From 7aa588cc962d5c2fb4cbcedd548f444621292cce Mon Sep 17 00:00:00 2001
From: Joseph Stahl <1269177+josephst@users.noreply.github.com>
Date: Tue, 26 Mar 2024 13:43:26 -0400
Subject: llama-cpp: rename cuBLAS to CUDA

Matches change from upstream https://github.com/ggerganov/llama.cpp/commit/280345968dabc00d212d43e31145f5c9961a7604
---
 pkgs/by-name/ll/llama-cpp/package.nix | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pkgs/by-name/ll/llama-cpp/package.nix b/pkgs/by-name/ll/llama-cpp/package.nix
index ca2f4d5149d63..928fbf6dfe27c 100644
--- a/pkgs/by-name/ll/llama-cpp/package.nix
+++ b/pkgs/by-name/ll/llama-cpp/package.nix
@@ -107,7 +107,7 @@ effectiveStdenv.mkDerivation (finalAttrs: {
     (cmakeBool "BUILD_SHARED_LIBS" true)
     (cmakeBool "LLAMA_BLAS" blasSupport)
     (cmakeBool "LLAMA_CLBLAST" openclSupport)
-    (cmakeBool "LLAMA_CUBLAS" cudaSupport)
+    (cmakeBool "LLAMA_CUDA" cudaSupport)
     (cmakeBool "LLAMA_HIPBLAS" rocmSupport)
     (cmakeBool "LLAMA_METAL" metalSupport)
     (cmakeBool "LLAMA_MPI" mpiSupport)
-- 
cgit 1.4.1


From e1ef3aaaccb554da9c581d684e96071b2f8cea95 Mon Sep 17 00:00:00 2001
From: Joseph Stahl <1269177+josephst@users.noreply.github.com>
Date: Tue, 26 Mar 2024 13:48:11 -0400
Subject: llama-cpp: embed (don't pre-compile) metal shaders

port of https://github.com/ggerganov/llama.cpp/pull/6118, although compiling shaders with XCode disabled as it requires disabling sandbox (and only works on MacOS anyways)
---
 pkgs/by-name/ll/llama-cpp/package.nix | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/pkgs/by-name/ll/llama-cpp/package.nix b/pkgs/by-name/ll/llama-cpp/package.nix
index 928fbf6dfe27c..ecd3167b9d930 100644
--- a/pkgs/by-name/ll/llama-cpp/package.nix
+++ b/pkgs/by-name/ll/llama-cpp/package.nix
@@ -131,7 +131,10 @@ effectiveStdenv.mkDerivation (finalAttrs: {
         # Should likely use `rocmPackages.clr.gpuTargets`.
         "-DAMDGPU_TARGETS=gfx803;gfx900;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102"
       ]
-      ++ optionals metalSupport [ (cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1") ];
+      ++ optionals metalSupport [
+        (cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1")
+        (cmakeBool "LLAMA_METAL_EMBED_LIBRARY" true)
+      ];
 
   # upstream plans on adding targets at the cmakelevel, remove those
   # additional steps after that
-- 
cgit 1.4.1


From a06a03ed7c92c5482f6d260c596a4b593f201318 Mon Sep 17 00:00:00 2001
From: Joseph Stahl <1269177+josephst@users.noreply.github.com>
Date: Thu, 28 Mar 2024 20:56:55 -0400
Subject: llama-cpp: update from b2481 to b2568

---
 pkgs/by-name/ll/llama-cpp/package.nix | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/pkgs/by-name/ll/llama-cpp/package.nix b/pkgs/by-name/ll/llama-cpp/package.nix
index ecd3167b9d930..74be7dc68541c 100644
--- a/pkgs/by-name/ll/llama-cpp/package.nix
+++ b/pkgs/by-name/ll/llama-cpp/package.nix
@@ -71,13 +71,13 @@ let
 in
 effectiveStdenv.mkDerivation (finalAttrs: {
   pname = "llama-cpp";
-  version = "2481";
+  version = "2568";
 
   src = fetchFromGitHub {
     owner = "ggerganov";
     repo = "llama.cpp";
     rev = "refs/tags/b${finalAttrs.version}";
-    hash = "sha256-40GSZZEnjM9L9KVVKdSKtBoSRy996l98ORM4NeltsSM=";
+    hash = "sha256-yBlLChtzfAi2TAGUO1zdnpHCvi5YDCzjdflQgTWh98Y=";
   };
 
   postPatch = ''
-- 
cgit 1.4.1