Practical overview
When it fits
Use ML Forge when you want to build an image-classification training pipeline through a node-based desktop interface, inspect its training run, or export the result as PyTorch code.
Who it is for
- People who want to assemble image-classification workflows visually rather than write the pipeline code by hand.
- PyTorch users who want a visual way to connect data preparation, model layers, and training components.
- Developers who want to export a visual pipeline as a standalone PyTorch training script.
Documented uses
- Prepare supported image datasets and connect transforms and data loaders.
- Assemble a neural-network model from layer nodes and link it to a loss and optimizer.
- Train a model while viewing live loss curves and saving checkpoints.
- Inspect training metrics and run checkpoint-based inference with top-k predictions.
- Save an editor project or export it as a standalone PyTorch training script.
Requirements & limits
The repository documents installation through pip and identifies an MIT license. Training requires a separate PyTorch installation.
- PyTorch 2.0 or newer and torchvision must be installed separately for training.
- The documented workflow centers on image classification and the listed datasets.
- The editor can use CUDA when available and also documents CPU and Apple MPS support.
- Exported projects produce a standalone PyTorch training script; saved editor projects use the .mlf format.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install and open ML Forge
Install the package and launch the editor from a command prompt.
Before you start
- Python 3.10 or newer
- PyTorch 2.0 or newer and torchvision
- PyTorch is required for training and is not installed as a dependency.
Steps
Install PyTorch and torchvision with the command below.
pip install torch torchvisionInstall ML Forge with the command below.
pip install zaina-ml-forgeLaunch it with
ml-forge.
Expected outcome
The ML Forge visual editor opens for building a pipeline.
Guide limitations
- The documented install and launch instructions do not establish that training has been run successfully.
Build and train an image-classification pipeline
Assemble a dataset, model, and training flow in the editor and start a training run.
Before you start
- A running ML Forge editor
- Python 3.10 or newer, PyTorch 2.0 or newer, and torchvision for training
- A supported dataset: MNIST, FashionMNIST, CIFAR-10, CIFAR-100, or ImageFolder
Steps
In Data Prep, add a Dataset node, chain transforms including the required ToTensor, and end with a training DataLoader; add a validation chain with
train=Falsefor validation.In Model, connect an Input node through layer nodes to an Output node; input shape and class count are filled from the dataset.
In Training, wire DataLoader images to Model images, Model predictions to Loss pred, DataLoader labels to Loss target, and Loss loss to Optimizer params.
Configure epochs, device, checkpointing, and early stopping, then press RUN.
Expected outcome
The editor generates and runs the connected training pipeline; live loss curves are available during training.
Guide limitations
- The guide describes image classification and the listed datasets; it does not document a general custom-block workflow.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 7 October 2026
- Original website title
- zaina-ml/ml_forge · Official homepage title · retained in its source language