Python client library for moondream
Project description
Moondream Python Client Library
Python client library for moondream. This library is an alpha preview -- it is in an early stage of development, and backward compatibility is not yet guaranteed. If you are using this in production, please pin the revision you are using.
This library currently offers optimized CPU inference, but will be slower than the PyTorch implementation for CUDA and MPS backends. If you are running on a Mac with M1/M2/M3 etc. chips, or if you have a GPU available, this library is not recommended yet.
Setup
Install the library using pip:
pip install moondream==0.0.1
Then download the model weights:
# int8 weights (recommended):
wget "https://huggingface.co/vikhyatk/moondream2/resolve/client/moondream-latest-int8.bin.gz?download=true" -O - | gunzip > moondream-latest-int8.bin
# ...or, for FP16 weights:
wget "https://huggingface.co/vikhyatk/moondream2/resolve/client/moondream-latest-f16.bin.gz?download=true" -O - | gunzip > moondream-latest-f16.bin
Usage
import moondream as md
from PIL import Image
model = md.VL("moondream-latest-int8.bin")
image = Image.open("path/to/image.jpg")
# Optional -- encode the image to efficiently run multiple queries on the same
# image. This is not mandatory, since the model will automatically encode the
# image if it is not already encoded.
encoded_image = model.encode_image(image)
# Caption the image.
caption = model.caption(encoded_image)
# ...or, if you want to stream the output:
for t in model.caption(encoded_image, stream=True)["caption"]:
print(t, end="", flush=True)
# Ask a question about the image.
question = "How many people are in this image?"
answer = model.answer_question(encoded_image, question)["answer"]
# ...or again, if you want to stream the output:
for t in model.answer_question(encoded_image, question, stream=True)["answer"]:
print(t, end="", flush=True)
Accelerators
(TK -- document how ONNX execution providers work.)
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for moondream-0.0.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | e3bdcb66bf6b5f494dc8e66bd7fbca50234f24de30054a16418456920c34ad44 |
|
MD5 | 1547e8e77b9acd073f649d1efa18fe74 |
|
BLAKE2b-256 | aa2b3663e3a582519dc2c67a1bfd9105ee039c4387abb6912a7432556b3629b8 |