fla-core
Core operations for flash-linear-attention
Decision gist · record as of 2026-08-14
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.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- Requires Python 3.10 or later; hardware acceleration benefits depend on NVIDIA, AMD, or Intel accelerator availability.
- Low install friction with a single pure-Python dependency (einops).
- Active maintenance with recent commits and steady releases; repository shows 5560 stars and receives regular updates.
License · maintenance · safety
permissive license (permissive) — MIT License permits free use, modification, and distribution with minimal restrictions, making it suitable for both open-source and commercial projects.
last release 2026-07-27 (18 days) · last repo commit 2026-08-14 · 5,560 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 696,013 downloads/mo, #5,306 on PyPI
Alternatives
Verify before relying
pip install fla-core
import fla
# Use a model implementation or kernel from fla.layers or fla.ops- Whether all advertised model implementations (RetNet, Mamba, GLA, DeltaNet, etc.) are production-ready or research prototypes
- Performance benchmarks and actual speedup claims compared to standard attention implementations
- Compatibility with specific PyTorch or CUDA versions beyond the Python 3.10+ requirement
What it is and 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—researchers 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.
The 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.
Use it for
- 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.
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
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.
Install
fla-core on PyPI
Before you install
Low install friction with a single pure-Python dependency (einops). Active maintenance with recent commits and steady releases; repository shows 5560 stars and receives regular updates.
Requires Python 3.10 or later; hardware acceleration benefits depend on NVIDIA, AMD, or Intel accelerator availability.
License in practice
MIT License permits free use, modification, and distribution with minimal restrictions, making it suitable for both open-source and commercial projects.
Quickstart
pip install fla-core
import fla
# Use a model implementation or kernel from fla.layers or fla.opsVerify before relying
- Whether all advertised model implementations (RetNet, Mamba, GLA, DeltaNet, etc.) are production-ready or research prototypes
- Performance benchmarks and actual speedup claims compared to standard attention implementations
- Compatibility with specific PyTorch or CUDA versions beyond the Python 3.10+ requirement
Package facts
| License | permissive license permissive |
| Python support | Supports the current Python release >=3.10 |
| Install friction | Low. Pure-Python wheel |
| Runtime dependencies | 1 packageeinops |
| Maintenance | Actively maintained 18 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 696,013 / month, #5,306 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
Evidence: fla_core-0.5.2-py3-none-any.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “linear attention kernels”
- fla-coreProvides hardware-efficient implementations of linear attention,…
- tokamaxTokamax provides custom accelerator kernels for JAX, including…
- nvidia-cudnn-frontendProvides Python and C++ APIs to NVIDIA's cuDNN library, exposing…
Give your agent the search over MCP, or paste the wish link into any chat.
More Artificial Intelligence packages
LiteLLM provides a unified Python interface to call 100+ LLM providers (OpenAI, Anthropic, Gemini, Bedrock, Azure, and others) using OpenAI-compatible API format, available as both a Python SDK and a self-hosted AI Gateway proxy server.
Install it if you need to work with multiple LLM providers or want to centralize LLM routing in your organization.
Client library and CLI tool for downloading, uploading, and managing models, datasets, and repositories on the Hugging Face Hub platform.
Install it if you work with Hugging Face Hub models or datasets.
LangChain provides a framework for building agents and LLM-powered applications by composing language models, tools, and memory through a unified API that abstracts over multiple model providers.
hf-xet provides chunk-based deduplication and efficient file transfer for the Hugging Face Hub, enabling faster uploads and downloads of large files with local disk caching.
Tokenizers converts raw text into token sequences for NLP models, with support for training custom vocabularies and using pre-built tokenizers (BPE, WordPiece) optimized for speed via Rust.
Transformers provides a unified framework for loading, fine-tuning, and running state-of-the-art pretrained models across text, vision, audio, video, and multimodal tasks using PyTorch, JAX, or TensorFlow.
Install it if you need to run or train any transformer-based model for NLP, vision, audio, or multimodal tasks.
See also bytesparse · sageattention · xformers · flashinfer-python · nvidia-cudnn-frontend · local-attention · ring-flash-attn · cpm-kernels · tokamax · flash-attn