# asappresearch/sru

Training RNNs as Fast as CNNs (https://arxiv.org/abs/1709.02755)

Repository: https://github.com/asappresearch/sru
Canonical: https://ross.abutalabs.com/products/sru
Language: Python
License: MIT
License Family: permissive
Topics: pytorch, deep-learning, recurrent-neural-networks, nlp
Last push: 2022-01-04T21:17:53+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 3292, "days_push": 1702, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2106, forks 304 (observed 2026-08-28T04:06:13.908308+00:00)

## What it is
SRU is a PyTorch library implementing the Simple Recurrent Unit, a highly parallelizable RNN that runs 10-16x faster than cuDNN LSTM with comparable accuracy. It also includes SRU++, a variant combining fast recurrence with attention for efficient language modeling.

## Use cases
- train RNNs as fast as CNNs on GPU
- replace LSTM with a faster recurrent unit in PyTorch models
- train language models with reduced compute
- speed up sequence modeling for NLP tasks
- build bidirectional recurrent layers with layer norm and highway connections

## When to choose
- you need recurrent sequence models but LSTM training speed is a bottleneck
- you want a drop-in PyTorch alternative to nn.LSTM with GPU acceleration
- you are training language models on limited compute budgets

## When to avoid
- you need transformer-only architectures or the latest attention variants
- you require active community development and frequent updates
- you work outside PyTorch or without CUDA GPUs

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, nlp
- domain: machine-learning, deep-learning
- platform: python, cross-platform
- tags: pytorch, recurrent-neural-networks, rnn, lstm-alternative, gpu-accelerated, language-modeling, sru, natural-language-processing, gpu, linux

## Member repositories
- asappresearch/sru (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:13.908308+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:53:59.066964+00:00, confidence not recorded.
  - readme: https://github.com/asappresearch/sru (fetched 2026-08-28T04:06:13.908308+00:00, sha 3866009c5f37)
- Data as of 2026-08-30T08:39:29.467469+00:00.