mlx-data

0.2.0 verified Mon Apr 27 auth: no python

Universal data loaders for Apple MLX framework. Current version 0.2.0, released on 2024-09-20. Development is active with weekly releases.

pip install mlx-data

Common errors

error ModuleNotFoundError: No module named 'mlx.data' ↓

cause mlx-data is not installed or version is too old.

fix

Run pip install mlx-data --upgrade.

error AttributeError: module 'mlx.data' has no attribute 'DataLoader' ↓

cause Incorrect import path; users often try 'import mlx.data' then use mlx.data.DataLoader.

fix

Use from mlx.data import DataLoader.

Warnings

breaking DataLoader.batch() method was removed in v0.2.0; use DataLoader.from_iterable with batch_size argument instead. ↓

fix Replace `dataloader.batch(4)` with `DataLoader.from_iterable(data, batch_size=4)`.

deprecated The `shuffle` parameter in `DataLoader` is deprecated; use `.shuffle()` method instead. ↓

fix Remove shuffle from constructor and call `dataloader.shuffle()` before iteration.

Imports

DataLoader
wrong
```
import mlx.data
```
correct
```
from mlx.data import DataLoader
```
Correct usage is 'from mlx.data import DataLoader' rather than 'import mlx.data'.

Quickstart

Create a DataLoader from a simple iterable and iterate over batches.

from mlx.data import DataLoader
import mlx.core as mx

dataloader = DataLoader.from_iterable([1, 2, 3, 4, 5], batch_size=2)
for batch in dataloader:
    print(batch)