TiRex-2
NX-AI/tirex-2Pretrained multivariate time series forecasting model with past and future-known covariates and quantile forecasts, released for zero-shot use.
pip install tirex-2
Source code, pretrained models and documentation for xLSTM and TiRex, released by NXAI GmbH, Linz, Austria. Each project links to its repository, paper and, where available, documentation and packages.
Pretrained multivariate time series forecasting model with past and future-known covariates and quantile forecasts, released for zero-shot use.
pip install tirex-2
Zero-shot univariate forecasting model for long and short horizons with enhanced in-context learning, based on xLSTM.
pip install tirex-ts
Official PyTorch implementation of the xLSTM architecture, including the xLSTM 7B language model.
pip install xlstm
Official JAX implementation of xLSTM with efficient training and inference code.
Vision-LSTM (ViL) adapts xLSTM as a generic backbone for computer vision. Includes pretrained models and a full training pipeline.
Code and data for the scaling behaviour of xLSTM and Transformer models from 80M to 7B parameters.
Tiled Flash Linear Attention: kernel library for fast and efficient mLSTM training and inference.
pip install mlstm_kernels
CUDA and Triton kernels for LSTM, GRU, Elman networks and sLSTM, optimised for modern GPU hardware.
pip install flashrnn
Triton kernels and a PyTorch layer for KronLSTM, a Kronecker delta rule for long-context linear attention.
Efficient normalisation via shared RMS statistics, reducing the number of operations on the critical path of each Transformer block.
| Project | GitHub stars | PyPI downloads | Hugging Face downloads |
|---|---|---|---|
| xLSTMNX-AI/xlstm | 2,214 | 364,39431,069 last 30 days | 181,6633,512 last 30 days |
| Vision-LSTMNX-AI/vision-lstm | 491 | – | – |
| TiRexNX-AI/tirex | 307 | 100,1678,889 last 30 days | 3,915,236148,549 last 30 days |
| FlashRNNNX-AI/flashrnn | 190 | 71,15730,865 last 30 days | – |
| TiRex-2NX-AI/tirex-2 | 157 | 45,75024,861 last 30 days | 22,7576,469 last 30 days |
| xLSTM JAXNX-AI/xlstm-jax | 105 | – | – |
| mLSTM KernelsNX-AI/mlstm_kernels | 96 | 257,19837,655 last 30 days | – |
| KronLSTMNX-AI/KronLSTM | 62 | – | 333333 last 30 days |
| xLSTM Scaling LawsNX-AI/xlstm_scaling_laws | 24 | – | 00 last 30 days |
| CoupledNormNX-AI/coupled-norm | 0 | – | – |