This is the linux-2_17-x86_64 build, made for older systems (glibc 2.17
and newer -- CentOS 7 vintage).  It runs anywhere, at the cost of a few
features that need a newer toolchain:

  * rampart-onnx is NOT included in this build.  Use the embedding and
    reranking support in rampart-langtools instead.
  * rampart-ocr is NOT included either -- it runs its models through
    rampart-onnx, so it needs that module to be present.
  * Of the GPU builds, only CUDA 11 is available here.

If your system has glibc 2.28 or newer, re-run the installer to get the
linux-2_28-x86_64 build, which includes rampart-onnx, rampart-ocr, and
CUDA 12 / 13 as well as CUDA 11.

Optional packages
-----------------
  rampart --install --list      see what is available for this build
  rampart --install all         install everything except the opt-in
                                langtools variant below

Local LLM / vector search (rampart-langtools)
---------------------------------------------
`rampart --install all` installs the CPU build.  It works everywhere and
needs nothing else.

A CUDA 11 build of llamacpp, faiss and clip is available separately:

  rampart --install rampart-langtools-cu11

Installing it re-points rampart-llamacpp.so, rampart-faiss.so and
rampart-clip.so at the GPU builds.  Re-install rampart-langtools to go
back to CPU.

  IMPORTANT: no NVIDIA libraries are bundled.  The GPU package needs a
  CUDA 11 toolkit and cuDNN 9 already installed on the system.

rampart-graphicsmagick
----------------------
The image module links against a system GraphicsMagick, which is not
bundled:

  sudo yum install GraphicsMagick        (CentOS / RHEL 7)
  sudo apt install graphicsmagick        (Debian / Ubuntu)

require('rampart-gm') prints the right command for your system if the
library is missing.
