from transformers import AutoModelForVision2Seq, AutoProcessor
from peft import PeftModel

# Load base model
base_model = AutoModelForVision2Seq.from_pretrained(
    "Qwen/Qwen2.5-VL-32B",
    trust_remote_code=True
)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "yongxianwei/Qwen2.5-VL-32B-Grounding")
processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-32B", trust_remote_code=True)

# Inference
# ... your inference code ...

Training Details

Base Model: Qwen/Qwen2.5-VL-32B
Task: Visual Grounding
Training Method: LoRA (Low-Rank Adaptation)
Framework: Transformers

Fine-tuned using LoRA on specific Visual Grounding datasets.

Citation

bibtex

@misc{qwen2.5-vl-visual grounding,
  author = {Yongxian Wei},
  title = {Qwen2.5-VL LoRA for Visual Grounding},
  year = {2024},
  publisher = {Hugging Face},
  url = {https://huggingface.co/yongxianwei/Qwen2.5-VL-32B-Grounding}
}