Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy

A fine-tuned version of Qwen3-8B for news media bias detection and neutral rewriting, developed by the Vector Institute as part of the UnBias-Plus project.

Given a news article, the model identifies biased language segments, classifies their bias type and severity, provides neutral replacements, and returns a fully rewritten unbiased version of the article — all in a single structured JSON response.

This is the Instruct variant — trained without chain-of-thought thinking blocks (enable_thinking=False). It produces clean structured JSON directly, making it faster and more reliable for production inference, including deployment via vLLM or other OpenAI-compatible serving backends.

Trained on: train_1 — UnBias-Plus

Model Details

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch, json

model_id = "vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
    device_map="auto",
)
model.eval()

SYSTEM_PROMPT = """You are an expert linguist and bias detection specialist.
Your task is to carefully read a news article, detect ALL biased language,
and return a structured JSON response. Return ONLY valid JSON, no extra text."""

article = "Your news article here..."

messages = [
    {"role": "system", "content": SYSTEM_PROMPT},
    {"role": "user", "content": f"Analyze the following article for bias and return the result in the required JSON format.\n\nARTICLE:\n{article}"},
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    enable_thinking=False,   # must be False for this variant
    return_tensors="pt",
    return_dict=True,
    truncation=True,
    max_length=8192,
)

with torch.no_grad():
    outputs = model.generate(
        input_ids=inputs["input_ids"].to(model.device),
        attention_mask=inputs["attention_mask"].to(model.device),
        max_new_tokens=4096,
        do_sample=False,       # greedy decoding for deterministic JSON
        temperature=None,
        top_p=None,
        pad_token_id=tokenizer.eos_token_id,
    )

new_tokens = outputs[0][inputs["input_ids"].shape[1]:]
response = tokenizer.decode(new_tokens, skip_special_tokens=True)
result = json.loads(response)

Using with the UnBias-Plus toolkit

from unbias_plus import UnBiasPlus

pipe = UnBiasPlus(
    model_name_or_path="vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy",
    load_in_4bit=False,   # set True for ~5GB VRAM
)

result = pipe.analyze("Your article text here...")
print(result.binary_label)      # "biased" or "unbiased"
print(result.severity)          # 0, 2, 3, or 4
print(len(result.biased_segments))
print(result.unbiased_text)

Using with vLLM

vllm serve vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy \
  --max-model-len 8192

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")

completion = client.chat.completions.create(
    model="vector-institute/Qwen3-8B-UnBias-Plus-SFT-Instruct-Legacy",
    messages=messages,
    max_tokens=4096,
    temperature=0,
    extra_body={"chat_template_kwargs": {"enable_thinking": False}},
)
result = json.loads(completion.choices[0].message.content)

Output Schema

{
  "binary_label": "biased" | "unbiased",
  "severity": 0 | 2 | 3 | 4,
  "bias_found": true | false,
  "biased_segments": [
    {
      "original": "exact substring from input article",
      "replacement": "neutral alternative phrase",
      "severity": "high" | "medium" | "low",
      "bias_type": "loaded language | dehumanizing framing | false generalizations | framing bias | euphemism/dysphemism | politically charged terminology | sensationalism",
      "reasoning": "1-2 sentence explanation"
    }
  ],
  "unbiased_text": "Full rewritten neutral article"
}

Severity Scale

Bias Types Detected

• Loaded language — words with strong emotional connotations
• Dehumanizing framing — language that strips dignity from groups
• False generalizations — sweeping statements ("they always", "all of them")
• Framing bias — selective wording that implies a viewpoint
• Euphemism/dysphemism — softening or hardening language to manipulate perception
• Politically charged terminology — labels used to provoke rather than describe
• Sensationalism — exaggerated language to evoke emotional responses

Training Data

Fine-tuned on vector-institute/Unbias-plus, a curated dataset of news articles with expert-annotated bias labels, segment-level annotations, and neutral rewrites.

Hardware Requirements

Model Variants

Limitations

• Trained primarily on English-language news articles
• Political bias detection reflects patterns in the training data
• Best performance on articles under 5000 characters
• As with all language models, outputs should be reviewed by a human before use in production

Citation

If you use this model in your research or application, please cite:

  title={UnBias-Plus: Detect, Explain, and Rewrite Bias},
  author={Radwan, Ahmed Y and ElKady, Ahmed and Chaduvula, Sindhuja and Hafez, Mohamed and Krishnan, Amrit and Raza, Shaina},
  journal={arXiv preprint arXiv:2606.23412},
  year={2026}
}

Links

• 📊 Leaderboard: UnBias-Plus Leaderboard
• 📁 Dataset: vector-institute/unbias-plus-dataset
• 🏛️ Organization: Vector Institute
• 🌐 Project: AIXpert