What's New: Intel® Distribution for Python* 2025.1

[StefR_Intel]

The Intel® Distribution for Python, 2025.1 has been updated to include functional and security updates. Users should update to the latest version. For more details on security and known issues, please check out Intel® Distribution for Python* Known Issues.

What’s new

Data Parallel Extension for Python* includes the following improvements:

• Achieved 100% compliance with Python Array API specification (revision "2023.12").
• Added mathematical functions (dpnp.gcd, dpnp.i0, dpnp.lcm, dpnp.ldexp, dpnp.sinc, dpnp.spacing).
• Added manipulation functions (dpnp.delete, dpnp.insert, dpnp.pad, dpnp.unstack)
• Added statistics functions (dpnp.bincount, dpnp.corrcoef, dpnp.correlate, dpnp.histogram2d, dpnp.histogramdd, dpnp.nanmedian).
• Added linear algebra functions (dpnp.linalg.diagonal, dpnp.linalg.cross, dpnp.linalg.matrix_norm, dpnp.linalg.outer, dpnp.linalg.tensordot, dpnp.linalg.trace, dpnp.linalg.svdvals, dpnp.linalg.vecdot, dpnp.linalg.vector_norm, dpnp.matrix_transpose, dpnp.matvec, dpnp.vecmat).
• Added indexing function dpnp.compress
• Improved performance of dpnp.histogram, dpnp.choose and dpnp.nan_to_num functions, and dpnp.ndarray.fill method.
• Provided compatibility with NumPy 2.2.3.

Data Parallel Control Library* includes the following improvements:

• Added tensor.top_k function for getting the k smallest or largest elements, which will be coming to a future array API release.
• Added tensor.dldevice_to_sycl_device, tensor.sycl_device_to_dldevice, and dpctl.SyclDevice.get_device_id methods to improve interoperability with DLPack.
• Improved performance of copy-and-cast operations for numpy.ndarray to tensor.usm_ndarray for contiguous inputs and for strided arrays to contiguous arrays.
• Improved performance of tensor accumulation functions (tensor.cumulative_sum, tensor.cumulative_prod, and tensor.cumulative_logsumexp).
• Improved performance of tensor reductions (tensor.sum, tensor.max, etc.)
• Improved performance of sort operations for some data types (especially small integral types) and added ‘radixsort’ sort kind to tensor.sort and tensor.argsort.