metadata
license: apache-2.0
pipeline_tag: image-text-to-text
moondream2 is a small vision language model designed to run efficiently on edge devices. Check out the GitHub repository for details, or try it out on the Hugging Face Space!
Benchmarks
| Release | VQAv2 | GQA | TextVQA | TallyQA (simple) | TallyQA (full) |
|---|---|---|---|---|---|
| 2024-03-04 | 74.2 | 58.5 | 36.4 | - | - |
| 2024-03-06 | 75.4 | 59.8 | 43.1 | 79.5 | 73.2 |
| 2024-03-13 | 76.8 | 60.6 | 46.4 | 79.6 | 73.3 |
| 2024-04-02 | 77.7 | 61.7 | 49.7 | 80.1 | 74.2 |
| 2024-05-08 | 79.0 | 62.7 | 53.1 | 81.6 | 76.1 |
| 2024-05-20 (latest) | 79.4 | 63.1 | 57.2 | 82.1 | 76.6 |
Usage
pip install transformers einops
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
model_id = "vikhyatk/moondream2"
revision = "2024-05-20"
model = AutoModelForCausalLM.from_pretrained(
model_id, trust_remote_code=True, revision=revision
)
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
image = Image.open('<IMAGE_PATH>')
enc_image = model.encode_image(image)
print(model.answer_question(enc_image, "Describe this image.", tokenizer))
The model is updated regularly, so we recommend pinning the model version to a specific release as shown above.