Skip to content

CLI-REGISTRY.md

Latest commit

 

History

History
269 lines (193 loc) · 7.34 KB

CLI-REGISTRY.md

File metadata and controls

269 lines (193 loc) · 7.34 KB

CLI & Model Registry

Install, search, and manage models from the command line using the openmodelstudio CLI. Models are published in the Open Model Registry, a public GitHub repository that acts as a decentralized model package manager.

Installation

pip install openmodelstudio

This installs both the Python SDK and the openmodelstudio CLI command.

Commands

Search for Models

openmodelstudio search classification

Output:

NAME              VERSION  FRAMEWORK  CATEGORY        DESCRIPTION
---------------   -------  ---------  --------------  -------------------------------------------
iris-svm          1.0.0    sklearn    classification  Support Vector Machine classifier for the...
titanic-rf        1.0.0    sklearn    classification  Random Forest classifier for Titanic surv...

Filter by framework or category:

openmodelstudio search cnn --framework pytorch
openmodelstudio search "" --category nlp
openmodelstudio search "" --framework sklearn --category classification

Browse All Registry Models

openmodelstudio registry

Output:

NAME              VERSION  FRAMEWORK  CATEGORY        AUTHOR            DESCRIPTION
---------------   -------  ---------  --------------  ----------------  ---------------------------
iris-svm          1.0.0    sklearn    classification  openmodelstudio   Support Vector Machine cla...
mnist-cnn         1.0.0    pytorch    computer-vision openmodelstudio   Convolutional Neural Netwo...
sentiment-lstm    1.0.0    pytorch    nlp             openmodelstudio   Bidirectional LSTM for tex...
timeseries-arima  1.0.0    python     time-series     openmodelstudio   ARIMA model for univariate...
titanic-rf        1.0.0    sklearn    classification  openmodelstudio   Random Forest classifier f...

Get Model Details

openmodelstudio info mnist-cnn

Output:

Name:        mnist-cnn
Version:     1.0.0
Author:      openmodelstudio
Framework:   pytorch
Category:    computer-vision
License:     MIT
Description: Convolutional Neural Network for MNIST digit classification.
Tags:        image-classification, cnn, mnist, beginner, deep-learning
Dependencies: torch>=2.0, torchvision>=0.15, numpy>=1.24
Homepage:    https://github.com/GACWR/open-model-registry

Install a Model

openmodelstudio install titanic-rf

Output:

Installing 'titanic-rf' from registry...
Installed to /home/user/.openmodelstudio/models/titanic-rf

This downloads the model files and a model.json manifest to your local models directory. The model is then available for import and registration with the platform.

Force-reinstall an existing model:

openmodelstudio install titanic-rf --force

List Installed Models

openmodelstudio list

Output:

NAME        VERSION  FRAMEWORK  PATH
----------  -------  ---------  -------------------------------------------
titanic-rf  1.0.0    sklearn    /home/user/.openmodelstudio/models/titanic-rf
mnist-cnn   1.0.0    pytorch    /home/user/.openmodelstudio/models/mnist-cnn

Uninstall a Model

openmodelstudio uninstall titanic-rf

Using an Installed Model

After installing, use oms.use_model() to load the model and register it in your project:

import openmodelstudio as oms

# Load the installed registry model
iris = oms.use_model("iris-svm")

# Register it in your project under any name
handle = oms.register_model("my-iris", model=iris)
print(handle)

# Train it
job = oms.start_training(handle.model_id, wait=True)
print(f"Training: {job['status']}")

use_model() resolves models via the platform API, so it works inside workspace containers (K8s pods) without requiring filesystem access. If the model isn't installed yet, it auto-installs from the registry.

You can also install directly from the UI on the Model Registry page (sidebar > Develop > Model Registry). Each model card shows an Installed or Not Installed badge.

Configuration

View Current Config

openmodelstudio config

Output:

registry_url: https://raw.githubusercontent.com/GACWR/open-model-registry/main/registry/index.json
models_dir: /home/user/.openmodelstudio/models

Change Registry URL

Point to a custom registry (your own fork, a private registry, etc.):

openmodelstudio config set registry_url https://raw.githubusercontent.com/myorg/my-registry/main/registry/index.json

Or set via environment variable:

export OPENMODELSTUDIO_REGISTRY_URL="https://raw.githubusercontent.com/myorg/my-registry/main/registry/index.json"

Change Models Directory

openmodelstudio config set models_dir /opt/models

Or set via environment variable:

export OPENMODELSTUDIO_MODELS_DIR="/opt/models"

Python SDK (Programmatic Access)

All CLI commands are available as Python functions:

import openmodelstudio as oms

# Search
results = oms.registry_search("classification")
results = oms.registry_search("cnn", framework="pytorch")
results = oms.registry_search("", category="nlp")

# List all
models = oms.registry_list()

# Get info
info = oms.registry_info("titanic-rf")
print(info["description"])
print(info["dependencies"])

# Install
path = oms.registry_install("titanic-rf")
path = oms.registry_install("mnist-cnn", force=True)

# Use an installed model (works in workspace containers)
iris = oms.use_model("iris-svm")
handle = oms.register_model("my-iris", model=iris)

# Uninstall (removes locally + unregisters from platform)
oms.registry_uninstall("titanic-rf")

# List installed
installed = oms.list_installed()

# Switch registry
oms.set_registry("https://raw.githubusercontent.com/myorg/my-registry/main/registry/index.json")

How the Registry Works

The Open Model Registry is a GitHub repository with this structure:

open-model-registry/
  models/
    iris-svm/
      model.py           # Model code (train + infer functions)
    mnist-cnn/
      model.py
    sentiment-lstm/
      model.py
    ...
  registry/
    index.json           # Aggregated metadata for all models
  scripts/
    build_index.py       # Generates index.json from model directories

Each model directory contains:

  • model.py -- the model code following the train(ctx) / infer(ctx) interface
  • Additional files as needed (configs, weights, etc.)

The registry/index.json is an aggregated index with metadata for every model (name, version, description, framework, category, tags, dependencies, file list). Both the CLI and the web UI read this single JSON file to discover available models.

Using a Custom Registry

  1. Fork open-model-registry
  2. Add your model directories under models/
  3. Run python scripts/build_index.py to regenerate index.json
  4. Push to your fork
  5. Point the CLI or SDK to your fork's raw URL

Web UI

The Model Registry page in the sidebar (Develop > Model Registry) provides:

  • Browse all models with search and category/framework filters
  • Click any model card to view full details, source code, dependencies, and tags
  • Install models directly into a project from the UI
  • Link to the model's GitHub page

Each model detail page shows:

  • Full description
  • Source code viewer (Monaco editor, read-only)
  • Tags, dependencies, license, author
  • Quick install command (click to copy)
  • Install-to-project dialog