Janeodum/tsaro-e2b
Janeodum/tsaro-e2b is a 5.1 billion parameter Gemma 4 E2B model, fine-tuned by Janeodum for threat extraction from unstructured reports in Hausa, Pidgin, or English. This model specializes in identifying threat types, locations, perpetrators, and other critical details, returning a structured threat signal. It is specifically designed for community safety reporting in Northern Nigeria, serving as an on-device extraction model for the Tsaro shared safety system.
Loading preview...
Overview
Janeodum/tsaro-e2b is a specialized 5.1 billion parameter model, fine-tuned from Google's gemma-4-e2b-it base model. Its primary function is to extract structured threat signals from unstructured reports, specifically designed for the Tsaro shared safety system in Northern Nigeria. The model processes inputs in Hausa, Pidgin, or English, identifying key threat components and assessing report authenticity.
Key Capabilities
- Threat Extraction: Identifies threat type, location, perpetrator and vehicle counts, direction of movement, and time references.
- Language Support: Processes reports in Hausa, Pidgin, and English.
- Report Validation: Judges whether a message constitutes a genuine security report.
- On-Device Deployment: Functions as a smaller, on-device extraction model, suitable for older or low-RAM Android devices.
Training and Limitations
The model was fine-tuned using a LoRA adapter on Tsaro's proprietary threat-report data, which includes examples derived from the ACLED Nigeria conflict archive with Hausa and Pidgin translations. It is built for a specific regional context and is not intended as a general-purpose model. Outputs are designed as extraction assistance rather than verified intelligence.