1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
|
Build Profiles
==============
This package supports a number of build profiles:
pkg.ggml.noblas: Disable the BLAS backend build
pkg.ggml.novulkan: Disable the Vulkan backend build
pkg.ggml.nohip: Disable the HIP build (very time consuming)
pkg.ggml.minhip: Build HIP only for gfx1030, saves time, useful for Salsa CI
(only active if pkg.ggml.nohip is inactive)
src:ggml-cuda
=============
The source package src:ggml can be converted to src:ggml-cuda (for building the
CUDA backend) by running the debian/cudabuild.sh in an unpacked source tree.
Note that you'll need to rename the .orig.tar.gz from ggml_ to ggml-cuda_
before building anything.
src:ggml-vendor-cuda-xx-y
=======================
In order to build against the Nvidia provided packages,
the process is similar as for src:ggml-cuda, that is
running debian/cudabuild-nvidia-cuda-xx-y.sh before building
(xx being the major CUDA version and y the minor CUDA version).
|
