Merge remote-tracking branch 'upstream/concedo'

.editorconfig

@@ -24,6 +24,16 @@ insert_final_newline = unset
 [examples/server/public/*]
 indent_size = 2
 
+[examples/server/public/deps_*]
+trim_trailing_whitespace = unset
+indent_style = unset
+indent_size = unset
+
+[examples/server/deps_*]
+trim_trailing_whitespace = unset
+indent_style = unset
+indent_size = unset
+
 [examples/llama.swiftui/llama.swiftui.xcodeproj/*]
 indent_style = tab
 

.github/workflows/kcpp-build-release-win-cuda.yaml

@@ -25,7 +25,7 @@ jobs:
           mkdir build
           cd build
           cmake .. -DLLAMA_CUBLAS=ON -DCMAKE_SYSTEM_VERSION="10.0.19041.0"
-          cmake --build . --config Release -j $((${env:NUMBER_OF_PROCESSORS} - 1))
+          cmake --build . --config Release -j 1
 
       - name: Save artifact
         uses: actions/upload-artifact@v3

.github/workflows/kcpp-build-release-win-cuda12.yaml

@@ -25,7 +25,7 @@ jobs:
           mkdir build
           cd build
           cmake .. -DLLAMA_CUBLAS=ON -DCMAKE_SYSTEM_VERSION="10.0.19041.0"
-          cmake --build . --config Release -j $((${env:NUMBER_OF_PROCESSORS} - 1))
+          cmake --build . --config Release -j 1
 
       - name: Save artifact
         uses: actions/upload-artifact@v3

.github/workflows/kcpp-build-release-win-full-cu12.yaml

@@ -33,6 +33,15 @@ jobs:
         run: |
           echo "$(Get-Location)\w64devkit\bin" | Out-File -Append -FilePath $env:GITHUB_PATH -Encoding utf8
 
+      - name: Print System Environment Variables
+        id: printvars
+        run: |
+          echo "Number of processors: ${env:NUMBER_OF_PROCESSORS}"
+          echo "Processor Architecture: ${env:PROCESSOR_ARCHITECTURE}"
+          echo "Computer Name: ${env:COMPUTERNAME}"
+          wmic cpu get name
+          wmic os get TotalVisibleMemorySize, FreePhysicalMemory
+
       - name: Build Non-CUDA
         id: make_build
         run: |
@@ -49,7 +58,7 @@ jobs:
           mkdir build
           cd build
           cmake .. -DLLAMA_CUBLAS=ON -DCMAKE_SYSTEM_VERSION="10.0.19041.0"
-          cmake --build . --config Release -j $((${env:NUMBER_OF_PROCESSORS} - 1))
+          cmake --build . --config Release -j 1
           mv bin/Release/koboldcpp_cublas.dll ../koboldcpp_cublas.dll
           cd ..
 

.github/workflows/kcpp-build-release-win-full.yaml

@@ -33,6 +33,15 @@ jobs:
         run: |
           echo "$(Get-Location)\w64devkit\bin" | Out-File -Append -FilePath $env:GITHUB_PATH -Encoding utf8
 
+      - name: Print System Environment Variables
+        id: printvars
+        run: |
+          echo "Number of processors: ${env:NUMBER_OF_PROCESSORS}"
+          echo "Processor Architecture: ${env:PROCESSOR_ARCHITECTURE}"
+          echo "Computer Name: ${env:COMPUTERNAME}"
+          wmic cpu get name
+          wmic os get TotalVisibleMemorySize, FreePhysicalMemory
+
       - name: Build Non-CUDA
         id: make_build
         run: |
@@ -49,7 +58,7 @@ jobs:
           mkdir build
           cd build
           cmake .. -DLLAMA_CUBLAS=ON -DCMAKE_SYSTEM_VERSION="10.0.19041.0"
-          cmake --build . --config Release -j $((${env:NUMBER_OF_PROCESSORS} - 1))
+          cmake --build . --config Release -j 1
           mv bin/Release/koboldcpp_cublas.dll ../koboldcpp_cublas.dll
           cd ..
 

.github/workflows/kcpp-build-release-win-oldcpu-full.yaml

@@ -33,6 +33,15 @@ jobs:
         run: |
           echo "$(Get-Location)\w64devkit\bin" | Out-File -Append -FilePath $env:GITHUB_PATH -Encoding utf8
 
+      - name: Print System Environment Variables
+        id: printvars
+        run: |
+          echo "Number of processors: ${env:NUMBER_OF_PROCESSORS}"
+          echo "Processor Architecture: ${env:PROCESSOR_ARCHITECTURE}"
+          echo "Computer Name: ${env:COMPUTERNAME}"
+          wmic cpu get name
+          wmic os get TotalVisibleMemorySize, FreePhysicalMemory
+
       - name: Build Non-CUDA
         id: make_build
         run: |
@@ -49,7 +58,7 @@ jobs:
           mkdir build
           cd build
           cmake .. -DLLAMA_CUBLAS=ON -DLLAMA_AVX2=OFF -DCMAKE_SYSTEM_VERSION="10.0.19041.0"
-          cmake --build . --config Release -j ${env:NUMBER_OF_PROCESSORS}
+          cmake --build . --config Release -j 1
           mv bin/Release/koboldcpp_cublas.dll ../koboldcpp_cublas.dll
           cd ..
 

CMakeLists.txt

@@ -469,11 +469,14 @@ endif()
 add_library(ggml
             ggml/src/ggml.c
             ggml/include/ggml.h
+            ggml/src/ggml-cpu.c
+            ggml/include/ggml-cpu.h
             ggml/src/ggml-alloc.c
             ggml/include/ggml-alloc.h
             ggml/src/ggml-backend.cpp
             ggml/src/ggml-backend-impl.h
             ggml/include/ggml-backend.h
+            ggml/include/ggml-cpp.h
             ggml/src/ggml-quants.c
             ggml/src/ggml-quants.h
             ggml/src/llamafile/sgemm.cpp

Makefile

@@ -452,6 +452,18 @@ ggml_v4_vulkan.o: ggml/src/ggml.c ggml/include/ggml.h
 ggml_v4_vulkan_noavx2.o: ggml/src/ggml.c ggml/include/ggml.h
 	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) $(VULKAN_FLAGS) -c $< -o $@
 
+# cpu and clblast separated
+ggml-cpu.o: ggml/src/ggml-cpu.c ggml/include/ggml-cpu.h
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) -c $< -o $@
+ggml-cpu_v4_failsafe.o: ggml/src/ggml-cpu.c ggml/include/ggml-cpu.h
+	$(CC)  $(FASTCFLAGS) $(NONECFLAGS) -c $< -o $@
+ggml-cpu_v4_noavx2.o: ggml/src/ggml-cpu.c ggml/include/ggml-cpu.h
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) -c $< -o $@
+ggml-cpu_v4_clblast.o: ggml/src/ggml-cpu.c ggml/include/ggml-cpu.h
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+ggml-cpu_v4_clblast_noavx2.o: ggml/src/ggml-cpu.c ggml/include/ggml-cpu.h
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+
 #quants
 ggml-quants.o: ggml/src/ggml-quants.c ggml/include/ggml.h ggml/src/ggml-quants.h ggml/src/ggml-common.h
 	$(CC)  $(CFLAGS) $(FULLCFLAGS) -c $< -o $@
@@ -574,7 +586,7 @@ whispercpp_cublas.o: otherarch/whispercpp/whisper_adapter.cpp
 	$(CXX) $(CXXFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
 
 # idiotic "for easier compilation"
-GPTTYPE_ADAPTER = gpttype_adapter.cpp otherarch/llama_v2.cpp otherarch/llama_v3.cpp src/llama.cpp src/llama-grammar.cpp src/llama-sampling.cpp src/llama-vocab.cpp otherarch/utils.cpp otherarch/gptj_v1.cpp otherarch/gptj_v2.cpp otherarch/gptj_v3.cpp otherarch/gpt2_v1.cpp otherarch/gpt2_v2.cpp otherarch/gpt2_v3.cpp otherarch/rwkv_v2.cpp otherarch/rwkv_v3.cpp otherarch/neox_v2.cpp otherarch/neox_v3.cpp otherarch/mpt_v3.cpp ggml/include/ggml.h ggml/include/ggml-cuda.h include/llama.h otherarch/llama-util.h
+GPTTYPE_ADAPTER = gpttype_adapter.cpp otherarch/llama_v2.cpp otherarch/llama_v3.cpp src/llama.cpp src/llama-grammar.cpp src/llama-sampling.cpp src/llama-vocab.cpp otherarch/utils.cpp otherarch/gptj_v1.cpp otherarch/gptj_v2.cpp otherarch/gptj_v3.cpp otherarch/gpt2_v1.cpp otherarch/gpt2_v2.cpp otherarch/gpt2_v3.cpp otherarch/rwkv_v2.cpp otherarch/rwkv_v3.cpp otherarch/neox_v2.cpp otherarch/neox_v3.cpp otherarch/mpt_v3.cpp ggml/include/ggml.h ggml/include/ggml-cpu.h ggml/include/ggml-cuda.h include/llama.h otherarch/llama-util.h
 gpttype_adapter_failsafe.o: $(GPTTYPE_ADAPTER)
 	$(CXX) $(CXXFLAGS) $(FAILSAFE_FLAGS) -c $< -o $@
 gpttype_adapter.o: $(GPTTYPE_ADAPTER)
@@ -596,18 +608,18 @@ clean:
 	rm -vrf ggml/src/ggml-cuda/template-instances/*.o
 
 # useful tools
-main: examples/main/main.cpp common/json-schema-to-grammar.cpp common/arg.cpp build-info.h ggml.o llama.o console.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+main: examples/main/main.cpp common/json-schema-to-grammar.cpp common/arg.cpp build-info.h ggml.o ggml-cpu.o llama.o console.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
 	@echo '====  Run ./main -h for help.  ===='
-sdmain: otherarch/sdcpp/util.cpp otherarch/sdcpp/main.cpp otherarch/sdcpp/stable-diffusion.cpp otherarch/sdcpp/upscaler.cpp otherarch/sdcpp/model.cpp otherarch/sdcpp/thirdparty/zip.c build-info.h ggml.o llama.o console.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+sdmain: otherarch/sdcpp/util.cpp otherarch/sdcpp/main.cpp otherarch/sdcpp/stable-diffusion.cpp otherarch/sdcpp/upscaler.cpp otherarch/sdcpp/model.cpp otherarch/sdcpp/thirdparty/zip.c build-info.h ggml.o ggml-cpu.o llama.o console.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
-whispermain: otherarch/whispercpp/main.cpp otherarch/whispercpp/whisper.cpp build-info.h ggml.o llama.o console.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+whispermain: otherarch/whispercpp/main.cpp otherarch/whispercpp/whisper.cpp build-info.h ggml.o ggml-cpu.o llama.o console.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
-imatrix: examples/imatrix/imatrix.cpp build-info.h ggml.o llama.o console.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+imatrix: examples/imatrix/imatrix.cpp build-info.h ggml.o ggml-cpu.o llama.o console.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
-gguf: examples/gguf/gguf.cpp build-info.h ggml.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+gguf: examples/gguf/gguf.cpp build-info.h ggml.o ggml-cpu.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
-gguf-split: examples/gguf-split/gguf-split.cpp ggml.o llama.o build-info.h llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+gguf-split: examples/gguf-split/gguf-split.cpp ggml.o ggml-cpu.o llama.o build-info.h llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
 
 vulkan-shaders-gen: ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp
@@ -617,30 +629,30 @@ vulkan-shaders-gen: ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp
 	$(shell) vulkan-shaders-gen --glslc glslc --input-dir ggml/src/vulkan-shaders --target-hpp ggml/src/ggml-vulkan-shaders.hpp --target-cpp ggml/src/ggml-vulkan-shaders.cpp
 
 #generated libraries
-koboldcpp_default: ggml.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+koboldcpp_default: ggml.o ggml-cpu.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(DEFAULT_BUILD)
 
 ifdef FAILSAFE_BUILD
-koboldcpp_failsafe: ggml_v4_failsafe.o ggml_v3_failsafe.o ggml_v2_failsafe.o ggml_v1_failsafe.o expose.o gpttype_adapter_failsafe.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FAILSAFE) $(OBJS)
+koboldcpp_failsafe: ggml_v4_failsafe.o ggml-cpu_v4_failsafe.o ggml_v3_failsafe.o ggml_v2_failsafe.o ggml_v1_failsafe.o expose.o gpttype_adapter_failsafe.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FAILSAFE) $(OBJS)
 	$(FAILSAFE_BUILD)
 else
 koboldcpp_failsafe:
 	$(DONOTHING)
 endif
 
 ifdef NOAVX2_BUILD
-koboldcpp_noavx2: ggml_v4_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_failsafe.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_SIMPLE) $(OBJS)
+koboldcpp_noavx2: ggml_v4_noavx2.o ggml-cpu_v4_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_failsafe.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_SIMPLE) $(OBJS)
 	$(NOAVX2_BUILD)
 else
 koboldcpp_noavx2:
 	$(DONOTHING)
 endif
 
 ifdef CLBLAST_BUILD
-koboldcpp_clblast: ggml_v4_clblast.o ggml_v3_clblast.o ggml_v2_clblast.o ggml_v1.o expose.o gpttype_adapter_clblast.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
+koboldcpp_clblast: ggml_v4_clblast.o ggml-cpu_v4_clblast.o ggml_v3_clblast.o ggml_v2_clblast.o ggml_v1.o expose.o gpttype_adapter_clblast.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
 	$(CLBLAST_BUILD)
 ifdef NOAVX2_BUILD
-koboldcpp_clblast_noavx2: ggml_v4_clblast_noavx2.o ggml_v3_clblast_noavx2.o ggml_v2_clblast_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_clblast_noavx2.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_SIMPLE) $(OBJS)
+koboldcpp_clblast_noavx2: ggml_v4_clblast_noavx2.o ggml-cpu_v4_clblast_noavx2.o ggml_v3_clblast_noavx2.o ggml_v2_clblast_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_clblast_noavx2.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_SIMPLE) $(OBJS)
 	$(CLBLAST_BUILD)
 else
 koboldcpp_clblast_noavx2:
@@ -654,26 +666,26 @@ koboldcpp_clblast_noavx2:
 endif
 
 ifdef CUBLAS_BUILD
-koboldcpp_cublas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o gpttype_adapter_cublas.o sdcpp_cublas.o whispercpp_cublas.o llavaclip_cublas.o llava.o ggml-backend_cublas.o $(CUBLAS_OBJS) $(OBJS_FULL) $(OBJS)
+koboldcpp_cublas: ggml_v4_cublas.o ggml-cpu.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o gpttype_adapter_cublas.o sdcpp_cublas.o whispercpp_cublas.o llavaclip_cublas.o llava.o ggml-backend_cublas.o $(CUBLAS_OBJS) $(OBJS_FULL) $(OBJS)
 	$(CUBLAS_BUILD)
 else
 koboldcpp_cublas:
 	$(DONOTHING)
 endif
 
 ifdef HIPBLAS_BUILD
-koboldcpp_hipblas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o gpttype_adapter_cublas.o sdcpp_cublas.o whispercpp_cublas.o llavaclip_cublas.o llava.o ggml-backend_cublas.o $(HIP_OBJS) $(OBJS_FULL) $(OBJS)
+koboldcpp_hipblas: ggml_v4_cublas.o ggml-cpu.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o gpttype_adapter_cublas.o sdcpp_cublas.o whispercpp_cublas.o llavaclip_cublas.o llava.o ggml-backend_cublas.o $(HIP_OBJS) $(OBJS_FULL) $(OBJS)
 	$(HIPBLAS_BUILD)
 else
 koboldcpp_hipblas:
 	$(DONOTHING)
 endif
 
 ifdef VULKAN_BUILD
-koboldcpp_vulkan: ggml_v4_vulkan.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter_vulkan.o ggml-vulkan.o sdcpp_vulkan.o whispercpp_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o $(OBJS_FULL) $(OBJS)
+koboldcpp_vulkan: ggml_v4_vulkan.o ggml-cpu.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter_vulkan.o ggml-vulkan.o sdcpp_vulkan.o whispercpp_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o $(OBJS_FULL) $(OBJS)
 	$(VULKAN_BUILD)
 ifdef NOAVX2_BUILD
-koboldcpp_vulkan_noavx2: ggml_v4_vulkan_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_vulkan_noavx2.o ggml-vulkan.o sdcpp_vulkan.o whispercpp_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o $(OBJS_SIMPLE) $(OBJS)
+koboldcpp_vulkan_noavx2: ggml_v4_vulkan_noavx2.o ggml-cpu_v4_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_vulkan_noavx2.o ggml-vulkan.o sdcpp_vulkan.o whispercpp_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o $(OBJS_SIMPLE) $(OBJS)
 	$(VULKAN_BUILD)
 else
 koboldcpp_vulkan_noavx2:
@@ -687,17 +699,17 @@ koboldcpp_vulkan_noavx2:
 endif
 
 # tools
-quantize_gguf: examples/quantize/quantize.cpp ggml.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
+quantize_gguf: examples/quantize/quantize.cpp ggml.o ggml-cpu.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
-quantize_gptj: otherarch/tools/gptj_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
+quantize_gptj: otherarch/tools/gptj_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o ggml-cpu.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
-quantize_gpt2: otherarch/tools/gpt2_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
+quantize_gpt2: otherarch/tools/gpt2_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o ggml-cpu.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
-quantize_neox: otherarch/tools/neox_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
+quantize_neox: otherarch/tools/neox_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o ggml-cpu.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
-quantize_mpt: otherarch/tools/mpt_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
+quantize_mpt: otherarch/tools/mpt_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o ggml-cpu.o llama.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
-quantize_clip: examples/llava/clip.cpp examples/llava/clip.h examples/llava/quantclip.cpp ggml_v3.o ggml.o llama.o ggml-backend_default.o $(OBJS_FULL)
+quantize_clip: examples/llava/clip.cpp examples/llava/clip.h examples/llava/quantclip.cpp ggml_v3.o ggml.o ggml-cpu.o llama.o ggml-backend_default.o $(OBJS_FULL)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
 perplexity: examples/perplexity/perplexity.cpp                build-info.h ggml_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o k_quants.o ggml-alloc.o $(CUBLAS_OBJS) $(HIP_OBJS) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS) $(HIPLDFLAGS)

cmake/arm64-apple-clang.cmake

@@ -0,0 +1,16 @@
+set( CMAKE_SYSTEM_NAME Darwin )
+set( CMAKE_SYSTEM_PROCESSOR arm64 )
+
+set( target arm64-apple-darwin-macho )
+
+set( CMAKE_C_COMPILER    clang )
+set( CMAKE_CXX_COMPILER  clang++ )
+
+set( CMAKE_C_COMPILER_TARGET   ${target} )
+set( CMAKE_CXX_COMPILER_TARGET ${target} )
+
+set( arch_c_flags "-march=armv8.4-a -fvectorize -ffp-model=fast -fno-finite-math-only" )
+set( warn_c_flags "-Wno-format -Wno-unused-variable -Wno-unused-function" )
+
+set( CMAKE_C_FLAGS_INIT   "${arch_c_flags} ${warn_c_flags}" )
+set( CMAKE_CXX_FLAGS_INIT "${arch_c_flags} ${warn_c_flags}" )