You can deploy this model and start earning money today!

mabera/
nigeria-fertilizer-access-model

meta-llama/Llama-3.3-70B-Instruct

finetune

apache-2.0

Views

No views yet

Nigeria Fertilizer Access Model

Author: Hussein Adeiza (mabera) Role: Licensed Environmental Health Officer, Abuja Nigeria Base Model: Llama 3.3 70B Fine-tuned with: AutoScientist by Adaption Labs

Model Description

A LoRA adapter fine-tuned to interpret raw Nigerian fertilizer market statistics, price data, subsidy structure, yield gaps, and import dependency, and produce structured economic analytical reasoning. Raw statistics go in, expert interpretation comes out, grounded strictly in cited source data.

Training Data

Source: IFPRI/Food Security Portal (Kirui et al. 2026), IFDC, AfDB, NSIA, Nexdigm, ScienceDirect, FEPSAN
Dataset: 5 original cited prompt-completion pairs, expanded via Adaptive Data (House Special + Hallucination mitigation)
Languages: English, Hausa, Yoruba
Kaggle: https://www.kaggle.com/datasets/yunusahusseinadeiza/nigeria-fertilizer-access-interpreter

Training Metrics

Win rate (on dataset): 62% adapted vs 38% base model
General Win Rate (unseen Agriculture-domain tasks): 75% adapted vs 25% base
Base model: meta-llama/Llama-3.3-70B-Instruct
Method: LoRA — House Special + Hallucination mitigation
Dataset quality: 8.0 → 8.0 (0% relative improvement, Grade B → B)

Honest Disclosure

This submission's Adaptive Data quality score did not improve during adaptation (8.0 to 8.0, Grade B held steady), the first time in this portfolio this has happened. The win rates remain solid (62% training, 75% General Win Rate), so the model itself performs well, but the quality metric specifically did not move. A plausible contributing factor: this run used a smaller datapoint expansion (1,011 rows) than prior submissions (20,000+ rows on Oil Spill and Desertification), though this has not been confirmed as the cause. Disclosed transparently rather than only reporting the stronger win-rate figures.

Key Cited Findings (from original source data only)

Nigeria urea prices rose 43% (₦35,000 to ₦50,000 per 50kg bag) between February and May 2025, despite weak planting-season demand, indicating the driver was upstream/FX rather than domestic market pressure
Maize yields average 1.2 tons/ha in poorly fertilized northern regions versus 3.5 tons/ha in better-input southern regions, a threefold gap; over 45% of smallholders cut fertilizer purchases in 2022-2023 citing cost, not availability, as the barrier
The Presidential Fertilizer Initiative's 2016 subsidized price (₦5,500 vs ₦10,000 market) created a documented arbitrage incentive for middleman diversion; NAFDAC and SON have found adulteration with sand and sawdust fillers
Nigeria has zero domestic potash or phosphate production and remains fully import-dependent for two of three core NPK components, even as urea production grew nearly 40x (70,000 to 4.2 million MT, 2010-2025)

mabera/nigeria-fertilizer-access-model