recallclear-smollm2-135m-lora

RecallClear — plain-language vehicle recall explainer

A LoRA adapter for HuggingFaceTB/SmolLM2-135M-Instruct that rewrites official U.S. vehicle safety recall notices into a five-part summary an owner can act on:

WHAT'S WRONG / WHAT COULD HAPPEN / HOW URGENT / WHAT TO DO / WHAT IT COSTS

Training data

4,651 recall notices published by NHTSA between 2013 and 2025 (Recalls Data, public domain). Reference cards were derived from the notices' own structured fields using a curated plain-language lexicon and a four-level urgency ladder anchored on NHTSA's official do_not_drive and fire_risk_when_parked flags.

Seven manufacturers (Subaru, Mazda, Tesla, Mitsubishi, Volvo, Porsche, Winnebago) were held out entirely, so evaluation measures generalisation to unseen brands.

Intended use and limits

This is a reading aid, not safety or legal advice. The official notice remains authoritative and should always be shown alongside the rewrite. Do-not-drive and park-outside warnings should be taken from NHTSA's flags, not from this model: evaluation found the notice text alone often lacks the evidence needed to recover them.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("HuggingFaceTB/SmolLM2-135M-Instruct")
model = PeftModel.from_pretrained(base, "HanfuZhao781/recallclear-smollm2-135m-lora")
tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM2-135M-Instruct")

Full pipeline, evaluation, and web app: https://github.com/hanfuzhao/Module4-RecallClear