qwen3.6-27b-abliterated-journalist

LoRA adapter for investigative journalism and OSINT workflows, fine-tuned from huihui-ai/Huihui-Qwen3.6-27B-abliterated.

This is the adapter checkpoint. For local inference, use the merged or GGUF variants:

• qwen3.6-27b-abliterated-journalist-merged
• qwen3.6-27b-abliterated-journalist-GGUF

Training

• Method: LoRA with Unsloth FastModel + TRL SFT, following the official Unsloth Qwen3.5 fine-tune recipe (canonical for Qwen3.6 too).
• Base model: huihui-ai/Huihui-Qwen3.6-27B-abliterated (multimodal, hybrid-thinking, dense 27B, qwen3_5 architecture)
• Dataset: tomvaillant/investigative-journalism-training (687 examples, OSINT methodology)
• LoRA config: r=16, alpha=16, dropout=0; targets q/k/v/o/gate/up/down projections; bias="none"; use_gradient_checkpointing="unsloth"
• Precision: bf16 (load_in_16bit=True; 4-bit QLoRA explicitly not recommended for Qwen3.5/3.6 per Unsloth docs)
• Optimizer: adamw_8bit, lr 2e-4, warmup 10 steps, 2 epochs, effective batch = 1 × 4 grad-accum
• Vision tower: frozen (finetune_vision_layers=False) — text-only LoRA, vision capability preserved byte-identical to base
• Final loss: 0.52 (step 300, epoch ~1.75)
• Task: investigative reporting assistance, OSINT methodology, verification, public-records research, source handling, and ethics

Sources And Attribution

Training examples were synthesized and curated from the following sources.

OSINT tool data (Buried Signals / Indicator Media)

• tomvaillant/osint-tool-database — 7,524 OSINT tools, weekly crawl of 9 open-source toolkits. Creator: Tom Vaillant / Buried Signals.
• OSINT Navigator Cached Q&A — 238 production query/answer pairs with Wilson-score ranking. Creator: Tom Vaillant / Buried Signals.
• Indicator Media Briefing Tools Database — 101 tool entries extracted from 37 briefings. Creator: Indicator Media. Used with permission.

Buried Signals skills (buriedsignals/osint, buriedsignals/spotlight)

• OSINT Investigation Methodology, Follow the Money, OPSEC Basics, Spotlight Investigator, Spotlight Fact-Checker, Evidence Grounding. Creator: Tom Vaillant / Buried Signals.
• Social Media Intelligence skill — adapted from jamditis/claude-skills-journalism by Jay Amditis (MIT License).

Follow the Money primary sources

• Jim Shultz, Follow the Money: A Guide to Monitoring Budgets and Oil and Gas Revenues, Revenue Watch / Open Society Institute, 2005 (ISBN 1-891385-40-2).
• Miranda Patrucic & Jelena Cosic, Introduction to Investigative Journalism: Following the Money, GIJN, 2024 (CC BY-ND 4.0).
• Derek Bowler, Tracing Beneficial Ownership with OSINT for Financial Crime, EBU Eurovision News Spotlight, 2025.

Investigative manuals (PDFs)

• GIJN, Citizen Investigations: A Practical Guide (2019).
• UNESCO / Mark Lee Hunter, Story-Based Inquiry: A Manual for Investigative Journalists.
• UNESCO, Global Investigative Journalism Casebook.
• Al Jazeera Media Institute, Investigative Journalism Handbook.
• EJF / TEMPO Institute, Investigative Journalism Training Manual.
• CiFAR, Investigate — The Manual (illicit financial flows, 2021).
• CIPE, Investigative Reporting: A Toolkit for Reporters (2009).

Web references

• European Journalism Centre, Verification Handbook 3.
• Society of Professional Journalists, Code of Ethics.
• Bellingcat resources and how-tos (CC BY-NC-SA 4.0 where applicable).
• GIJN Resource Center.
• Reporters Committee for Freedom of the Press (RCFP).

Synthetic generation

• 700 Claude Opus 4.6 (Anthropic) instruction/response pairs generated from the materials above. System prompt embeds investigative ethics (SIFT, evidence grounding, proportionality, source protection).

Intended Use

This model is intended for journalist-facing assistant workflows: investigation planning, OSINT tool selection, verification checklists, public-source research methods, and evidence-grounded drafting. Verify model outputs before use in reporting.

This was trained with Unsloth.