import torch
from peft import PeftModel
from transformers import AutoModelForImageTextToText, AutoProcessor

base = "Qwen/Qwen3-VL-8B-Instruct"
adapter = "nhantran214/damage-vlm-qwen3vl-8b-lora"

processor = AutoProcessor.from_pretrained(base, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
    base, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True
)
model = PeftModel.from_pretrained(model, adapter)

Low-VRAM (4-bit)

Source dir: outputs/runs/full_winner_v2
Package kind: lora_adapter
Exported (UTC): 2026-08-06 01:10:28Z

bash

pip install bitsandbytes
# In this repo:
python scripts/08_infer_full_winner.py --load-in-4bit \
  --adapter nhantran214/damage-vlm-qwen3vl-8b-lora --image path/to.jpg

Project

Offline competition pipeline — see the training repository README for data tiers, holdout bake-off, and full-winner retrain steps.