What problem does it solve? Running global weather forecasting models like FengWu, FourCastNet, FuXi, GraphCast, and Pangu-Weather on Hygon DCU hardware requires complex container setup, ERA5 data preprocessing, and consistent accuracy evaluation. This Skill automates the full inference and benchmarking workflow so you get standardized RMSE/ACC metrics without manual orchestration. ## Core Features & Use Cases - Multi-Model Inference: Supports five mainstream weather models (FengWu, FourCastNet, FuXi, GraphCast, Pangu-Weather) with a unified Docker-based execution flow on Hygon DCU. - Flexible Data Scenarios: Handles preprocessed ERA5 H5 data, raw NC files requiring the four-step preprocessing pipeline, or dummy data for quick pipeline validation. - Standardized Metrics Collection: Computes RMSE and ACC from .npy result files and writes them to a fixed result.json schema, with visualization outputs like loss curves and prediction comparison plots. - Use Case: You want to benchmark FuXi inference accuracy on a Hygon DCU server. The Skill guides container startup with correct device mounts, runs inference.py and result.py, and produces a result.json with per-channel and average RMSE/ACC. ## Quick Start Ask the agent to run FengWu weather model inference on Hygon DCU and collect the RMSE and ACC metrics into result.json.
