File size: 2,621 Bytes
ae7108c
 
 
 
 
 
 
 
 
 
 
 
 
 
eb8ddcd
 
 
ae7108c
 
eb8ddcd
ae7108c
 
 
 
4600f83
793feba
ae7108c
 
 
 
 
 
 
 
 
793feba
ae7108c
 
 
 
 
eb8ddcd
 
9c1b034
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
language: en
tags:
- weather-forecasting
- diffusion-models
- rectified-flow
- meteorology
- pytorch
- deep-learning
license: mit
datasets:
- meteolibre
---




# MeteoLibre Rectified Flow Model

In the folder models_shortcut/:

## Model Description

- **Model type**: Rectified Flow Diffusion Model
- **Architecture**: 3D U-Net with FiLM conditioning
- **Input**: Meteorological data patches (12 channels + 1 lightning channels, 3D spatio-temporal)
- **Output**: Generated weather forecast data
- **Training data**: MeteoLibre meteorological dataset
- **Language(s)**: Python
- **License**: MIT

## Training

The model was trained using:
- **Framework**: PyTorch with Hugging Face Accelerate
- **Optimizer**: Adam (lr=5e-4) OR SOAP
- **Batch size**: 64
- **Epochs**: 200
- **Precision**: Mixed precision (bf16)
- **Distributed training**: Multi-GPU support

And there is different video exemple for the inference.

Performance summary for the first wave of shortcut model:

## Performance Summary

| Model          | Optimizer | Steps  | sat_mse | sat_psnr | sat_ssim | light_mae | light_precision | light_recall | light_f1 | light_iou |
|----------------|-----------|--------|---------|----------|----------|-----------|-----------------|--------------|----------|-----------|
| RF (Run 1)     | -         | 128    | 0.0952  | 28.5327  | 0.8042   | 0.0221    | 0.5482          | 0.6535       | 0.5950   | -         |
| RF (Run 2)     | -         | 128    | 0.1076  | 27.8870  | 0.8000   | 0.0221    | 0.5157          | 0.6454       | 0.5724   | -         |
| Baseline       | Persistence| baseline| 0.2368 | 24.5138  | 0.7266   | 0.0154    | 0.6714          | 0.6665       | 0.6678   | 0.1023    |
| Shortcut       | Adam      | 16     | 0.0981  | 28.3788  | 0.8106   | 0.0216    | 0.6339          | 0.5192       | 0.5686   | 0.0791    |
| Shortcut       | Adam      | 64     | 0.0983  | 28.3702  | 0.8114   | 0.0207    | 0.6609          | 0.5304       | 0.5860   | 0.0791    |
| Shortcut       | Adam      | 128    | 0.0983  | 28.3581  | 0.8112   | 0.0208    | 0.6518          | 0.5208       | 0.5769   | 0.0791    |
| Shortcut       | SOAP      | 16     | 0.0601  | 30.5008  | 0.8663   | 0.0156    | 0.8654          | 0.6958       | 0.7710   | 0.0818    |
| Shortcut       | SOAP      | 64     | 0.0606  | 30.4786  | 0.8661   | 0.0151    | 0.8658          | 0.6879       | 0.7663   | 0.0818    |
| Shortcut       | SOAP      | 128    | 0.0605  | 30.4848  | 0.8660   | 0.0151    | 0.8635          | 0.6886       | 0.7656   | 0.0818    |

Metrics from evaluation on 64x20 elements (satellite and lightning data).