OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Developer Tools & SDKs

Framework

Framework is a TypeScript machine-learning library backed by Rust native code, providing tensor operations, automatic differentiation, neural-network modules and training optimizers.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Framework — OpenViST project profile in AI Developer Tools & SDKs.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Framework when you want a TypeScript API for model experiments or educational work involving tensors, autograd and neural-network training, with a documented Rust backend option for CPU, CUDA or WebGPU.

Who it is for

  • TypeScript developers building or experimenting with machine-learning models.
  • Learners exploring automatic differentiation and how ML frameworks work internally.
  • Developers interested in CPU or GPU-backed tensor and neural-network operations.

Documented uses

  • Create tensors and perform arithmetic, matrix operations, convolutions, pooling and reductions.
  • Define neural-network layers, compute losses, backpropagate gradients and update parameters.
  • Build experiments using the CPU backend or a selected CUDA or WebGPU backend.
Before you choose

Requirements & limits

The repository specifies an MIT license. The README documents npm installation; source builds require Rust, and CUDA builds additionally require the CUDA toolkit.

  • The API is TypeScript, while native operations are backed by Rust through an N-API bridge.
  • CPU is the default backend; CUDA and WebGPU require selecting corresponding build feature flags, which are mutually exclusive at compile time.
  • The documented API includes SGD and Adam optimizers, with AdamW and learning-rate scheduling also listed.
  • The repository describes the project as built to understand ML frameworks and models internally.

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

Install Framework and prepare a CPU native build

Install the npm package and build the native CPU backend from source.

Before you start

  • Node.js package tooling capable of running npm commands.
  • Rust, when building the native code from source.

Steps

  1. Install the package with the command below.

    Documented command
    npm install @mni-ml/framework

  2. For a source build, run the command below for the default CPU backend.

    Documented command
    npm run build:native

  3. Run the command below to build the TypeScript layer.

    Documented command
    npm run build

Expected outcome

The documented package installation and CPU/TypeScript build steps are complete.

Guide limitations

  • The source-build instructions are intended for contributors or custom builds.
  • This setup recipe does not cover running a model or verifying a successful build.
Documentation 1 ↗
Source-based practical guide

Build a small neural-network training flow

Use the documented API stages to define a two-layer model, calculate loss, backpropagate and update parameters.

Before you start

  • The @mni-ml/framework package installed in a TypeScript project.
  • Input tensors and target labels shaped for the model.

Steps

  1. Create input data with Tensor and define layers with Linear.

  2. Pass the input through the layers and calculate a loss with crossEntropyLoss.

  3. Call loss.backward() to run the backward pass.

  4. Collect layer parameters, create an Adam optimizer, then call step() and zeroGrad().

Expected outcome

The documented code flow computes a loss, propagates gradients and applies an optimizer update.

Guide limitations

  • The example uses placeholder target labels and does not provide a dataset or a complete training loop.
  • The recipe follows the README's example; it does not represent an execution test.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
mni-ml/framework · Official homepage title · retained in its source language
How we review sourcesSuggest a correction