---
license: apache-2.0
base_model: huihui-ai/Huihui-Qwen3.5-9B-abliterated
language:
  - en
tags:
  - safetensors
  - qwen3_5
  - investigative-journalism
  - osint
  - merged
  - conversational
---

# qwen3.5-9b-abliterated-journalist-merged

Merged safetensors checkpoint for [tomvaillant/qwen3.5-9b-abliterated-journalist](https://huggingface.co/tomvaillant/qwen3.5-9b-abliterated-journalist), an investigative journalism and OSINT fine-tune based on [huihui-ai/Huihui-Qwen3.5-9B-abliterated](https://huggingface.co/huihui-ai/Huihui-Qwen3.5-9B-abliterated).

For local llama.cpp use, prefer the GGUF export:

- [qwen3.5-9b-abliterated-journalist-GGUF](https://huggingface.co/tomvaillant/qwen3.5-9b-abliterated-journalist-GGUF)

## Training

- **Adapter:** [tomvaillant/qwen3.5-9b-abliterated-journalist](https://huggingface.co/tomvaillant/qwen3.5-9b-abliterated-journalist)
- **Method:** QLoRA with [Unsloth](https://github.com/unslothai/unsloth) + TRL SFT, then merged into the base model
- **Dataset:** `tomvaillant/investigative-journalism-training`
- **Task:** investigative reporting, OSINT methodology, verification, source handling, and ethics

## Sources And Attribution

The dataset is grounded in OSINT Navigator tool data, Indicator Media briefing tools, Buried Signals investigative skills, GIJN and Bellingcat materials, Verification Handbook 3, SPJ ethics, RCFP resources, and public investigative manuals from UNESCO, Al Jazeera Media Institute, CiFAR, CIPE, and EJF/TEMPO Institute.

It contains 700 source-grounded instruction examples generated with Claude Opus 4.6. Full attribution is maintained in the training repository as `SOURCES.md`.

## Intended Use

This checkpoint can support newsroom and research workflows, but its outputs are not primary-source evidence. Verify generated URLs, facts, legal claims, and tool recommendations independently.