rules-qa-instruct

LoRA-specialized then merged Qwen2.5-0.5B-Instruct for grounded rules answers (@geck018/rules-engine). Trained on multi-domain synthetic Q&A (distractors + abstention).

Usage (Transformers.js)

import { pipeline } from '@huggingface/transformers';
const generator = await pipeline('text-generation', 'geck018/rules-qa-instruct', { dtype: 'q4' });

If a quantized ONNX variant is missing, the library falls back to onnx-community/Qwen2.5-0.5B-Instruct.