coderop12/fnd-qwen2vl7b-stage-a

FND – Stage A (Qwen2-VL-7B LoRA)

Goal: General image↔caption alignment on COCO 2017. Vision frozen; LoRA on language.

Use

from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
from peft import PeftModel

base = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-7B-Instruct", device_map="auto", torch_dtype="auto", trust_remote_code=True)
proc = AutoProcessor.from_pretrained("coderop12/fnd-qwen2vl7b-stage-a", trust_remote_code=True)
model = PeftModel.from_pretrained(base, "coderop12/fnd-qwen2vl7b-stage-a").eval()

See metadata.json for training knobs.