# humanize-e Adapter A QLoRA LoRA adapter for Qwen2.5-3B-Instruct trained to humanize AI-generated text. ## Model Details - **Base Model:** Qwen/Qwen2.5-3B-Instruct - **Fine-tuning Method:** QLoRA (LoRA r=32, alpha=64) - **Training Data:** 156 synthetic pairs from Kaggle AI vs Human Text dataset - **Developer:** Mike Elistra (Mike Koci) - **Training Notebook:** https://www.kaggle.com/code/mick009/notebook9c4e543da9 ## Usage ```python from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig from peft import PeftModel import torch base_model = "Qwen/Qwen2.5-3B-Instruct" adapter_path = "mike-elistra/humanize-e-adapter" tokenizer = AutoTokenizer.from_pretrained(adapter_path) model = AutoModelForCausalLM.from_pretrained( base_model, quantization_config=BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4"), device_map="auto" ) model = PeftModel.from_pretrained(model, adapter_path) model.eval() # Use with humanize-e inference code ``` ## Inference API The adapter works with the humanize-e FastAPI service: ```bash HUMANIZE_ADAPTER=mike-elistra/humanize-e-adapter python app.py ``` ## Training See Kaggle notebook for full training pipeline: https://www.kaggle.com/code/mick009/notebook9c4e543da9 ## License MIT