102524ae13c5feac13a7a264ba  sqg/cascaded-si-S40/model.pt
+eeaae845658b253fcb0736a17692a5906ac065a01740fab433e10851b212f5bb  sqg/direct-si/config.yaml
+043edcf9b92836ab185e2400f38af9da51217ee9314513546c560139c833dd4b  sqg/direct-si/model.pt
+390b86e0316eab905bfbcceb40673a7863527e264a96eb64333a89415abaf58e  sqg/edm/config.yaml
+7387034ad6413ff6073a115602533645c6f31b8005ee1a1a89ea9579e6297a99  sqg/edm/model.pt

euler/cascaded-si-S02/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.5
+    - 1.0
+    n_segments: 2
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 32.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/cascaded-si-S02/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b6e0173fabe9528bf63a1bb13b05454630f046f05d17a94b9fbc3467272b4a0
+size 18578154

euler/cascaded-si-S05/config.yaml

@@ -0,0 +1,55 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.2
+    - 0.4
+    - 0.6
+    - 0.8
+    - 1.0
+    n_segments: 5
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0001
+  sigma_train: 2.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/cascaded-si-S05/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9833a124ffaa155d569140630eb8bd91f7854c4965eb0042a437462a4f0c8407
+size 18578218

euler/cascaded-si-S10/config.yaml

@@ -0,0 +1,60 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.1
+    - 0.2
+    - 0.3
+    - 0.4
+    - 0.5
+    - 0.6
+    - 0.7
+    - 0.8
+    - 0.9
+    - 1.0
+    n_segments: 10
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0003
+  sigma_train: 8.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/cascaded-si-S10/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d467e250984f326c17ab3daf4b6a956e47976ddd9948dc56ab87c00b96308dbb
+size 18578282

euler/cascaded-si-S20/config.yaml

@@ -0,0 +1,70 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.05
+    - 0.1
+    - 0.15
+    - 0.2
+    - 0.25
+    - 0.3
+    - 0.35
+    - 0.4
+    - 0.45
+    - 0.5
+    - 0.55
+    - 0.6
+    - 0.65
+    - 0.7
+    - 0.75
+    - 0.8
+    - 0.85
+    - 0.9
+    - 0.95
+    - 1.0
+    n_segments: 20
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0001
+  sigma_train: 4.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/cascaded-si-S20/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64edb1d25448691ecae8edb69a1090300bb30f56504ebe4ea8b60ee25103b466
+size 18578346

euler/cascaded-si-S40/config.yaml

@@ -0,0 +1,90 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.025
+    - 0.05
+    - 0.075
+    - 0.1
+    - 0.125
+    - 0.15
+    - 0.175
+    - 0.2
+    - 0.225
+    - 0.25
+    - 0.275
+    - 0.3
+    - 0.325
+    - 0.35
+    - 0.375
+    - 0.4
+    - 0.425
+    - 0.45
+    - 0.475
+    - 0.5
+    - 0.525
+    - 0.55
+    - 0.575
+    - 0.6
+    - 0.625
+    - 0.65
+    - 0.675
+    - 0.7
+    - 0.725
+    - 0.75
+    - 0.775
+    - 0.8
+    - 0.825
+    - 0.85
+    - 0.875
+    - 0.9
+    - 0.925
+    - 0.95
+    - 0.975
+    - 1.0
+    n_segments: 40
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 8.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/cascaded-si-S40/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c7d254bda7e5f9d528666be727a1c2825c7bede04500212e12b560f75e842d6
+size 18578538

euler/direct-si/config.yaml

@@ -0,0 +1,51 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 1.0
+    n_segments: 1
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0001
+  sigma_train: 8.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/direct-si/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d670d439ea0e36979cef8bcee46bc78fd0b47b4fcc4cc517ffe0219b2d255c32
+size 18578154

euler/edm/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: euler_v2_uniform
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: edm
+  sigma_data: 1.0
+  P_std: 1.2
+  sampling:
+    sigma_min: 0.002
+    sigma_max: 80.0
+    rho: 7
+    t_values:
+    - 0.0
+    - 1.0
+    n_steps: 20
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  lr_rampup_kimg: 0
+  data_mode: fine_endpoint_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  P_mean: -1.6
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

euler/edm/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d1d855977e423cc7929d7c2c2be749b2651142bb7d0d8a1a479cee3d0aad715
+size 18578154

het-waves/cascaded-si-S02/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.5
+    - 1.0
+    n_segments: 2
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 32.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/cascaded-si-S02/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81ad0a0ffd6813fdf0d29049755205130592d292168db6eb405815b6752d9843
+size 25042002

het-waves/cascaded-si-S05/config.yaml

@@ -0,0 +1,55 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.2
+    - 0.4
+    - 0.6
+    - 0.8
+    - 1.0
+    n_segments: 5
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.003
+  sigma_train: 2.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/cascaded-si-S05/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55d7268475ea02239b848aea1f0e848303bed5fce1b942db4d39d2acaafdd1a3
+size 25042002

het-waves/cascaded-si-S10/config.yaml

@@ -0,0 +1,60 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.1
+    - 0.2
+    - 0.3
+    - 0.4
+    - 0.5
+    - 0.6
+    - 0.7
+    - 0.8
+    - 0.9
+    - 1.0
+    n_segments: 10
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0001
+  sigma_train: 2.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/cascaded-si-S10/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19ae726f34a4633f2b0cff8c7e18bf77838798647ab4100d786f9c13f9cc62f0
+size 25042066

het-waves/cascaded-si-S20/config.yaml

@@ -0,0 +1,70 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.05
+    - 0.1
+    - 0.15
+    - 0.2
+    - 0.25
+    - 0.3
+    - 0.35
+    - 0.4
+    - 0.45
+    - 0.5
+    - 0.55
+    - 0.6
+    - 0.65
+    - 0.7
+    - 0.75
+    - 0.8
+    - 0.85
+    - 0.9
+    - 0.95
+    - 1.0
+    n_segments: 20
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0003
+  sigma_train: 16.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/cascaded-si-S20/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8af387e8de03c0137dc29786fad6b7838e3a870fa31d3051725211c56981204f
+size 25042130

het-waves/cascaded-si-S40/config.yaml

@@ -0,0 +1,90 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.025
+    - 0.05
+    - 0.075
+    - 0.1
+    - 0.125
+    - 0.15
+    - 0.175
+    - 0.2
+    - 0.225
+    - 0.25
+    - 0.275
+    - 0.3
+    - 0.325
+    - 0.35
+    - 0.375
+    - 0.4
+    - 0.425
+    - 0.45
+    - 0.475
+    - 0.5
+    - 0.525
+    - 0.55
+    - 0.575
+    - 0.6
+    - 0.625
+    - 0.65
+    - 0.675
+    - 0.7
+    - 0.725
+    - 0.75
+    - 0.775
+    - 0.8
+    - 0.825
+    - 0.85
+    - 0.875
+    - 0.9
+    - 0.925
+    - 0.95
+    - 0.975
+    - 1.0
+    n_segments: 40
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 1.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/cascaded-si-S40/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a01579cc242c5c0af7bff9f23263e47177e0db5a9b6dd211c99d0230899faf7
+size 25042322

het-waves/direct-si/config.yaml

@@ -0,0 +1,51 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 1.0
+    n_segments: 1
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 32.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/direct-si/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f13c4ecaef486f3cc746dc4d1ac8c17af474e88863de88b99f2adbca34c0f449
+size 25042002

het-waves/edm/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: heterogeneous_waves_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord_aug
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: edm
+  sigma_data: 1.0
+  P_std: 1.2
+  sampling:
+    sigma_min: 0.002
+    sigma_max: 80.0
+    rho: 7
+    t_values:
+    - 0.0
+    - 1.0
+    n_steps: 20
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  lr_rampup_kimg: 0
+  data_mode: fine_endpoint_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 3.0e-05
+  P_mean: -0.4
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

het-waves/edm/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:239c3b5a212dc128d49607c7d3dd701116b5419ba8c0e58f085dace1b6b24327
+size 25042002

nonlinear-kg/cascaded-si-S02/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 6.0
+    - 12.0
+    n_segments: 2
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 3.0e-05
+  sigma_train: 8.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/cascaded-si-S02/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30c313ffaf84bec673566d35268f152e88dccd32686890f70bee271b79727b10
+size 25032338

nonlinear-kg/cascaded-si-S05/config.yaml

@@ -0,0 +1,55 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 2.4000000000000004
+    - 4.800000000000001
+    - 7.2
+    - 9.600000000000001
+    - 12.0
+    n_segments: 5
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0001
+  sigma_train: 0.25
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/cascaded-si-S05/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b01eeee29892c90a77ccb156738e1d4024167f2cbd30b2f624124eb0d19342c8
+size 25032338

nonlinear-kg/cascaded-si-S10/config.yaml

@@ -0,0 +1,60 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 1.2000000000000002
+    - 2.4000000000000004
+    - 3.6
+    - 4.800000000000001
+    - 6.0
+    - 7.2
+    - 8.4
+    - 9.600000000000001
+    - 10.8
+    - 12.0
+    n_segments: 10
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.003
+  sigma_train: 0.5
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/cascaded-si-S10/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c21f902a8a64fa6e7e03bf8aa2a5d0d1d647fd02bf0cf0d9520651bede97185
+size 25032402

nonlinear-kg/cascaded-si-S20/config.yaml

@@ -0,0 +1,70 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.6000000000000001
+    - 1.2000000000000002
+    - 1.8
+    - 2.4000000000000004
+    - 3.0
+    - 3.6
+    - 4.2
+    - 4.800000000000001
+    - 5.4
+    - 6.0
+    - 6.6000000000000005
+    - 7.2
+    - 7.800000000000001
+    - 8.4
+    - 9.0
+    - 9.600000000000001
+    - 10.200000000000001
+    - 10.8
+    - 11.4
+    - 12.0
+    n_segments: 20
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0001
+  sigma_train: 1.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/cascaded-si-S20/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:546ee87205aaf7ac0b7f688abf0de10309b212781d5f4c6c5f945323fb243c4d
+size 25032466

nonlinear-kg/cascaded-si-S40/config.yaml

@@ -0,0 +1,90 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.30000000000000004
+    - 0.6000000000000001
+    - 0.9
+    - 1.2000000000000002
+    - 1.5
+    - 1.8
+    - 2.1
+    - 2.4000000000000004
+    - 2.7
+    - 3.0
+    - 3.3000000000000003
+    - 3.6
+    - 3.9000000000000004
+    - 4.2
+    - 4.5
+    - 4.800000000000001
+    - 5.1000000000000005
+    - 5.4
+    - 5.7
+    - 6.0
+    - 6.300000000000001
+    - 6.6000000000000005
+    - 6.9
+    - 7.2
+    - 7.5
+    - 7.800000000000001
+    - 8.1
+    - 8.4
+    - 8.700000000000001
+    - 9.0
+    - 9.3
+    - 9.600000000000001
+    - 9.9
+    - 10.200000000000001
+    - 10.5
+    - 10.8
+    - 11.100000000000001
+    - 11.4
+    - 11.700000000000001
+    - 12.0
+    n_segments: 40
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0003
+  sigma_train: 2.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/cascaded-si-S40/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:356ffe38e95681e7c84dec6927e5fb6da2032816ec7ab77118e4223640439bd9
+size 25032658

nonlinear-kg/direct-si/config.yaml

@@ -0,0 +1,51 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 12.0
+    n_segments: 1
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 128.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/direct-si/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c4ee6e6261d46f193cf328a3e5198df3e7e1bcce29ab31a91ca567737dd9646
+size 25032338

nonlinear-kg/edm/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: nlkg_v3_candidate
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_2comp
+  D: 392
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 12.0
+method:
+  name: edm
+  sigma_data: 1.0
+  P_std: 1.2
+  sampling:
+    sigma_min: 0.002
+    sigma_max: 80.0
+    rho: 7
+    t_values:
+    - 0.0
+    - 12.0
+    n_steps: 20
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  lr_rampup_kimg: 0
+  data_mode: fine_endpoint_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  P_mean: -4.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

nonlinear-kg/edm/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:633bbcce5f6d8f39ccb29e6c674bd497cdc4337e05d51de9d2a6db90a64e5843
+size 25032274

passive-scalar/cascaded-si-S02/config.yaml

@@ -0,0 +1,52 @@
+data:
+  name: passive_scalar_v1
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.5
+    - 1.0
+    n_segments: 2
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.0003
+  sigma_train: 4.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

passive-scalar/cascaded-si-S02/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7faa6a6466ddcae430d77e501e0f76f3b92836baa415dfd606425ed863493b9f
+size 18587370

passive-scalar/cascaded-si-S05/config.yaml

@@ -0,0 +1,55 @@
+data:
+  name: passive_scalar_v1
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.2
+    - 0.4
+    - 0.6
+    - 0.8
+    - 1.0
+    n_segments: 5
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 32.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

passive-scalar/cascaded-si-S05/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41be255ea7bb0beaef16caa37d2bc408c6c5b525a6c26958916f5e27618e801b
+size 18587434

passive-scalar/cascaded-si-S10/config.yaml

@@ -0,0 +1,60 @@
+data:
+  name: passive_scalar_v1
+  packaged_dataset_path: <DATASET>/
+  train_ics: 80
+  train_realizations: 200
+model:
+  name: unet2_5ch_coord
+  D: 196
+  base_channels: 64
+  n_grid: 32
+  k_c: 8
+  t_start: 0.0
+  t_end: 1.0
+method:
+  name: cascaded_si
+  t_eps: 0.0
+  beta_power: 1
+  max_grad_norm: 1.0
+  sampling:
+    noise_scale: 1.0
+    t_values:
+    - 0.0
+    - 0.1
+    - 0.2
+    - 0.3
+    - 0.4
+    - 0.5
+    - 0.6
+    - 0.7000000000000001
+    - 0.8
+    - 0.9
+    - 1.0
+    n_segments: 10
+    n_sde_steps_per_seg: 20
+  eval_noise_scales:
+  - 1.0
+  data_mode: fine_segment_pairs
+  fine_val_members: 20
+  fine_horizon_idx: 40
+  lr: 0.001
+  sigma_train: 32.0
+training:
+  batch_size: 256
+  max_steps: 30000
+  val_every_updates: 1000
+  val_extra_steps:
+  - 0
+  - 100
+  - 200
+  - 300
+  - 500
+  - 750
+  log_loss_every_epochs: 1
+  ema_halflife_kimg: 500
+  ema_rampup_ratio: 0.05
+  allow_tf32: false
+  deterministic: true
+  num_workers: 0
+  val_seed: 777
+seed: 0

passive-scalar/cascaded-si-S10/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f730fbbc1bfd042b771cee1500b863c191bfb3e41722a19859987683fb12180b
+size 18587434