{"categories":[{"label":"Artificial Intelligence","url":"https://skillfed.io/packages/category/scientific-engineering-artificial-intelligence/11"}],"enrichment":{"capability":"Provides axial positional embeddings for transformer networks, enabling position encoding across multi-dimensional data like images, videos, or sequences using decomposed attention axes.","skillfed_tags":["transformer-architecture","positional-encoding","vision-transformer"],"use_cases":["Encoding position information in vision transformers processing image patches arranged in 2D grids","Adding positional context to video transformers with temporal and spatial dimensions","Improving transformer efficiency on multi-dimensional data by factorizing position encoding across axes","Building language models that benefit from axial decomposition of sequence positions","Extending transformers to handle variable-sized multi-dimensional inputs with better generalization"],"what_it_does":"Axial Positional Embedding implements a specialized form of positional encoding designed for transformer networks working with multi-dimensional data. Instead of encoding position as a single sequence, it decomposes position information across multiple axes\u2014useful for images, videos, or any structured tensor input. The package provides two main variants: a standard AxialPositionalEmbedding that combines embeddings from separate axes, and a ContinuousAxialPositionalEmbedding that uses MLPs for better extrapolation across unseen dimensions.\n\nThe package depends on torch for tensor operations and einops for flexible tensor reshaping. It targets developers building attention-based models on non-sequential data, where traditional sequence-based positional encodings are inefficient or inappropriate. The implementation is lightweight and integrates directly into PyTorch models by adding the embedding to token representations.","worth_installing":"Yes, if you are building transformer models on multi-dimensional data (images, videos, or structured tensors) and want a lightweight, permissively licensed positional encoding layer. The aging maintenance status is a minor concern for a stable utility, but verify compatibility with your specific PyTorch version. No known vulnerabilities."},"id":"axial-positional-embedding","links":{"html":"https://skillfed.io/packages/axial-positional-embedding","md":"https://skillfed.io/packages/axial-positional-embedding.md","pypi":"https://pypi.org/project/axial-positional-embedding/"},"maintenance":{"status":"aging"},"meta":{"latest_release":"2025-02-25","license_spdx":null,"license_treatment":"permissive","name":"axial-positional-embedding","python_support":"supports_current","summary":"Axial Positional Embedding"},"popularity":{"monthly_downloads":84739,"position":13979,"tier":"top_15000"},"security":{"n_vulnerabilities":0},"version":"0.3.12"}