LettuceDetect is a transformer-based model for hallucination detection on context and answer pairs, designed for Retrieval-Augmented Generation (RAG) applications. This model is built on ModernBERT, which has been specifically chosen and trained becasue of its extended context support (up to 8192 tokens). This long-context capability is critical for tasks where detailed and extensive documents need to be processed to accurately determine if an answer is supported by the provided context.

This is our Large model based on ModernBERT-large

Model Details

Architecture: mmBERT-base with extended context support (up to 8192 tokens)
Task: Token Classification / Hallucination Detection
Training Dataset: RagTruth-HU
Language: Hungarian

How It Works

The model is trained to identify tokens in the answer text that are not supported by the given context. During inference, the model returns token-level predictions which are then aggregated into spans. This allows users to see exactly which parts of the answer are considered hallucinated.

Usage

Installation

Install the 'lettucedetect' repository

pip install lettucedetect

Using the model

python

from lettucedetect.models.inference import HallucinationDetector

detector = HallucinationDetector(
    method="transformer",
    model_path="KRLabsOrg/lettucedect-mmbert-base-hu-v1",
    lang="hu",
    trust_remote_code=True
)

contexts = [
    "Franciaország fővárosa Párizs. Franciaország népessége 67 millió fő. Franciaország területe 551 695 km²."
]
question = "Mennyi Franciaország népessége?"
answer = "Franciaország népessége 125 millió fő."

predictions = detector.predict(context=contexts, question=question, answer=answer, output_format="spans")
print("Predictions:", predictions)

# Predictions: [{'start': 0, 'end': 38, 'confidence': 0.9475189447402954, 'text': 'Franciaország népessége 125 millió fő.'}]

Performance

Results on Translated RAGTruth-HU (Class 1: Hallucination)

We evaluate our Hungarian models on the translated RAGTruth dataset. As a prompt baseline we include meta-llama/Llama-4-Maverick-17B-128E-Instruct.

Language	Model	Precision (%)	Recall (%)	F1 (%)	Maverick F1 (%)	Δ F1 (%)
Hungarian	meta-llama/Llama-4-Maverick-17B-128E-Instruct	38.70	96.82	55.30	55.30	+0.00
Hungarian	lettucedect-mmBERT-small (ours)	70.20	72.51	71.33	55.30	+16.03
Hungarian	lettucedect-mmBERT-base (ours)	76.62	69.21	72.73	55.30	+17.43

Note: Percentages are reported for the hallucination class (Class 1). Δ F1 is measured in percentage points vs. the Maverick baseline.

Citing

If you use the model or the tool, please cite the following paper:

bibtex

@misc{Kovacs:2025,
      title={LettuceDetect: A Hallucination Detection Framework for RAG Applications}, 
      author={Ádám Kovács and Gábor Recski},
      year={2025},
      eprint={2502.17125},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.17125}, 
}