File size: 1,591 Bytes
28291b5 ac7d0b1 9bd11ae ac7d0b1 28291b5 ac7d0b1 28291b5 ac7d0b1 28291b5 ac7d0b1 28291b5 ac7d0b1 28291b5 ac7d0b1 28291b5 ac7d0b1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 | ---
license: gemma
base_model: google/gemma-4-E2B-it
base_model_relation: finetune
library_name: transformers
tags:
- gemma-4
- tsaro
- threat-extraction
language:
- ha
- en
pipeline_tag: text-generation
---
# Tsaro Gemma 4 E2B
Fine-tuned Gemma 4 E2B threat extraction model for Tsaro, a shared safety
system for Northern Nigeria.
## What this model does
Given an unstructured report in Hausa, Pidgin, or English, this model returns
a structured threat signal — threat type, location, perpetrator and vehicle
counts, direction of movement, time references, and a confidence score — and
judges whether the message is a genuine security report at all.
## Model details
- **Base model:** [`google/gemma-4-e2b-it`](https://huggingface.co/google/gemma-4-e2b-it)
- **Fine-tuning:** LoRA adapter trained on Tsaro threat-report data, then merged
into the base weights
- **Role in Tsaro:** the E2B variant is the smaller of two on-device extraction
models, used as the fallback for older or low-RAM Android devices
## Derived models
- [`Janeodum/tsaro-e2b-gguf`](https://huggingface.co/Janeodum/tsaro-e2b-gguf) — GGUF quantization for
on-device inference via llama.cpp / llama.rn
## Training data
Fine-tuned on threat-report examples spanning Hausa, Pidgin, and English,
including examples derived from the ACLED Nigeria conflict archive with
Hausa and Pidgin translations.
## Intended use and limitations
Built for community safety reporting in a specific regional context. Not a
general-purpose model. Outputs are extraction assistance, not verified
intelligence.
|