Dala.Ml.Nx (dala v0.3.2)

Nx integration helpers for Dala.

Nx itself is pure Elixir and works on any platform. This module provides backend selection logic and inference helpers.

Summary

Functions

axon_available?()

Checks if Axon is available.

default_backend()

Returns the default backend for the current platform.

emlx_available?()

Checks if EMLX is available.

inference(model, params, input_data)

Runs inference with an Axon model using the best available backend.

init()

Initializes Nx with the best available backend for the current platform.

tensor(data, opts \\ [])

Creates a tensor with the platform-appropriate backend.

Functions

axon_available?()

@spec axon_available?() :: boolean()

Checks if Axon is available.

default_backend()

@spec default_backend() :: module() | tuple()

Returns the default backend for the current platform.

emlx_available?()

@spec emlx_available?() :: boolean()

Checks if EMLX is available.

inference(model, params, input_data)

@spec inference(term(), term(), term()) :: {:ok, term()} | {:error, term()}

Runs inference with an Axon model using the best available backend.

init()

@spec init() :: :emlx | :nx_binary

Initializes Nx with the best available backend for the current platform.

Priority:

EMLX (if available) - best for Apple Silicon
Nx.BinaryBackend (pure Elixir fallback)

tensor(data, opts \\ [])

@spec tensor(
  term(),
  keyword()
) :: Nx.Tensor.t()

Creates a tensor with the platform-appropriate backend.