metadata

title: README
emoji: ⚡
colorFrom: blue
colorTo: green
sdk: static
pinned: false

MLX Vision

A community org for model weights compatible with (mlx-im)[https://github.com/riccardomusmeci/mlx-im/tree/main] powered by MLX.

These are weights converted from timm/torchvision and ready to be used.

How to install

git clone https://github.com/riccardomusmeci/mlx-im
cd mlx-im
pip install .

Models

To create a model with weights:

from mlxim.model import create_model

# loading weights from mlx-vision
model = create_model("resnet18")

# loading weights from local file
model = create_model("resnet18", weights="path/to/weights.npz")

As of today (2024-03-05) mlx does not support nn.Conv2d with group or dilation greater than 1 (e.g. resnext, regnet, efficientnet).