low-level / high-level API improvements

tangledgroup · Dec 11, 2024 · d603691 · d603691
1 parent b2d0f52
commit d603691
Show file tree

Hide file tree

Showing 4 changed files with 26 additions and 12 deletions.
diff --git a/Makefile_6.patch b/Makefile_6.patch
@@ -20,9 +20,9 @@
  	common/base64.hpp \
  	$(OBJ_ALL)
 -	$(CXX) $(CXXFLAGS) -static -fPIC -c $< -o $@ -Wno-cast-qual
-+	$(CXX) $(CXXFLAGS) -DLLAVA_LOG_OFF -c examples/llava/llava.cpp -o examples/llava/llava.o
-+	$(CXX) $(CXXFLAGS) -DLLAVA_LOG_OFF -c examples/llava/clip.cpp -o examples/llava/clip.o
-+	$(CXX) $(CXXFLAGS) -DLLAVA_LOG_OFF -c examples/llava/mllama.cpp -o examples/llava/mllama.o
++	$(CXX) $(CXXFLAGS) -c examples/llava/llava.cpp -o examples/llava/llava.o
++	$(CXX) $(CXXFLAGS) -c examples/llava/clip.cpp -o examples/llava/clip.o
++	$(CXX) $(CXXFLAGS) -c examples/llava/mllama.cpp -o examples/llava/mllama.o
 +	ar rcs $@ examples/llava/llava.o examples/llava/clip.o examples/llava/mllama.o
 
  llama-llava-cli: examples/llava/llava-cli.cpp \

diff --git a/llama/llama.py b/llama/llama.py
@@ -263,7 +263,7 @@ def sampler_free(sampler: llama_sampler_p):
 
 def clip_init_context(options: Options) -> clip_ctx_p:
     mmproj_path: str = hf_hub_download(repo_id=options.model.hf_repo, filename=options.model.mmproj_hf_file)
-    clip_context: clip_ctx_p = lib.clip_model_load(mmproj_path.encode(), 1)
+    clip_context: clip_ctx_p = lib.clip_model_load(mmproj_path.encode(), 0) # path, verbosity
     return clip_context
 
 

diff --git a/pyproject.toml b/pyproject.toml
@@ -57,7 +57,19 @@ if [[ $AUDITWHEEL_PLAT == manylinux* ]]; then
     dnf update -y
     dnf install -y epel-release
     dnf update -y
-    dnf install -y unzip p7zip p7zip-plugins upx clang-tools-extra
+    dnf group install -y "Development Tools"
+    # dnf install -y unzip p7zip p7zip-plugins upx clang-tools-extra libcurl-devel
+    dnf install -y clang-tools-extra libcurl-devel ninja-build
+
+    git clone https://github.com/google/shaderc
+    cd shaderc
+    ./utils/git-sync-deps
+    mkdir build
+    cd build
+    cmake -GNinja -DCMAKE_BUILD_TYPE=Release ../
+    ninja
+    ninja install
+    cd ../..
 
     # vulkan
     if [[ "$ARCH" == "x86_64" ]]; then
@@ -66,19 +78,21 @@ if [[ $AUDITWHEEL_PLAT == manylinux* ]]; then
         dnf install -y https://pkgs.sysadmins.ws/el8/extras/x86_64/glslc-2023.1-3.el8.x86_64.rpm https://pkgs.sysadmins.ws/el8/extras/x86_64/glslang-12.0.0-1.el8.x86_64.rpm
     fi
 else
-    apk -U add unzip 7zip upx clang-extra-tools
+    # apk -U add unzip 7zip upx clang-extra-tools
+    apk -U add clang-extra-tools curl-dev ninja-build
 
     # vulkan
     if [[ "$ARCH" == "x86_64" ]]; then
         apk -U add mesa-vulkan-ati mesa-vulkan-intel mesa-vulkan-layers mesa-vulkan-swrast vulkan-headers vulkan-loader vulkan-loader-dev vulkan-tools vulkan-validation-layers vulkan-validation-layers-dev vulkan-validation-layers-static shaderc glslang glslang-dev glslang-libs glslang-static
     fi
 fi
 """
-skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "pp37-*", "pp38-*", "pp39-*", "*i686"]
+# skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "pp37-*", "pp38-*", "pp39-*", "*i686"]
 # skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "cp313-*", "pp37-*", "pp38-*", "pp39-*", "*i686"]
 # skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "pp37-*", "pp38-*", "pp39-*", "*i686", "*manylinux*"]
 # skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "cp310-*", "cp311-*", "pp37-*", "pp38-*", "pp39-*", "pp310-*", "*i686"]
 # skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "cp310-*", "cp311-*", "pp37-*", "pp38-*", "pp39-*", "pp310-*", "*i686", "*musllinux*"]
+skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "cp311-*", "cp313-*", "pp37-*", "pp38-*", "pp39-*", "pp310-*", "*i686", "*musllinux*"]
 # skip = ["cp36-*", "cp37-*", "cp38-*", "cp39-*", "cp310-*", "cp311-*", "cp313-*", "pp37-*", "pp38-*", "pp39-*", "pp310-*", "*i686", "*musllinux*"]
 manylinux-x86_64-image = "quay.io/pypa/manylinux_2_28_x86_64:latest"
 manylinux-aarch64-image = "quay.io/pypa/manylinux_2_28_aarch64:latest"

diff --git a/scripts/build.py b/scripts/build.py
@@ -847,11 +847,11 @@ def build(*args, **kwargs):
         clean_llama_cpp()
         build_vulkan_1_x(*args, **kwargs)
 
-    # # cuda 12.6.3
-    # if env.get('GGML_CUDA', '1') != '0':
-    #     if env.get('AUDITWHEEL_POLICY') in ('manylinux2014', 'manylinux_2_28', None) and env.get('AUDITWHEEL_ARCH') in ('x86_64', None):
-    #         clean_llama_cpp()
-    #         build_linux_cuda_12_6_3(*args, **kwargs)
+    # cuda 12.6.3
+    if env.get('GGML_CUDA', '1') != '0':
+        if env.get('AUDITWHEEL_POLICY') in ('manylinux2014', 'manylinux_2_28', None) and env.get('AUDITWHEEL_ARCH') in ('x86_64', None):
+            clean_llama_cpp()
+            build_linux_cuda_12_6_3(*args, **kwargs)
 
 
 if __name__ == '__main__':