mabera/nigeria-carbon-emissions-interpreter

Nigeria Carbon Emissions & Economic Trajectory Interpreter

Author: Hussein Adeiza (mabera) Role: Licensed Environmental Health Officer, Abuja Nigeria Base Model: Mixtral 8x7B Fine-tuned with: AutoScientist by Adaption Labs

⚠️ Weights Upload Pending

This model's weights are not yet uploaded due to a confirmed platform-side issue: the AutoScientist download endpoint returns a 161GB full fine-tuned model (not a LoRA adapter) with no support for resumable downloads (HTTP Range requests are ignored, always returning the full file) and 1-hour token expiry, making the file effectively undownloadable at typical connection speeds. This was reported to and confirmed by the Adaption team, who are investigating. Weights will be added once resolved. All training metrics below are confirmed from the AutoScientist dashboard.

Model Description

A model fine-tuned to interpret Nigeria's carbon emissions and economic trajectory data, producing structured analytical reasoning grounded entirely in real, directly-downloaded statistics from Our World in Data.

Training Method: Full Fine-Tuning, Not LoRA

Unlike every other submission in this author's portfolio, this job trained via full fine-tuning rather than LoRA. This was not a deliberate selection, the training configuration screen defaulted to this setting without an explicit choice being made, consistent with the platform's default behavior on this job. Confirmed via direct correspondence with the Adaption team, who are looking into why.

Training Data

• Source: Our World in Data, owid/co2-data (GitHub), downloaded directly
• Dataset: 5 original prompt-completion pairs, every number computed via pandas directly from the raw source file, expanded via Adaptive Data (Hallucination Mitigation, full 20K+ datapoint expansion)
• Kaggle: https://www.kaggle.com/datasets/yunusahusseinadeiza/nigeria-carbon-emissions-interpreter

Training Metrics

• Win rate (on dataset): 87% adapted vs 14% base model
• General Win Rate (Science domain): 79% adapted vs 21% base
• Base model: mistralai/Mixtral-8x7B-Instruct-v0.1
• Method: Full fine-tuning (platform default, not selected), Hallucination Mitigation, full 20,000+ datapoint expansion
• Dataset quality: 8.0 → 8.9 (+11.3% relative improvement, Grade B)
• Percentile: 31.5
• Domain classification: Science (60%) / Data-analysis-visualization (40%)

Key Cited Findings (from the raw downloaded source only)

• Nigeria's total CO2 emissions rose 40.3% since 2000 (96.8 to 135.8 million tonnes), while per-capita emissions fell 23.8% (0.766 to 0.584 tonnes), population growth outpacing emissions growth
• South Africa's per-capita emissions are 11.7x Nigeria's, driven overwhelmingly by coal (83% of South Africa's total emissions)
• Nigeria's carbon intensity of GDP (0.116 kg CO2/$) is lower than both South Africa's (5.1x higher) and the World average (2.5x higher), despite Nigeria's status as a major oil producer, reflecting an export-oriented rather than domestically carbon-intensive economy
• Nigeria is a net importer of embodied carbon in trade, while South Africa is a large net exporter, consistent with South Africa's heavy industrial and mining export sector

Credits

Powered by Adaptive Data — Adaption Labs AutoScientist Challenge 2026, Part 2 — Market Analysis & News Category