AtaraxiaDev/nixos-config
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
nixos-config/patches/ollama-0.2.8.patch
Dmitriy Kholkin c3b15d99ec
upgrade system
2024-08-04 13:39:08 +03:00

512 lines
17 KiB
Diff
Raw Blame History

From 26a309c6d6d7b62af1fa7a3ba88f5af0e854b2da Mon Sep 17 00:00:00 2001
From: Pol Dellaiera <pol.dellaiera@protonmail.com>
Date: Tue, 23 Jul 2024 10:15:42 +0200
Subject: [PATCH 1/4] ollama: 0.2.7 -> 0.2.8
changelog: https://github.com/ollama/ollama/releases/tag/v0.2.8
diff: https://github.com/ollama/ollama/compare/v0.2.7..v0.2.8
---
pkgs/by-name/ol/ollama/package.nix | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/pkgs/by-name/ol/ollama/package.nix b/pkgs/by-name/ol/ollama/package.nix
index 6dd421ec6176cf3..05b33ae57241ad7 100644
--- a/pkgs/by-name/ol/ollama/package.nix
+++ b/pkgs/by-name/ol/ollama/package.nix
@@ -32,13 +32,13 @@
let
pname = "ollama";
# don't forget to invalidate all hashes each update
- version = "0.2.7";
+ version = "0.2.8";
src = fetchFromGitHub {
owner = "ollama";
repo = "ollama";
rev = "v${version}";
- hash = "sha256-YHBGS615dxz6pZR+8awMinp+ZRf3J8Re5BVeDEIyt2E=";
+ hash = "sha256-uir/GoFs0nhhHpznyYkwVjAUWxwjWmGZ7liU2htyQ04=";
fetchSubmodules = true;
};
@@ -52,11 +52,11 @@ let
(preparePatch "02-clip-log.diff" "sha256-rMWbl3QgrPlhisTeHwD7EnGRJyOhLB4UeS7rqa0tdXM=")
(preparePatch "03-load_exception.diff" "sha256-NJkT/k8Mf8HcEMb0XkaLmyUNKV3T+384JRPnmwDI/sk=")
(preparePatch "04-metal.diff" "sha256-bPBCfoT3EjZPjWKfCzh0pnCUbM/fGTj37yOaQr+QxQ4=")
- (preparePatch "05-default-pretokenizer.diff" "sha256-50+mzQBQZmYEhYvARHw/dliH0M/gDOYm2uy/yJupDF4=")
- (preparePatch "06-qwen2.diff" "sha256-FdDqEIblPy47z3yavKUnaV93Yk+3oboEzj4vHq+R66M=")
- (preparePatch "07-embeddings.diff" "sha256-lqg2SI0OapD9LCoAG6MJW6HIHXEmCTv7P75rE9yq/Mo=")
- (preparePatch "08-clip-unicode.diff" "sha256-1qMJoXhDewxsqPbmi+/7xILQfGaybZDyXc5eH0winL8=")
- (preparePatch "09-pooling.diff" "sha256-7meKWbr06lbVrtxau0AU9BwJ88Z9svwtDXhmHI+hYBk=")
+ (preparePatch "05-default-pretokenizer.diff" "sha256-Mgx+xi59rz3d5yEXp90QPQMiUr9InlA0Wo1mOSuRcec=")
+ (preparePatch "06-embeddings.diff" "sha256-lqg2SI0OapD9LCoAG6MJW6HIHXEmCTv7P75rE9yq/Mo=")
+ (preparePatch "07-clip-unicode.diff" "sha256-1qMJoXhDewxsqPbmi+/7xILQfGaybZDyXc5eH0winL8=")
+ (preparePatch "08-pooling.diff" "sha256-7meKWbr06lbVrtxau0AU9BwJ88Z9svwtDXhmHI+hYBk=")
+ (preparePatch "09-lora.diff" "sha256-HVDYiqNkuWO9K7aIiT73iiMj5lxMsJC1oqIG4madAPk=")
];
preparePatch = patch: hash: fetchpatch {
From ef6a3aec105ecdf21fb3d0d92f1e22f80419d82a Mon Sep 17 00:00:00 2001
From: Pol Dellaiera <pol.dellaiera@protonmail.com>
Date: Tue, 23 Jul 2024 10:29:48 +0200
Subject: [PATCH 2/4] ollama: reformat with `nixfmt-rfc-style`
---
.git-blame-ignore-revs | 3 +
pkgs/by-name/ol/ollama/package.nix | 320 +++++++++++++++--------------
2 files changed, 169 insertions(+), 154 deletions(-)
diff --git a/.git-blame-ignore-revs b/.git-blame-ignore-revs
index 012889bb2e8247f..ae22633837b6fe5 100644
--- a/.git-blame-ignore-revs
+++ b/.git-blame-ignore-revs
@@ -141,3 +141,6 @@ acd0e3898feb321cb9a71a0fd376f1157d0f4553
# poptracker: format with nixfmt-rfc-style (#326697)
ff5c8f6cc3d1f2e017e86d50965c14b71f00567b
+
+# ollama: reformat with nixfmt-rfc-style (#329353)
+8f91c4628a6b2e3f486c150c15b7407214d9b9de
diff --git a/pkgs/by-name/ol/ollama/package.nix b/pkgs/by-name/ol/ollama/package.nix
index 05b33ae57241ad7..7b1f2c88e92e591 100644
--- a/pkgs/by-name/ol/ollama/package.nix
+++ b/pkgs/by-name/ol/ollama/package.nix
@@ -1,32 +1,33 @@
-{ lib
-, buildGoModule
-, fetchFromGitHub
-, fetchpatch
-, buildEnv
-, linkFarm
-, overrideCC
-, makeWrapper
-, stdenv
-, addDriverRunpath
-
-, cmake
-, gcc12
-, clblast
-, libdrm
-, rocmPackages
-, cudaPackages
-, darwin
-, autoAddDriverRunpath
-
-, nixosTests
-, testers
-, ollama
-, ollama-rocm
-, ollama-cuda
-
-, config
+{
+ lib,
+ buildGoModule,
+ fetchFromGitHub,
+ fetchpatch,
+ buildEnv,
+ linkFarm,
+ overrideCC,
+ makeWrapper,
+ stdenv,
+ addDriverRunpath,
+
+ cmake,
+ gcc12,
+ clblast,
+ libdrm,
+ rocmPackages,
+ cudaPackages,
+ darwin,
+ autoAddDriverRunpath,
+
+ nixosTests,
+ testers,
+ ollama,
+ ollama-rocm,
+ ollama-cuda,
+
+ config,
# one of `[ null false "rocm" "cuda" ]`
-, acceleration ? null
+ acceleration ? null,
}:
let
@@ -59,25 +60,28 @@ let
(preparePatch "09-lora.diff" "sha256-HVDYiqNkuWO9K7aIiT73iiMj5lxMsJC1oqIG4madAPk=")
];
- preparePatch = patch: hash: fetchpatch {
- url = "file://${src}/llm/patches/${patch}";
- inherit hash;
- stripLen = 1;
- extraPrefix = "llm/llama.cpp/";
- };
-
+ preparePatch =
+ patch: hash:
+ fetchpatch {
+ url = "file://${src}/llm/patches/${patch}";
+ inherit hash;
+ stripLen = 1;
+ extraPrefix = "llm/llama.cpp/";
+ };
- accelIsValid = builtins.elem acceleration [ null false "rocm" "cuda" ];
- validateFallback = lib.warnIf (config.rocmSupport && config.cudaSupport)
- (lib.concatStrings [
- "both `nixpkgs.config.rocmSupport` and `nixpkgs.config.cudaSupport` are enabled, "
- "but they are mutually exclusive; falling back to cpu"
- ])
- (!(config.rocmSupport && config.cudaSupport));
- shouldEnable = assert accelIsValid;
- mode: fallback:
- (acceleration == mode)
- || (fallback && acceleration == null && validateFallback);
+ accelIsValid = builtins.elem acceleration [
+ null
+ false
+ "rocm"
+ "cuda"
+ ];
+ validateFallback = lib.warnIf (config.rocmSupport && config.cudaSupport) (lib.concatStrings [
+ "both `nixpkgs.config.rocmSupport` and `nixpkgs.config.cudaSupport` are enabled, "
+ "but they are mutually exclusive; falling back to cpu"
+ ]) (!(config.rocmSupport && config.cudaSupport));
+ shouldEnable =
+ assert accelIsValid;
+ mode: fallback: (acceleration == mode) || (fallback && acceleration == null && validateFallback);
rocmRequested = shouldEnable "rocm" config.rocmSupport;
cudaRequested = shouldEnable "cuda" config.cudaSupport;
@@ -85,7 +89,6 @@ let
enableRocm = rocmRequested && stdenv.isLinux;
enableCuda = cudaRequested && stdenv.isLinux;
-
rocmLibs = [
rocmPackages.clr
rocmPackages.hipblas
@@ -95,9 +98,7 @@ let
rocmPackages.rocm-device-libs
rocmPackages.rocm-smi
];
- rocmClang = linkFarm "rocm-clang" {
- llvm = rocmPackages.llvm.clang;
- };
+ rocmClang = linkFarm "rocm-clang" { llvm = rocmPackages.llvm.clang; };
rocmPath = buildEnv {
name = "rocm-path";
paths = rocmLibs ++ [ rocmClang ];
@@ -121,114 +122,125 @@ let
appleFrameworks.MetalPerformanceShaders
];
- wrapperOptions = [
- # ollama embeds llama-cpp binaries which actually run the ai models
- # these llama-cpp binaries are unaffected by the ollama binary's DT_RUNPATH
- # LD_LIBRARY_PATH is temporarily required to use the gpu
- # until these llama-cpp binaries can have their runpath patched
- "--suffix LD_LIBRARY_PATH : '${addDriverRunpath.driverLink}/lib'"
- ] ++ lib.optionals enableRocm [
- "--suffix LD_LIBRARY_PATH : '${rocmPath}/lib'"
- "--set-default HIP_PATH '${rocmPath}'"
- ];
+ wrapperOptions =
+ [
+ # ollama embeds llama-cpp binaries which actually run the ai models
+ # these llama-cpp binaries are unaffected by the ollama binary's DT_RUNPATH
+ # LD_LIBRARY_PATH is temporarily required to use the gpu
+ # until these llama-cpp binaries can have their runpath patched
+ "--suffix LD_LIBRARY_PATH : '${addDriverRunpath.driverLink}/lib'"
+ ]
+ ++ lib.optionals enableRocm [
+ "--suffix LD_LIBRARY_PATH : '${rocmPath}/lib'"
+ "--set-default HIP_PATH '${rocmPath}'"
+ ];
wrapperArgs = builtins.concatStringsSep " " wrapperOptions;
-
goBuild =
- if enableCuda then
- buildGoModule.override { stdenv = overrideCC stdenv gcc12; }
- else
- buildGoModule;
+ if enableCuda then buildGoModule.override { stdenv = overrideCC stdenv gcc12; } else buildGoModule;
inherit (lib) licenses platforms maintainers;
in
-goBuild ((lib.optionalAttrs enableRocm {
- ROCM_PATH = rocmPath;
- CLBlast_DIR = "${clblast}/lib/cmake/CLBlast";
-}) // (lib.optionalAttrs enableCuda {
- CUDA_LIB_DIR = "${cudaToolkit}/lib";
-}) // {
- inherit pname version src vendorHash;
-
- nativeBuildInputs = [
- cmake
- ] ++ lib.optionals enableRocm [
- rocmPackages.llvm.bintools
- ] ++ lib.optionals enableCuda [
- cudaPackages.cuda_nvcc
- ] ++ lib.optionals (enableRocm || enableCuda) [
- makeWrapper
- autoAddDriverRunpath
- ] ++ lib.optionals stdenv.isDarwin
- metalFrameworks;
-
- buildInputs = lib.optionals enableRocm
- (rocmLibs ++ [ libdrm ])
- ++ lib.optionals enableCuda [
- cudaPackages.cuda_cudart
- cudaPackages.cuda_cccl
- cudaPackages.libcublas
- ] ++ lib.optionals stdenv.isDarwin
- metalFrameworks;
-
- patches = [
- # disable uses of `git` in the `go generate` script
- # ollama's build script assumes the source is a git repo, but nix removes the git directory
- # this also disables necessary patches contained in `ollama/llm/patches/`
- # those patches are added to `llamacppPatches`, and reapplied here in the patch phase
- ./disable-git.patch
- # disable a check that unnecessarily exits compilation during rocm builds
- # since `rocmPath` is in `LD_LIBRARY_PATH`, ollama uses rocm correctly
- ./disable-lib-check.patch
- ] ++ llamacppPatches;
- postPatch = ''
- # replace inaccurate version number with actual release version
- substituteInPlace version/version.go --replace-fail 0.0.0 '${version}'
- '';
- preBuild = ''
- # disable uses of `git`, since nix removes the git directory
- export OLLAMA_SKIP_PATCHING=true
- # build llama.cpp libraries for ollama
- go generate ./...
- '';
- postFixup = ''
- # the app doesn't appear functional at the moment, so hide it
- mv "$out/bin/app" "$out/bin/.ollama-app"
- '' + lib.optionalString (enableRocm || enableCuda) ''
- # expose runtime libraries necessary to use the gpu
- wrapProgram "$out/bin/ollama" ${wrapperArgs}
- '';
-
- ldflags = [
- "-s"
- "-w"
- "-X=github.com/ollama/ollama/version.Version=${version}"
- "-X=github.com/ollama/ollama/server.mode=release"
- ];
+goBuild (
+ (lib.optionalAttrs enableRocm {
+ ROCM_PATH = rocmPath;
+ CLBlast_DIR = "${clblast}/lib/cmake/CLBlast";
+ })
+ // (lib.optionalAttrs enableCuda { CUDA_LIB_DIR = "${cudaToolkit}/lib"; })
+ // {
+ inherit
+ pname
+ version
+ src
+ vendorHash
+ ;
+
+ nativeBuildInputs =
+ [ cmake ]
+ ++ lib.optionals enableRocm [ rocmPackages.llvm.bintools ]
+ ++ lib.optionals enableCuda [ cudaPackages.cuda_nvcc ]
+ ++ lib.optionals (enableRocm || enableCuda) [
+ makeWrapper
+ autoAddDriverRunpath
+ ]
+ ++ lib.optionals stdenv.isDarwin metalFrameworks;
+
+ buildInputs =
+ lib.optionals enableRocm (rocmLibs ++ [ libdrm ])
+ ++ lib.optionals enableCuda [
+ cudaPackages.cuda_cudart
+ cudaPackages.cuda_cccl
+ cudaPackages.libcublas
+ ]
+ ++ lib.optionals stdenv.isDarwin metalFrameworks;
+
+ patches = [
+ # disable uses of `git` in the `go generate` script
+ # ollama's build script assumes the source is a git repo, but nix removes the git directory
+ # this also disables necessary patches contained in `ollama/llm/patches/`
+ # those patches are added to `llamacppPatches`, and reapplied here in the patch phase
+ ./disable-git.patch
+ # disable a check that unnecessarily exits compilation during rocm builds
+ # since `rocmPath` is in `LD_LIBRARY_PATH`, ollama uses rocm correctly
+ ./disable-lib-check.patch
+ ] ++ llamacppPatches;
+ postPatch = ''
+ # replace inaccurate version number with actual release version
+ substituteInPlace version/version.go --replace-fail 0.0.0 '${version}'
+ '';
+ preBuild = ''
+ # disable uses of `git`, since nix removes the git directory
+ export OLLAMA_SKIP_PATCHING=true
+ # build llama.cpp libraries for ollama
+ go generate ./...
+ '';
+ postFixup =
+ ''
+ # the app doesn't appear functional at the moment, so hide it
+ mv "$out/bin/app" "$out/bin/.ollama-app"
+ ''
+ + lib.optionalString (enableRocm || enableCuda) ''
+ # expose runtime libraries necessary to use the gpu
+ wrapProgram "$out/bin/ollama" ${wrapperArgs}
+ '';
+
+ ldflags = [
+ "-s"
+ "-w"
+ "-X=github.com/ollama/ollama/version.Version=${version}"
+ "-X=github.com/ollama/ollama/server.mode=release"
+ ];
- passthru.tests = {
- inherit ollama;
- service = nixosTests.ollama;
- version = testers.testVersion {
- inherit version;
- package = ollama;
+ passthru.tests =
+ {
+ inherit ollama;
+ service = nixosTests.ollama;
+ version = testers.testVersion {
+ inherit version;
+ package = ollama;
+ };
+ }
+ // lib.optionalAttrs stdenv.isLinux {
+ inherit ollama-rocm ollama-cuda;
+ service-cuda = nixosTests.ollama-cuda;
+ service-rocm = nixosTests.ollama-rocm;
+ };
+
+ meta = {
+ description =
+ "Get up and running with large language models locally"
+ + lib.optionalString rocmRequested ", using ROCm for AMD GPU acceleration"
+ + lib.optionalString cudaRequested ", using CUDA for NVIDIA GPU acceleration";
+ homepage = "https://github.com/ollama/ollama";
+ changelog = "https://github.com/ollama/ollama/releases/tag/v${version}";
+ license = licenses.mit;
+ platforms = if (rocmRequested || cudaRequested) then platforms.linux else platforms.unix;
+ mainProgram = "ollama";
+ maintainers = with maintainers; [
+ abysssol
+ dit7ya
+ elohmeier
+ roydubnium
+ ];
};
- } // lib.optionalAttrs stdenv.isLinux {
- inherit ollama-rocm ollama-cuda;
- service-cuda = nixosTests.ollama-cuda;
- service-rocm = nixosTests.ollama-rocm;
- };
-
- meta = {
- description = "Get up and running with large language models locally"
- + lib.optionalString rocmRequested ", using ROCm for AMD GPU acceleration"
- + lib.optionalString cudaRequested ", using CUDA for NVIDIA GPU acceleration";
- homepage = "https://github.com/ollama/ollama";
- changelog = "https://github.com/ollama/ollama/releases/tag/v${version}";
- license = licenses.mit;
- platforms =
- if (rocmRequested || cudaRequested) then platforms.linux
- else platforms.unix;
- mainProgram = "ollama";
- maintainers = with maintainers; [ abysssol dit7ya elohmeier roydubnium ];
- };
-})
+ }
+)
From d86d9340c5314a10d0ba3e7ef498d108def03538 Mon Sep 17 00:00:00 2001
From: Pol Dellaiera <pol.dellaiera@protonmail.com>
Date: Tue, 23 Jul 2024 11:08:55 +0200
Subject: [PATCH 3/4] ollama: move the assert after the top level parameters
---
pkgs/by-name/ol/ollama/package.nix | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/pkgs/by-name/ol/ollama/package.nix b/pkgs/by-name/ol/ollama/package.nix
index 7b1f2c88e92e591..c325192a583bbe7 100644
--- a/pkgs/by-name/ol/ollama/package.nix
+++ b/pkgs/by-name/ol/ollama/package.nix
@@ -30,6 +30,13 @@
acceleration ? null,
}:
+assert builtins.elem acceleration [
+ null
+ false
+ "rocm"
+ "cuda"
+];
+
let
pname = "ollama";
# don't forget to invalidate all hashes each update
@@ -69,18 +76,11 @@ let
extraPrefix = "llm/llama.cpp/";
};
- accelIsValid = builtins.elem acceleration [
- null
- false
- "rocm"
- "cuda"
- ];
validateFallback = lib.warnIf (config.rocmSupport && config.cudaSupport) (lib.concatStrings [
"both `nixpkgs.config.rocmSupport` and `nixpkgs.config.cudaSupport` are enabled, "
"but they are mutually exclusive; falling back to cpu"
]) (!(config.rocmSupport && config.cudaSupport));
shouldEnable =
- assert accelIsValid;
mode: fallback: (acceleration == mode) || (fallback && acceleration == null && validateFallback);
rocmRequested = shouldEnable "rocm" config.rocmSupport;
From 1b3b0049526c02ef652b2bc6bdca5042c6a59f31 Mon Sep 17 00:00:00 2001
From: Pol Dellaiera <pol.dellaiera@protonmail.com>
Date: Tue, 23 Jul 2024 11:51:43 +0200
Subject: [PATCH 4/4] ollama: replace variable with `with` keyword
---
pkgs/by-name/ol/ollama/package.nix | 11 +++++------
1 file changed, 5 insertions(+), 6 deletions(-)
diff --git a/pkgs/by-name/ol/ollama/package.nix b/pkgs/by-name/ol/ollama/package.nix
index c325192a583bbe7..a1f4dcc2062923f 100644
--- a/pkgs/by-name/ol/ollama/package.nix
+++ b/pkgs/by-name/ol/ollama/package.nix
@@ -114,12 +114,11 @@ let
];
};
- appleFrameworks = darwin.apple_sdk_11_0.frameworks;
- metalFrameworks = [
- appleFrameworks.Accelerate
- appleFrameworks.Metal
- appleFrameworks.MetalKit
- appleFrameworks.MetalPerformanceShaders
+ metalFrameworks = with darwin.apple_sdk_11_0.frameworks; [
+ Accelerate
+ Metal
+ MetalKit
+ MetalPerformanceShaders
];
wrapperOptions =
Reference in New Issue View Git Blame Copy Permalink