The latest in machine learning: jax-ml/ml_dtypes
The field of machine learning has seen significant advances with the introduction of jax-ml/ml_dtypes. This package broadens the capabilities of Python users by extending the data types available in NumPy, an essential tool in data computation. With this innovation, professionals can apply a variety of new data types, such as the bfloat16 and specialized 8-bit floating point formats, tailored for intricate computational work.
Further enhancing Python’s versatility, ml_dtypes supports versions 3.9 to 3.12 and offers installation from source. Acknowledging the critical role of precision in calculations, the creators also call attention to considerations when using these low-precision data types. Additionally, the initiative values openness, with source code available under the Apache 2.0 license and pre-compiled wheels benefiting from EIGEN, shared under MPL 2.0.
Read more: [
Github
](https://github.com/jax-ml/ml_dtypes)