{"categories":[{"label":"Artificial Intelligence","url":"https://skillfed.io/packages/category/scientific-engineering-artificial-intelligence/4"}],"enrichment":{"capability":"Provides hardware-efficient implementations of linear attention, sparse attention, state space models, and hybrid sequence model architectures, verified across NVIDIA, AMD, and Intel platforms.","skillfed_tags":["linear-attention","sequence-models","gpu-kernels"],"use_cases":["Replace standard Transformer attention with linear or sparse alternatives to reduce memory and compute during training or inference of large language models.","Experiment with state space models as sequence modeling primitives without implementing custom CUDA kernels.","Train hybrid models that combine multiple attention mechanisms in a single architecture.","Deploy sequence models on AMD or Intel accelerators where NVIDIA-only libraries are unavailable or suboptimal.","Benchmark and compare different subquadratic attention schemes on the same hardware and codebase."],"what_it_does":"Flash Linear Attention is a library of optimized building blocks for modern sequence models that move beyond standard quadratic-complexity attention. It bundles hardware-efficient kernels, training-ready layers, and complete model implementations spanning linear attention mechanisms, sparse attention patterns, state space models, and hybrid architectures that mix different attention types. The library targets practitioners building or experimenting with alternatives to Transformers\u2014researchers exploring state-of-the-art sequence modeling papers, engineers deploying models on resource-constrained hardware, and teams training large language models where memory and compute efficiency matter.\n\nThe package is organized around fused operations and platform-agnostic implementations verified on NVIDIA, AMD, and Intel hardware. It depends only on einops for tensor manipulation and requires Python 3.10 or later. The description and news section indicate active development with frequent additions of new model variants and backends throughout 2025 and 2026, suggesting this is a living research-to-production bridge rather than a static library.","worth_installing":"Yes. The package is actively maintained, has low install friction, carries a permissive MIT License, and fills a genuine need for practitioners exploring or deploying efficient sequence models. No known security vulnerabilities. Install if you are building or researching linear attention, state space models, or hybrid architectures; skip if you only need standard Transformer attention."},"id":"fla-core","links":{"html":"https://skillfed.io/packages/fla-core","md":"https://skillfed.io/packages/fla-core.md","pypi":"https://pypi.org/project/fla-core/"},"maintenance":{"status":"active"},"meta":{"latest_release":"2026-07-27","license_spdx":null,"license_treatment":"permissive","name":"fla-core","python_support":"supports_current","summary":"Core operations for flash-linear-attention"},"popularity":{"monthly_downloads":696013,"position":5306,"tier":"top_15000"},"security":{"n_vulnerabilities":0},"version":"0.5.2"}