---
language:
- en
license: apache-2.0
tags:
- rag
- faithfulness
- hallucination-detection
- lora
- microguard
datasets:
- galileo-ai/ragbench
- wandb/RAGTruth-processed
- PatronusAI/HaluBench
metrics:
- balanced_accuracy
- f1
pipeline_tag: text-classification
base_model: google/gemma-3-270m-it
---

# MicroGuard — Gemma-270M

A LoRA-adapted faithfulness classifier for RAG systems. Detects whether a generated answer is faithful to the retrieved context.

## Performance

| Metric | Value |
|--------|-------|
| Balanced Accuracy | 67.0% |
| F1 Score | 0.688 |
| Cohen's Kappa | 0.379 |
| Inference Latency | 60ms |

Evaluated on a combined test set of 15,976 examples from RAGBench, RAGTruth, and HaluBench.

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("google/gemma-3-270m-it")
model = PeftModel.from_pretrained(base, "tarun5986/MicroGuard-Gemma-270M")
tokenizer = AutoTokenizer.from_pretrained("google/gemma-3-270m-it")

# Or use the MicroGuard package
from microguard import MicroGuard
guard = MicroGuard(model="tarun5986/MicroGuard-Gemma-270M", base_model="google/gemma-3-270m-it")
result = guard.check(
    context="The Eiffel Tower was built in 1889 by Gustave Eiffel.",
    question="Who built the Eiffel Tower?",
    answer="The Eiffel Tower was built by Gustave Eiffel in 1889."
)
print(result)  # {'verdict': 'FAITHFUL', 'confidence': 74.2, 'latency_ms': 64.0}
```

## Training

- **Method**: LoRA (r=16, alpha=32, targets: q,k,v,o projections)
- **Data**: 127,932 examples from RAGBench + RAGTruth + HaluBench
- **Evaluation**: Constrained decoding via logit comparison (0% garbage outputs)

## Paper

[MicroGuard: Sub-Billion Parameter Faithfulness Classification for Real-Time RAG QA](https://github.com/tarun-ks/MicroGuard)

## Citation

```bibtex
@article{microguard2026,
  title={MicroGuard: Sub-Billion Parameter Faithfulness Classification for Real-Time RAG QA},
  author={Sharma, Tarun},
  journal={IEEE Access},
  year={2026}
}
```