onnx_load

Parses an .onnx file, builds a ggml computation graph, and allocates
tensors on the specified device. Weights are loaded via memory-mapped
file (zero-copy where possible).

Provides 'R' bindings to the 'GGML' tensor library for machine
learning, optimized for 'Vulkan' GPU acceleration with a transparent CPU
fallback. The package features a 'Keras'-like sequential API and a
'PyTorch'-style 'autograd' engine for building, training, and deploying
neural networks. Key capabilities include high-performance 5D tensor
operations, 'f16' precision, and efficient quantization. It supports
native 'ONNX' model import (50+ operators) and 'GGUF' weight loading
from the 'llama.cpp' and 'Hugging Face' ecosystems. Designed for
zero-overhead inference via dedicated weight buffering, it integrates
seamlessly as a 'parsnip' engine for 'tidymodels' and provides
first-class learners for the 'mlr3' framework.
See <https://github.com/ggml-org/ggml> for more information about the
underlying library.

Yuri Baramykov

ggmlR

'GGML' Tensor Operations for Machine Learning

onnx_load function

<dl><dt>path</dt>
<dd>Path to .onnx file.</dd>
<dt>device</dt>
<dd>Backend device: <code>"vulkan"</code> (default if available)
or <code>"cpu"</code>.</dd>
<dt>input_shapes</dt>
<dd>Optional named list of integer vectors specifying
fixed shapes for inputs with dynamic dimensions. Names must match
input tensor names. Each shape must include all dimensions including
batch, e.g. <code>list(image = c(1L, 3L, 224L, 224L))</code>.
Required when the model has dynamic dimensions and no default shape.</dd>
<dt>n_threads</dt>
<dd>Number of CPU threads. <code>NULL</code> (default) reads
<code>getOption("ggmlR.n_threads")</code>; if that is also unset, uses
<code>parallel::detectCores() - 1</code> (minimum 1).</dd>
<dt>dtype</dt>
<dd>Weight precision: <code>"f32"</code> (default) or <code>"f16"</code>.
When <code>"f16"</code>, large weight tensors (&gt;= 256 elements) are stored
in half-precision for faster Vulkan compute and lower VRAM usage.
Small tensors (bias, scalars, batch-norm params) remain in F32
for numerical stability. Inputs and outputs are always F32.</dd></dl>

Arguments

Load an ONNX model — onnx_load

<dl>

<dt>path</dt>
<dd>Path to .onnx file.</dd>


<dt>device</dt>
<dd>Backend device: <code>"vulkan"</code> (default if available)
or <code>"cpu"</code>.</dd>


<dt>input_shapes</dt>
<dd>Optional named list of integer vectors specifying
fixed shapes for inputs with dynamic dimensions. Names must match
input tensor names. Each shape must include all dimensions including
batch, e.g. <code>list(image = c(1L, 3L, 224L, 224L))</code>.
Required when the model has dynamic dimensions and no default shape.</dd>


<dt>n_threads</dt>
<dd>Number of CPU threads. <code>NULL</code> (default) reads
<code>getOption("ggmlR.n_threads")</code>; if that is also unset, uses
<code>parallel::detectCores() - 1</code> (minimum 1).</dd>


<dt>dtype</dt>
<dd>Weight precision: <code>"f32"</code> (default) or <code>"f16"</code>.
When <code>"f16"</code>, large weight tensors (&gt;= 256 elements) are stored
in half-precision for faster Vulkan compute and lower VRAM usage.
Small tensors (bias, scalars, batch-norm params) remain in F32
for numerical stability. Inputs and outputs are always F32.</dd>

</dl>

onnx_load: Load an ONNX model

Description

Usage

Value

Arguments