Downloads · 30 days
0
mvula-v5-student
C4E-Mvula/mvula-v5-student
mvula-v5-student is a other model from C4E-Mvula. Use it for the other task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as apache-2.0.
Programme: ECMWF Code for Earth 2026 — African Stream (Challenge 40) Code: github.com/msovara/lapai-forecast-africa Freeze tag: trackb-v5-c4e Checkpoint file: studentglobalstablev5.ckpt (~8.8 MiB, ~2.17 M parameters)
Access
Public
Updated Sep 22, 2026
Repo size
9.2 MB
Likes
1
Public
Repo makeup
Click a slice to open those files.
.ckpt9.2 MB · 100%
At a glance
- Task
- other
- Library
- pytorch
- License
- apache-2.0
- Access
- Public
- Created
- Sep 1, 2026
- Updated
- Sep 22, 2026
- SHA
- fc88d6c7
Spaces
Model card
From the Hugging Face model README
Mvula v5 — AIFS-distilled student (African short-range t2m)
Programme: ECMWF Code for Earth 2026 — African Stream (Challenge 40)
Code: github.com/msovara/lapai-forecast-africa
Freeze tag: trackb-v5-c4e
Checkpoint file: student_global_stable_v5.ckpt (~8.8 MiB, ~2.17 M parameters)
What this model is
A small CNN student distilled from a pruned AIFS-family teacher (K1 / CREDIT-style Track B). It is intended for laptop-scale experimentation on short-range African 2 m temperature under an analysis-forced protocol.
It is not a free-running 10-day global NWP emulator and not an operational NMHS forecast system.
Claim boundary
| Supported | Not supported / not claimed |
|---|---|
Cin=65 → Cout=3 (tp, msl, 2t) | Autonomous free-run / autoregressive / 10-day rollout |
| Analysis-forced one-step African t2m (headline = 00Z IC) | Full multi-variable Week-9 free-run table |
| CPU inference (~2.5 s / +6 h step on i7-11800H) | Finished ONNX / INT8 product (optional later) |
| Open eval artefacts in the GitHub repo | LoRA / ENACTS country adapters · smartphone-ready |
Precipitation (tp) for this head failed (dry-collapse) and is out of scope.
Packaged skill (Africa, 61 × 00Z inits, 2023)
Analysis-forced one-step only. Table columns are analysis IC hours (00Z campaign), not AR lead times.
| Analysis IC | RMSE (°C) | ACC | Bias (°C) |
|---|---|---|---|
| 00Z (headline) | 1.38 | 0.97 | −0.12 |
| 06Z | 7.80 | 0.45 | −5.33 (systematic cold) |
| 12Z | 4.38 | 0.75 | −1.12 |
| 18Z | 3.23 | 0.84 | +1.30 |
Full tables, maps, and draft paper: see the GitHub reports/ folder (including DRAFT_PAPER_MVULA_V5.pdf).
How to load (PyTorch)
import torch
from huggingface_hub import hf_hub_download
# After you upload this checkpoint to your model repo:
ckpt_path = hf_hub_download(
repo_id="C4E-Mvula/mvula-v5-student",
filename="student_global_stable_v5.ckpt",
)
blob = torch.load(ckpt_path, map_location="cpu", weights_only=False)
# Prefer the GitHub package for the CNN class:
# pip install "git+https://github.com/msovara/lapai-forecast-africa.git"
# from lapai_inference.model import LapAIStudentCNN, LapAIStudentConfig
End-user paths on GitHub: python run_mvula.py info|bench|dashboard or Apptainer — see repo README.
Intended users
NMHS / university experimenters, students, and AfriClimate AI community members who want a small, inspectable African temperature demo on a laptop or in a Hugging Face Space — not a replacement for official forecasts.
Licence
- Code / this card: Apache-2.0
- Student weights: research and evaluation use (see GitHub
NOTICE) - Teacher (AIFS / Anemoi) weights: remain under their original terms — not redistributed here
Citation / credit
Team Mvula (Code for Earth 2026). Mentors: Shruti Nath, Rendani Mbuvha, Mario Santa Cruz López.
GPU training/eval support: Cassava AI Factory. HPC: CHPC Lengau.
- Task
- other
- Library
- pytorch
- License
- apache-2.0
- Created
- Sep 1, 2026
- Updated
- Sep 22, 2026