How to install Torch.rb on Ubuntu with GPU support?

First, install CUDA and cuDNN, then download the matching LibTorch version for your CUDA setup, configure the gem with the LibTorch path, and reinstall. The README provides specific commands and notes on checking CUDA availability with Torch::CUDA.available?

Torch.rb vs PyTorch: which is better for Ruby projects?

Torch.rb is ideal if you're committed to Ruby and need deep learning within that ecosystem, offering a similar API. However, PyTorch in Python has a larger community, more models, and better documentation, so switch if ecosystem breadth is critical.

Does Torch.rb work on Mac M1 or M2 chips?

Yes, it supports Apple silicon via Metal Performance Shaders (MPS). Check availability with Torch::Backends::MPS.available? and move models to the GPU using net.to(device) with Torch.device("mps"), as explained in the Mac performance section.

Can I use pre-trained PyTorch models with Torch.rb?

Yes, but with caveats. You can load PyTorch-saved state dictionaries, but due to LibTorch bugs, you need to convert parameters to tensors in Python first, as described in the saving and loading models section.

What is the performance difference between Torch.rb and Python PyTorch?

Performance is comparable since both use LibTorch for tensor computations. The main differences arise from Ruby's runtime overhead, but GPU acceleration via CUDA or MPS helps mitigate this, especially for training neural networks.

How to save and load models in Torch.rb for deployment?

Use Torch.save(net.state_dict, "net.pth") to save and Torch.load("net.pth") with net.load_state_dict to load. Ensure the model is in eval mode with net.eval, and follow the conversion steps if loading from Python.

Is Torch.rb suitable for production machine learning pipelines?

It can be, especially for Ruby-centric deployments, but consider the maturity: limited Windows support, dependency on LibTorch compilation, and a smaller ecosystem might require extra effort compared to Python-based solutions.

torch-rb — Ruby Deep Learning Library

What is torch-rb?

Torch.rb is a deep learning library for Ruby that provides a PyTorch-like API, powered by LibTorch. It allows Ruby developers to perform tensor computations, build neural networks, and leverage GPU acceleration for machine learning tasks directly within Ruby applications.

Target Audience

Ruby developers and data scientists who want to implement deep learning models without leaving the Ruby ecosystem, especially those familiar with PyTorch's API.

Value Proposition

It offers a seamless bridge between Ruby and PyTorch's performance, with an idiomatic Ruby API, GPU support, and compatibility with PyTorch tutorials and models.

Deep learning for Ruby, powered by LibTorch

Use Cases

Best For

Implementing neural networks in Ruby applications
Converting PyTorch models and tutorials to Ruby
Performing tensor computations with GPU acceleration in Ruby
Building deep learning prototypes quickly within Ruby environments
Integrating machine learning models into Ruby on Rails projects
Educational purposes for learning deep learning concepts in Ruby

Not Ideal For

Windows-based development environments, as Torch.rb explicitly does not support Windows platforms
Teams requiring access to a vast library of pre-trained models or cutting-edge research implementations, due to Ruby's smaller deep learning ecosystem compared to Python
Projects with strict deployment timelines and no existing Ruby infrastructure, where Python's PyTorch offers more mature tooling and community support
Applications needing straightforward, no-compilation installation, since Torch.rb requires manual LibTorch download and extension compilation

Pros & Cons

Pros

High-Performance Backend

Leverages LibTorch, PyTorch's C++ backend, for efficient tensor computations and GPU acceleration, matching PyTorch's core performance.

Ruby-Idiomatic API

Adapts PyTorch's API to Ruby conventions, using methods like add! for in-place operations and tensor? for boolean checks, making it intuitive for Ruby developers.

GPU Acceleration Support

Supports CUDA on Linux and Metal Performance Shaders on Apple silicon, enabling faster training on compatible hardware, as detailed in the performance section.

Seamless Numo Integration

Allows easy conversion between Torch tensors and Numo arrays with x.numo and Torch.from_numo, facilitating integration with Ruby's scientific computing stack.

PyTorch Tutorial Compatibility

Users can follow PyTorch tutorials and convert code to Ruby, as noted in the API section, reducing the learning curve for those familiar with PyTorch.

Cons

Platform Limitations

Excludes Windows support entirely, limiting its use to Linux and Mac environments, which is a significant barrier for some development teams.

Complex Installation Process

Requires manual download of LibTorch and compilation of extensions, taking 5-10 minutes and potentially causing setup errors, as highlighted in the installation instructions.

Limited Ecosystem Maturity

While companion gems exist, the overall deep learning ecosystem in Ruby is smaller, with fewer community resources and pre-trained models compared to Python's PyTorch.

Interoperability Hurdles

Loading models saved in Python requires extra steps to convert parameters due to bugs in LibTorch, adding friction to cross-language workflows, as admitted in the saving and loading section.

Frequently Asked Questions

What is torch-rb?

Target Audience

Ruby developers and data scientists who want to implement deep learning models without leaving the Ruby ecosystem, especially those familiar with PyTorch's API.

Value Proposition

It offers a seamless bridge between Ruby and PyTorch's performance, with an idiomatic Ruby API, GPU support, and compatibility with PyTorch tutorials and models.

Use Cases

Best For

Implementing neural networks in Ruby applications
Converting PyTorch models and tutorials to Ruby
Performing tensor computations with GPU acceleration in Ruby
Building deep learning prototypes quickly within Ruby environments
Integrating machine learning models into Ruby on Rails projects
Educational purposes for learning deep learning concepts in Ruby

Not Ideal For

Windows-based development environments, as Torch.rb explicitly does not support Windows platforms
Teams requiring access to a vast library of pre-trained models or cutting-edge research implementations, due to Ruby's smaller deep learning ecosystem compared to Python
Projects with strict deployment timelines and no existing Ruby infrastructure, where Python's PyTorch offers more mature tooling and community support
Applications needing straightforward, no-compilation installation, since Torch.rb requires manual LibTorch download and extension compilation

Pros & Cons

Pros

High-Performance Backend

Leverages LibTorch, PyTorch's C++ backend, for efficient tensor computations and GPU acceleration, matching PyTorch's core performance.

Ruby-Idiomatic API

Adapts PyTorch's API to Ruby conventions, using methods like add! for in-place operations and tensor? for boolean checks, making it intuitive for Ruby developers.

GPU Acceleration Support

Supports CUDA on Linux and Metal Performance Shaders on Apple silicon, enabling faster training on compatible hardware, as detailed in the performance section.

Seamless Numo Integration

Allows easy conversion between Torch tensors and Numo arrays with x.numo and Torch.from_numo, facilitating integration with Ruby's scientific computing stack.

PyTorch Tutorial Compatibility

Users can follow PyTorch tutorials and convert code to Ruby, as noted in the API section, reducing the learning curve for those familiar with PyTorch.

Cons

Platform Limitations

Excludes Windows support entirely, limiting its use to Linux and Mac environments, which is a significant barrier for some development teams.

Complex Installation Process

Requires manual download of LibTorch and compilation of extensions, taking 5-10 minutes and potentially causing setup errors, as highlighted in the installation instructions.

Limited Ecosystem Maturity

While companion gems exist, the overall deep learning ecosystem in Ruby is smaller, with fewer community resources and pre-trained models compared to Python's PyTorch.

Interoperability Hurdles

Loading models saved in Python requires extra steps to convert parameters due to bugs in LibTorch, adding friction to cross-language workflows, as admitted in the saving and loading section.

Frequently Asked Questions

torch-rb

What is torch-rb?

Overview

Use Cases

Best For

Not Ideal For

Pros & Cons

Pros

Cons

Frequently Asked Questions

Related Projects

Found a gem we're missing?

torch-rb

What is torch-rb?

Overview

Use Cases

Best For

Not Ideal For

Pros & Cons

Pros

Cons

Frequently Asked Questions

Related Projects

Found a gem we're missing?