ithub847/Qwen2.5-VL-7B-Instruct-finetune

Qwen2.5-VL-7B-Instruct — Fine-tuned

Fine-tuned version of Qwen/Qwen2.5-VL-7B-Instruct by ithub847.

Model Description

Property	Value
Base model	Qwen/Qwen2.5-VL-7B-Instruct
Architecture	Qwen2.5-VL
Parameters	7B
License	Apache 2.0
Tasks	Image captioning, VQA, OCR, coding

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM, AutoProcessor
from PIL import Image
import torch

model_id = "ithub847/Qwen2.5-VL-7B-Instruct-finetune"

processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Text only
inputs = processor(text="Write a Python hello world", return_tensors="pt")
out = model.generate(**inputs, max_new_tokens=200)
print(processor.decode(out[0], skip_special_tokens=True))

Docker

docker run --gpus all -p 8000:8000 \
  -e HF_TOKEN=$HF_TOKEN \
  ghcr.io/ithub847/qwen-vl:latest

Training

Fine-tuned using LoRA / QLoRA on custom dataset. Details coming soon.

License

Apache 2.0