byoai-t5-liar-classifier

Fine-tuned T5-small to classify statements into six factuality labels: pants-fire, false, barely-true, half-true, mostly-true, true.

Source: Generated from the book Build Your Own AI dataset (BYOAI_LIAR). Includes short, structured inputs: classify: statement: context: tags: chapter:

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tok = AutoTokenizer.from_pretrained("gcuomo/byoai-t5-liar-classifier")
mdl = AutoModelForSeq2SeqLM.from_pretrained("gcuomo/byoai-t5-liar-classifier")
prompt = '''classify:
statement: RAG retrieves passages from a vector store like ChromaDB before generating.
context: RAG retrieval then generation
tags: data-prep, feature-engineering, rag
chapter: Prepping Data for AI'''
out = mdl.generate(**tok(prompt, return_tensors="pt", truncation=True, max_length=128))
print(tok.decode(out[0], skip_special_tokens=True))

## Training
    •	Base model: t5-small
    •	Epochs: 5
    •	Train/Val/Test sizes: 4549 / 520 / 489
    •	Labels: pants-fire, false, barely-true, half-true, mostly-true, true
    •	Prompt template as above.

## Limitations

Border classes (e.g., true vs mostly-true) can be confused. Provide short,
specific context and tags for best results.

## Citation

If you use this model in academic or educational work, please cite:

> Cuomo, G., & De Jesús, J. *Build Your Own AI*. BYOAI Project, 2025-2026.