Logo Lanfrica
  • Home
  • Atlas
  • Insights
  • Docs
  • Sign in

© 2026 Lanfrica. All rights reserved. All copyrights of the resources shown on the Lanfrica website belong to the original copyright holders, unless explicitly stated otherwise.

kherin/karoguard-adtc-2026-submission

Domain:

agricultureclimate

Record type:

model
Creator:
khe
Host:
KaroGuard is a compact agricultural decision-support model for smallholder farmers in Mauritius and other cyclone-exposed African and Indian Ocean communities. # KaroGuard — Offline Cyclone-Resilience Assistant KaroGuard is a compact agricultural decision-support model for smallholder farmers in Mauritius and other cyclone-exposed African and Indian Ocean communities. It runs locally through llama.cpp after the model is downloaded. It helps organize pre-cyclone preparation, post-cyclone observations and recovery actions while keeping human safety ahead of crops, equipment and access. ## Quick start bash download_model.sh llama-cli -m model/KaroGuard-Q4_K_M.gguf -p "A cyclone warning is active. What should a tomato farmer do first?" The model download is credential-free and the model does not require network access during inference. The hosted model artifact is available from the public KaroGuard Hugging Face repository. The included `download_model.sh` script downloads that exact artifact and verifies its SHA-256 checksum before use. ## Model - Base: Qwen/Qwen3-4B-Instruct-2507 - Runtime: llama.cpp - Quantization: GGUF Q4_K_M - Artifact SHA-256: 5ea2b969bd067f96fc9a26cdd4ed749e3a4f23b4838fcd6293be243301af6b76 ## Submission validation From a fresh public clone, run the exact ADTC participant smoke test: bash download_model.sh adtc-profiler run --submission . --mode participant --output submission.json --skip-accuracy The committed `submission.json` was regenerated from a clean clone with `adtc-profiler 0.1.0`. It reports `environment.measured_on` as `participant_laptop`, confirms the claimed 4B parameter count, and records no thermal throttling. The model download requires no credentials; after the download, profiling and inference use the local GGUF without network access. ## Safety KaroGuard is informational decision support. Follow current official warnings and local authorities. Do not enter floodwater, approach fallen electrical lines, or delay evacuation to protect crops. The model does not issue weather warnings, predict cyclone tracks, diagnose from images, prescribe regulated chemicals, or replace em …

Visit

github.com

Tasks

natural language generation

Languages

Sha

Tags

adtcadtc-2026aiai-modeldevpostdevpost-hackathonmodel

Licenses

GPL-3.0

Similar

ArtTechnologies-User/adtc-2026-submission-templateICON-XB/adtc-2026-agrimate-submissionGSS-creator/adtc-2026-submission-G1-360MWayazi/adtc-2026NourTi/rifqa-adtc-2026ifenium/farmgate-adtc-2026

ArtTechnologies-User/adtc-2026-submission-template

Official submission template for the Africa Deep Tech Challenge 2026 Laptop LLM track # ADTC 2026 —

ICON-XB/adtc-2026-agrimate-submission

AgriMate, an offline-first agricultural AI assistant for resource-constrained African environments,

GSS-creator/adtc-2026-submission-G1-360M

G1-360M V18.2 is a compact, production-ready enterprise AI model built by Gaston Software Solutions

Wayazi/adtc-2026

ADTC 2026 — On-Device LLM Challenge: Qwen3.5-2B fine-tuned for healthcare advisory (isiZulu, KZN) wi

NourTi/rifqa-adtc-2026

Rifqa: offline English tutoring AI for Algerian students with Arabic explanations (ADTC 2026 submiss

ifenium/farmgate-adtc-2026

Offline Qwen3 1.7B agricultural market-price assistant for ADTC 2026 (Africa Deep Tech Challenge)