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
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
|
{
description = "Port of Facebook's LLaMA model in C/C++";
inputs = {
nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
flake-parts.url = "github:hercules-ci/flake-parts";
};
# For inspection, use `nix flake show github:ggerganov/llama.cpp` or the nix repl:
#
# ```bash
# ❯ nix repl
# nix-repl> :lf github:ggerganov/llama.cpp
# Added 13 variables.
# nix-repl> outputs.apps.x86_64-linux.quantize
# { program = "/nix/store/00000000000000000000000000000000-llama.cpp/bin/quantize"; type = "app"; }
# ```
outputs =
{ self, flake-parts, ... }@inputs:
let
# We could include the git revisions in the package names but those would
# needlessly trigger rebuilds:
# llamaVersion = self.dirtyShortRev or self.shortRev;
# Nix already uses cryptographic hashes for versioning, so we'll just fix
# the fake semver for now:
llamaVersion = "0.0.0";
in
flake-parts.lib.mkFlake { inherit inputs; }
{
imports = [
.devops/nix/nixpkgs-instances.nix
.devops/nix/apps.nix
.devops/nix/devshells.nix
.devops/nix/jetson-support.nix
];
# An overlay can be used to have a more granular control over llama-cpp's
# dependencies and configuration, than that offered by the `.override`
# mechanism. Cf. https://nixos.org/manual/nixpkgs/stable/#chap-overlays.
#
# E.g. in a flake:
# ```
# { nixpkgs, llama-cpp, ... }:
# let pkgs = import nixpkgs {
# overlays = [ (llama-cpp.overlays.default) ];
# system = "aarch64-linux";
# config.allowUnfree = true;
# config.cudaSupport = true;
# config.cudaCapabilities = [ "7.2" ];
# config.cudaEnableForwardCompat = false;
# }; in {
# packages.aarch64-linux.llamaJetsonXavier = pkgs.llamaPackages.llama-cpp;
# }
# ```
#
# Cf. https://nixos.org/manual/nix/unstable/command-ref/new-cli/nix3-flake.html?highlight=flake#flake-format
flake.overlays.default =
(final: prev: {
llamaPackages = final.callPackage .devops/nix/scope.nix { inherit llamaVersion; };
inherit (final.llamaPackages) llama-cpp;
});
systems = [
"aarch64-darwin"
"aarch64-linux"
"x86_64-darwin" # x86_64-darwin isn't tested (and likely isn't relevant)
"x86_64-linux"
];
perSystem =
{
config,
lib,
system,
pkgs,
pkgsCuda,
pkgsRocm,
...
}:
{
# Unlike `.#packages`, legacyPackages may contain values of
# arbitrary types (including nested attrsets) and may even throw
# exceptions. This attribute isn't recursed into by `nix flake
# show` either.
#
# You can add arbitrary scripts to `.devops/nix/scope.nix` and
# access them as `nix build .#llamaPackages.${scriptName}` using
# the same path you would with an overlay.
legacyPackages = {
llamaPackages = pkgs.callPackage .devops/nix/scope.nix { inherit llamaVersion; };
llamaPackagesCuda = pkgsCuda.callPackage .devops/nix/scope.nix { inherit llamaVersion; };
llamaPackagesRocm = pkgsRocm.callPackage .devops/nix/scope.nix { inherit llamaVersion; };
};
# We don't use the overlay here so as to avoid making too many instances of nixpkgs,
# cf. https://zimbatm.com/notes/1000-instances-of-nixpkgs
packages =
{
default = config.legacyPackages.llamaPackages.llama-cpp;
}
// lib.optionalAttrs pkgs.stdenv.isLinux {
opencl = config.packages.default.override { useOpenCL = true; };
cuda = config.legacyPackages.llamaPackagesCuda.llama-cpp;
mpi-cpu = config.packages.default.override { useMpi = true; };
mpi-cuda = config.packages.default.override { useMpi = true; };
}
// lib.optionalAttrs (system == "x86_64-linux") {
rocm = config.legacyPackages.llamaPackagesRocm.llama-cpp;
};
# Packages exposed in `.#checks` will be built by the CI and by
# `nix flake check`. Currently we expose all packages, but we could
# make more granular choices
checks = config.packages;
};
};
}
|
