# LLaVA-Next-Video LoRA — Traffic Video QA

Base: llava-hf/LLaVA-NeXT-Video-7B-hf
Domain: InterAct urban-traffic video QA
Weights: LoRA adapter only (use PEFT).

Usage:
from transformers import LlavaNextVideoForConditionalGeneration, LlavaNextVideoProcessor
from peft import PeftModel
base_id = "llava-hf/LLaVA-NeXT-Video-7B-hf"
adapter_id = "UDVideoQA/llava-next-video-traffic-qa-lora"
processor = LlavaNextVideoProcessor.from_pretrained(base_id)
base = LlavaNextVideoForConditionalGeneration.from_pretrained(base_id, device_map="auto", torch_dtype="auto")
model = PeftModel.from_pretrained(base, adapter_id)