|
|
--- |
|
|
license: other |
|
|
base_model: Qwen/Qwen2-VL-2B-Instruct |
|
|
library_name: transformers |
|
|
tags: |
|
|
- qwen2-vl |
|
|
- lora |
|
|
- multimodal |
|
|
- amazon-listing |
|
|
- kaggle |
|
|
--- |
|
|
|
|
|
# Qwen2-VL LoRA — Amazon Listing Generator |
|
|
|
|
|
Lightweight LoRA adapter trained with **LLaMA-Factory** to turn a product image into an Amazon-style listing (title, bullets, description). |
|
|
|
|
|
> **Note:** This repo ships the **adapter only**. Load it on top of `Qwen/Qwen2-VL-2B-Instruct`. |
|
|
|
|
|
## Quickstart |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
from peft import PeftModel |
|
|
from PIL import Image |
|
|
|
|
|
base = "Qwen/Qwen2-VL-2B-Instruct" |
|
|
adapter = "soupstick/qwen2vl-amazon-ft-lora" |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(base, trust_remote_code=True, device_map="auto") |
|
|
model = PeftModel.from_pretrained(model, adapter) |
|
|
tok = AutoTokenizer.from_pretrained(base, trust_remote_code=True) |
|
|
|
|
|
img = Image.open("sample.png").convert("RGB") |
|
|
resp, _ = model.chat(tok, query="<image>\nGenerate Amazon listing.", history=[], image=img) |
|
|
print(resp) |
|
|
Training |
|
|
|
|
|
Framework: LLaMA-Factory (LoRA) |
|
|
|
|
|
Task: Multimodal instruction-following for e-commerce listings |
|
|
|
|
|
Data: community dataset (see the dataset card linked below) |
|
|
|
