Conversion to Other Array Frameworks

Conversion to Other Array Frameworks#

MLX arrays can interoperate with frameworks that support buffer sharing/DLPack. Use a Ruby-backed bridge (for example a DLPack adapter) to convert between MLX arrays and arrays in other ecosystems.

From MLX to another framework#

require "mlx"
mx = MLX::Core

module Bridge
  module_function
  def from_mlx(x)
    x.to_a
  end
end

x = mx.arange(3)

# Convert MLX -> external framework using its bridging helper
external = Bridge.from_mlx(x)
# Convert back
x_round_trip = mx.array(external)

From another framework back to MLX#

# x_other should be an array-like value from your target framework
x_other = [1.0, 2.0, 3.0]
x = mx.array(x_other)

Note

Many interop helpers currently flow through Python tooling in the wider MLX ecosystem. In Ruby, keep conversions focused on adapters available for your selected bridge library.

PyTorch#

When using a Torch-backed bridge, treat tensors as external arrays and avoid in-place mutation from non-MLX contexts while gradients are being computed.

# Ruby: require your Torch bridge and load a tensor
# then pass it through mx.array(...)

JAX#

For JAX workflows, use the same interoperability path through the adapter layer you use for NumPy-compatible buffers.

TensorFlow#

For TensorFlow workflows, follow your tensor bridge’s DLPack/array conversion API.

# TensorFlow tensor -> external buffer -> MLX
# x_tf = ...
# x = mx.array(Bridge.from_tensor(x_tf))
Contents