state-spaces/s4

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2919 stars · 369 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official implementations of Structured State Space (S4) sequence models and related variants like S4D, HiPPO, and SaShiMi, built in PyTorch. It provides self-contained model layers, custom CUDA/KeOps kernels, and experiment reproduction code for research on long-sequence modeling.

Use cases

• train state space models on long sequences
• reproduce S4 paper experiments
• use S4 layer in my own pytorch model
• model long-range dependencies in audio or text
• compare transformers with state space sequence models
• generate audio waveforms with SaShiMi

When to choose

• you need efficient modeling of very long sequences (thousands to millions of tokens)
• you want the official, research-faithful S4/S4D implementations
• you're doing research on sequence models beyond transformers

When to avoid

• you need a production-ready pip-installable library (packaging is still on the roadmap)
• you want a plug-and-play high-level API rather than research code
• you're unwilling to compile custom CUDA kernels or set up PyKeOps for best performance

Facets

library · maturity active

machine-learning deep-learning nlp audio-processing deep-learning machine-learning python windows state-space-models sequence-modeling pytorch s4 research-code hippo time-series natural-language-processing audio gpu linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="state-spaces/s4")