Package 'Rggml'

Title: Vendored 'GGML' Tensor Library with C-Callable Compute API
Description: Vendors the CPU backend of the 'GGML' tensor library (<https://github.com/ggml-org/ggml>) as a static library and exposes its core tensor-context and matrix-multiply compute path through 'R_RegisterCCallable' C-callable entry points. This is a carrier package: it has no high-level R modeling API of its own. Other R packages link to it ('LinkingTo') to build and compute 'GGML' tensor graphs (including quantized types such as Q4_K) from their own C or C++ code without re-vendoring 'GGML'. The CPU and BLAS backends are built (dense products offload to R's own BLAS); GPU ('Vulkan') support is a possible future addition.
Authors: Sounkou Mahamane Toure [aut, cre], Georgi Gerganov [cph] (Author of the GGML library), The ggml.ai / llama.cpp contributors [cph] (GGML CPU backend contributors), Yuri Baramykov [ctb] (ggmlR: source of the vendored GGML copy and its R/CRAN-compliance I/O shim, adapted here)
Maintainer: Sounkou Mahamane Toure <[email protected]>
License: GPL (>= 2)
Version: 0.1.0
Built: 2026-07-07 21:24:06 UTC
Source: https://github.com/sounkou-bioinfo/Rfmalloc

Help Index


Vendored 'GGML' library version

Description

Returns the version string reported by the vendored 'GGML' library at runtime (ggml_version()), resolved through Rggml's own registered C-callable rather than a compile-time constant, so it always reflects the code that was actually built.

Usage

ggml_version()

Value

A length-1 character vector.

Examples

ggml_version()