Feilian-3D
Feilian-3D is a deep learning project for 3D wind speed field prediction using U-Net architecture, optimized for HPC environments with AMD ROCm GPUs.
Overview
Feilian-3D leverages deep learning techniques to predict three-dimensional wind speed fields, making it particularly useful for applications in atmospheric modeling, renewable energy forecasting, and climate science.
Key Features
- 3D U-Net Architecture - Encoder-decoder structure with skip connections optimized for volumetric data
- AMD ROCm Support - Optimized for AMD MI250X GPUs with ROCm 6.3.3
- HPC Ready - Distributed training with PyTorch DDP and SLURM integration
- Mixed Precision Training - Automatic mixed precision for faster training and lower memory usage
- Physics-Informed Loss - Gradient loss functions for physical plausibility
- Flexible Configuration - Comprehensive config system for experiments and hyperparameter tuning
Documentation Sections
Installation
Setup guide for Setonix HPC and generic Linux systems
Quickstart
Get started with training your first model
Architecture
Deep dive into the 3D U-Net model architecture
Training
Distributed training and optimization techniques
SLURM Configuration
Running jobs on HPC clusters with SLURM
API Reference
Complete API documentation for all modules