gemma4-e4b-journalist

A compact investigative journalism model fine-tuned on google/gemma-4-E4B-it (8B params), designed for browser-sized deployment via WebGPU. Same training data and domain coverage as gemma4-31b-journalist.

Built by Buried Signals for edge/browser inference in OSINT Navigator.

Training

• Method: QLoRA (4-bit NF4 + LoRA r=16, alpha=32) via Oumi
• Data: 700 instruction/response pairs across 10 categories — see training dataset
• Epochs: 1
• Hardware: A10G (HF Jobs)

Sources

Same training corpus as the 31B variant. Covers OSINT tool selection, verification methodology, financial investigation (Follow the Money), digital security, media ethics, and investigative storytelling.

Key sources include the OSINT Navigator Tool Database (7,524 tools), Bellingcat guides, GIJN manuals, UNESCO/Al Jazeera/CiFAR handbooks, SPJ ethics, RCFP legal resources, and Buried Signals investigation skill repositories.

Full attribution: SOURCES.md

GGUF

This model was converted to GGUF format using Unsloth.

Example usage:

• Text only: llama-cli -hf tomvaillant/gemma4-e4b-journalist --jinja
• Multimodal: llama-mtmd-cli -hf tomvaillant/gemma4-e4b-journalist --jinja

Available files

• gemma-4-E4B-it.Q4_K_M.gguf
• gemma-4-E4B-it.BF16-mmproj.gguf

This was trained 2x faster with Unsloth