Mvula

Compressing ECMWF’s AIFS for Edge Deployment

The LapAI-Forecast project, Mvula, compresses ECMWF’s AIFS model into a lightweight, laptop-deployable forecasting tool. It targets reliable 10-day global forecasts at 1° resolution through direct architectural compression techniques such as grid coarsening, attention head pruning, LoRA and quantization, with evaluation focused on African weather and climate conditions.

Mentors

Mario Santa Cruz López
Rendani Mbuvha
Shruti Nath

Participants

Chimwemwe Chanda

Mthetho Sovara

Samuel Mathekga

Gabriel Elim

Fima Sichone