OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Xlstm

Xlstm is an open-source Python project implementing an extended long short-term memory recurrent neural-network architecture. Its code supports language modeling and token-based applications, integration into other architectures, and a separate optimized xLSTM Large 7B implementation.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Xlstm — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Xlstm when you want Python code for experimenting with an extended recurrent architecture, adding xLSTM components to a model, or examining the separate xLSTM Large implementation.

Who it is for

  • Researchers comparing recurrent architectures in language-modeling experiments.
  • Python developers integrating xLSTM blocks into other machine-learning architectures.
  • Developers exploring the xLSTM Large 7B model and its implementation.

Documented uses

  • Explore an extended LSTM architecture as an alternative in language-modeling research.
  • Use the xLSTM language-model component for token-based applications.
  • Integrate an xLSTM block stack into a non-language application or another architecture.
  • Initialize the xLSTM Large model and make a forward pass using the provided notebook example.
Before you choose

Requirements & limits

The repository is public and identifies the project as open source under Apache-2.0. The README links xLSTM Large 7B weights on Hugging Face.

  • The repository distinguishes the paper’s xLSTM architecture from the optimized xLSTM Large implementation.
  • The xLSTM Large 7B implementation requires the separate mlstm_kernels package.
  • The documented environment uses Conda and PyTorch; the repository specifies tested PyTorch versions of 1.8 or later.
  • The sLSTM CUDA kernels require NVIDIA Compute Capability 8.0 or higher, and CUDA and PyTorch versions must match in custom setups.
  • For Apple Metal, the repository recommends native PyTorch implementations; it identifies an MLX-native community port separately.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up the xLSTM Python package with Conda

Create the repository’s documented Conda environment and install the xLSTM package.

Before you start

  • Conda
  • The xLSTM repository, including environment_pt240cu124.yaml
  • PyTorch; the package is based on PyTorch and tested for versions 1.8 or later

Steps

  1. Create the environment with conda env create -n xlstm -f environment_pt240cu124.yaml.

  2. Activate it with conda activate xlstm.

  3. Install the package from PyPI with the command below, or clone the repository, enter its directory, and run the command below.

    Documented command
    pip install xlstm
    Documented command
    pip install -e .

Expected outcome

A Conda environment is created and the xLSTM package is installed using one of the repository’s documented installation routes.

Guide limitations

  • This setup guide does not run a model or establish compatibility with a particular machine.
  • The xLSTM Large 7B implementation additionally requires mlstm_kernels.
Documentation 1 ↗
Source-based practical guide

Initialize xLSTM Large and perform a forward pass

Follow the repository’s documented notebook example to configure and call an initialized xLSTM Large model.

Before you start

  • The xLSTM package and PyTorch
  • The separate mlstm_kernels package required by xLSTM Large
  • A CUDA device for the notebook’s shown configuration, which moves the model and inputs to cuda

Steps

  1. Open notebooks/xlstm_large/demo.ipynb in the repository.

  2. Import xLSTMLargeConfig and xLSTMLarge from xlstm.xlstm_large.model.

  3. Configure the model, instantiate it, and move it to CUDA as shown in the notebook.

  4. Create integer token inputs and call the model to obtain the forward-pass output.

Expected outcome

The documented example initializes a randomly initialized model and performs a forward pass; it is not a guide to loading the pretrained 7B weights.

Guide limitations

  • The example uses CUDA and Triton kernel settings; the repository gives separate native PyTorch kernel settings for other platforms.
  • The repository reports NVIDIA as its main tested GPU platform and says Triton kernels should also run on AMD GPUs; this is not a compatibility guarantee.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
11 October 2026
Original website title
NX-AI/xlstm · Official homepage title · retained in its source language
How we review sourcesSuggest a correction