[llama-cpp] Update to later version (#148)

* [llama-cpp] update to b1695 * ci: test llama-cpp[cublas] * [llama-cpp] update tool installation * ci: install llama-cpp in Linux * [llama-cpp] update baseline
luncliff · Dec 25, 2023 · 917e78a · 917e78a
1 parent 36feec9
commit 917e78a
Show file tree

Hide file tree

Showing 8 changed files with 29 additions and 132 deletions.
diff --git a/ports/llama-cpp/fix-cmake.patch b/ports/llama-cpp/fix-cmake.patch
diff --git a/ports/llama-cpp/fix-openblas.patch b/ports/llama-cpp/fix-openblas.patch
diff --git a/ports/llama-cpp/portfile.cmake b/ports/llama-cpp/portfile.cmake
@@ -5,12 +5,9 @@ endif()
 vcpkg_from_github(
     OUT_SOURCE_PATH SOURCE_PATH
     REPO ggerganov/llama.cpp
-    REF b1273 # commit 99115f3fa654b593099c6719ad30e3f54ce231e1
-    SHA512 2b3e8fd9673647f59a4fa96621afe2f77ab10a2bee88a96b662b493beb2b66f17c854c1077f01f8ea8998d0296f92225d3033aae0adc756810f80caf45b9a456
+    REF b1695 # commit 925e5584a058afb612f9c20bc472c130f5d0f891
+    SHA512 3f50216030fe022dbfdcdb7c96765bdf2f1995a4672e7ffbca672fc5f502149d47146428e3f5cdd4799724011b29941826cf66b9178337cce05b8aa0b5292f1c
     HEAD_REF master
-    PATCHES
-        # fix-openblas.patch
-        fix-cmake.patch
 )
 
 vcpkg_find_acquire_program(PKGCONFIG)
@@ -63,7 +60,7 @@ vcpkg_cmake_configure(
         ${BLAS_OPTIONS}
         -DPKG_CONFIG_EXECUTABLE:FILEPATH="${PKGCONFIG}"
         -DBUILD_COMMIT:STRING="99115f3fa654b593099c6719ad30e3f54ce231e1"
-        -DBUILD_NUMBER:STRING="1273"
+        -DBUILD_NUMBER:STRING="1695"
     OPTIONS_RELEASE
         -DLLAMA_METAL_NDEBUG=ON
 )
@@ -73,14 +70,16 @@ vcpkg_cmake_config_fixup(CONFIG_PATH "lib/cmake/Llama" PACKAGE_NAME "Llama")
 vcpkg_copy_pdbs()
 
 vcpkg_copy_tools(TOOL_NAMES
-    baby-llama beam-search benchmark convert-llama2c-to-ggml embd-input-test embedding llama-bench
+    baby-llama beam-search benchmark convert-llama2c-to-ggml embedding llama-bench
     main perplexity quantize-stats quantize save-load-state server simple speculative train-text-from-scratch    
+    batched-bench batched export-lora finetune infill llava-cli lookahead lookup parallel tokenize
     AUTO_CLEAN
 )
 if("test" IN_LIST FEATURES)
     vcpkg_copy_tools(TOOL_NAMES
         test-grad0 test-grammar-parser test-llama-grammar test-quantize-fns test-quantize-perf
         test-sampling test-tokenizer-0-falcon test-tokenizer-0-llama test-tokenizer-1-llama
+        test-backend-ops test-rope test-tokenizer-1-bpe
         AUTO_CLEAN
     )
 endif()

diff --git a/ports/llama-cpp/vcpkg.json b/ports/llama-cpp/vcpkg.json
@@ -1,6 +1,6 @@
 {
   "name": "llama-cpp",
-  "version-string": "b1273",
+  "version-string": "b1695",
   "description": "Port of Facebook's LLaMA model in C/C++",
   "homepage": "https://github.com/ggerganov/llama.cpp",
   "supports": "windows | linux | osx",

diff --git a/test/self-hosted-cuda.json b/test/self-hosted-cuda.json
@@ -5,6 +5,13 @@
   "homepage": "https://github.com/luncliff/vcpkg-registry",
   "supports": "windows",
   "dependencies": [
+    {
+      "name": "llama-cpp",
+      "features": [
+        "cublas"
+      ],
+      "platform": "x64 & windows"
+    },
     {
       "name": "onnxruntime",
       "features": [

diff --git a/test/vcpkg.json b/test/vcpkg.json
@@ -3,7 +3,7 @@
   "version-date": "2023-12-20",
   "description": "vcpkg registry maintained by @luncliff",
   "homepage": "https://github.com/luncliff/vcpkg-registry",
-  "supports": "windows | osx | ios",
+  "supports": "windows | linux | osx | ios",
   "dependencies": [
     {
       "name": "apple-crypto",
@@ -27,7 +27,7 @@
       "features": [
         "test"
       ],
-      "platform": "windows | osx"
+      "platform": "windows | osx | linux"
     },
     {
       "name": "metal-cpp",

diff --git a/versions/baseline.json b/versions/baseline.json
@@ -60,6 +60,10 @@
       "baseline": "swift-5.5",
       "port-version": 4
     },
+    "llama-cpp": {
+      "baseline": "b1695",
+      "port-version": 0
+    },
     "metal-cpp": {
       "baseline": "2023-04-05",
       "port-version": 0

diff --git a/versions/l-/llama-cpp.json b/versions/l-/llama-cpp.json
@@ -0,0 +1,9 @@
+{
+  "versions": [
+    {
+      "git-tree": "e740088bddef250c1962485745b0bc6cb45c33d0",
+      "version-string": "b1695",
+      "port-version": 0
+    }
+  ]
+}