import { AutoTokenizer, AutoModelForCausalLM } from "@huggingface/transformers";

const modelId = "tomvaillant/qwen3-4b-abliterated-v2-journalist-ONNX";
const tokenizer = await AutoTokenizer.from_pretrained(modelId);
const model = await AutoModelForCausalLM.from_pretrained(modelId, {
  dtype: "q4",
  device: "webgpu",
});

const messages = [
  { role: "user", content: "What records should I check to verify who owns a local company?" },
];

const inputs = tokenizer.apply_chat_template(messages, {
  add_generation_prompt: true,
  return_dict: true,
  enable_thinking: false,
});

const output = await model.generate({ ...inputs, max_new_tokens: 512 });
console.log(tokenizer.decode(output[0], { skip_special_tokens: true }));

Files

onnx/model_q4.onnx
onnx/model_q4.onnx_data*
genai_config.json
tokenizer and chat-template files

Training And Conversion

Adapter: tomvaillant/qwen3-4b-abliterated-v2-journalist
Merged checkpoint: tomvaillant/qwen3-4b-abliterated-v2-journalist-merged
Training: QLoRA with Unsloth + TRL SFT
ONNX export: onnxruntime-genai int4 export with external data split for browser loading
Dataset: tomvaillant/investigative-journalism-training

Sources And Attribution

Training data: tomvaillant/investigative-journalism-training — 687 instruction/response pairs synthesized by Claude Opus 4.6 (Anthropic) from the Buried Signals OSINT and investigative-journalism corpus: OSINT Navigator tool data, Indicator Media briefings, Buried Signals investigative skills, GIJN, Bellingcat, Verification Handbook 3, SPJ Code of Ethics, RCFP, and public manuals from UNESCO, Al Jazeera Media Institute, CiFAR, CIPE, and EJF/TEMPO Institute.

See the dataset card for the full source list, licenses, and per-partner attribution.

Intended Use

Built for in-browser investigative assistance in OSINT workflows. Treat generated tool recommendations, URLs, and factual claims as research leads requiring independent verification.