Update docs/backend/SYCL.md

Co-authored-by: Meng, Hengyu <hengyu.meng@intel.com>

README.md

@@ -173,7 +173,6 @@ Unless otherwise noted these projects are open-source with permissive licensing:
 - [akx/ggify](https://github.com/akx/ggify) – download PyTorch models from HuggingFace Hub and convert them to GGML
 - [crashr/gppm](https://github.com/crashr/gppm) – launch llama.cpp instances utilizing NVIDIA Tesla P40 or P100 GPUs with reduced idle power consumption
 - [gpustack/gguf-parser](https://github.com/gpustack/gguf-parser-go/tree/main/cmd/gguf-parser) - review/check the GGUF file and estimate the memory usage
-- [Styled Lines](https://marketplace.unity.com/packages/tools/generative-ai/styled-lines-llama-cpp-model-292902) (proprietary licensed, async wrapper of inference part for game development in Unity3d with prebuild Mobile and Web platform wrappers and a model example)
 
 **Infrastructure:**
 

convert_hf_to_gguf.py

@@ -1487,7 +1487,7 @@ class StableLMModel(Model):
                 raise ValueError(f"Unprocessed norms: {norms}")
 
 
-@Model.register("LLaMAForCausalLM", "LlamaForCausalLM", "MistralForCausalLM", "MixtralForCausalLM")
+@Model.register("LlamaForCausalLM", "MistralForCausalLM", "MixtralForCausalLM")
 class LlamaModel(Model):
     model_arch = gguf.MODEL_ARCH.LLAMA
 

docs/backend/SYCL.md

@@ -636,6 +636,14 @@ use 1 SYCL GPUs: [0] with Max compute units:512
 
   It's same for other projects including llama.cpp SYCL backend.
 
+- Meet issue: `Native API failed. Native API returns: -6 (PI_ERROR_OUT_OF_HOST_MEMORY) -6 (PI_ERROR_OUT_OF_HOST_MEMORY) -999 (UNKNOWN PI error)` or `failed to allocate SYCL0 buffer`
+
+  Device Memory is not enough.
+
+  |Reason|Solution|
+  |-|-|
+  |Default Context is too big. It leads to more memory usage.|Set `-c 8192` or smaller value.|
+  |Model is big and require more memory than device's.|Choose smaller quantized model, like Q5 -> Q4;<br>Use more than one devices to load model.|
 
 ### **GitHub contribution**:
 Please add the **[SYCL]** prefix/tag in issues/PRs titles to help the SYCL-team check/address them without delay.

examples/server/server.cpp

@@ -2993,8 +2993,6 @@ int main(int argc, char ** argv) {
                 }, [&](json error_data) {
                     server_sent_event(sink, "error", error_data);
                 });
-                static const std::string ev_done = "data: [DONE]\n\n";
-                sink.write(ev_done.data(), ev_done.size());
                 sink.done();
                 return true;
             };

examples/server/tests/features/steps/steps.py

@@ -1020,8 +1020,6 @@ async def oai_chat_completions(user_prompt,
                             event_data = line.split(': ', 1)
                             assert event_data[0] == 'data', f'Bad event code received: ```{event_data}```'
                             chunk_raw = event_data[1]
-                            if chunk_raw == '[DONE]':
-                                break
 
                             chunk = json.loads(chunk_raw)
                             assert len(chunk['choices']) == 1, f"no choices provided, line ```{line}```"

examples/sycl/run-llama2.sh

@@ -11,16 +11,17 @@ source /opt/intel/oneapi/setvars.sh
 #ZES_ENABLE_SYSMAN=1, Support to get free memory of GPU by sycl::aspect::ext_intel_free_memory. Recommended to use when --split-mode = layer.
 
 INPUT_PROMPT="Building a website can be done in 10 simple steps:\nStep 1:"
-MODEL_FILE=llama-2-7b.Q4_0.gguf
+MODEL_FILE=models/llama-2-7b.Q4_0.gguf
 NGL=33
+CONEXT=8192
 
 if [ $# -gt 0 ]; then
     GGML_SYCL_DEVICE=$1
     echo "use $GGML_SYCL_DEVICE as main GPU"
     #use signle GPU only
-    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -m models/${MODEL_FILE} -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -mg $GGML_SYCL_DEVICE -sm none
+    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -m ${MODEL_FILE} -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONEXT} -mg $GGML_SYCL_DEVICE -sm none
 
 else
     #use multiple GPUs with same max compute units
-    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -m models/${MODEL_FILE} -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0
+    ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -m ${MODEL_FILE} -p "${INPUT_PROMPT}" -n 400 -e -ngl ${NGL} -s 0 -c ${CONEXT}
 fi

ggml/src/CMakeLists.txt

@@ -26,9 +26,6 @@ if (NOT MSVC)
     endif()
 endif()
 
-unset(GGML_EXTRA_LIBS_PRIVATE)
-unset(GGML_EXTRA_LIBS_PUBLIC)
-
 if (APPLE AND GGML_ACCELERATE)
     find_library(ACCELERATE_FRAMEWORK Accelerate)
     if (ACCELERATE_FRAMEWORK)
@@ -38,7 +35,7 @@ if (APPLE AND GGML_ACCELERATE)
         add_compile_definitions(ACCELERATE_NEW_LAPACK)
         add_compile_definitions(ACCELERATE_LAPACK_ILP64)
 
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE ${ACCELERATE_FRAMEWORK})
+        set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} ${ACCELERATE_FRAMEWORK})
     else()
         message(WARNING "Accelerate framework not found")
     endif()
@@ -90,7 +87,7 @@ if (GGML_METAL)
             COMMENT "Generate assembly for embedded Metal library"
         )
 
-        list(APPEND GGML_SOURCES_METAL ${METALLIB_EMBED_ASM})
+        set(GGML_SOURCES_METAL ${GGML_SOURCES_METAL} ${METALLIB_EMBED_ASM})
     else()
         if (GGML_METAL_SHADER_DEBUG)
             # custom command to do the following:
@@ -135,7 +132,7 @@ if (GGML_METAL)
             )
     endif() # GGML_METAL_EMBED_LIBRARY
 
-    list(APPEND GGML_EXTRA_LIBS_PRIVATE
+    set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS}
         ${FOUNDATION_LIBRARY}
         ${METAL_FRAMEWORK}
         ${METALKIT_FRAMEWORK}
@@ -160,11 +157,11 @@ if (GGML_OPENMP)
 
         add_compile_definitions(GGML_USE_OPENMP)
 
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE OpenMP::OpenMP_C OpenMP::OpenMP_CXX)
+        set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} OpenMP::OpenMP_C OpenMP::OpenMP_CXX)
 
         if (GGML_MUSA)
-            list(APPEND GGML_EXTRA_INCLUDES     "/usr/lib/llvm-10/include/openmp")
-            list(APPEND GGML_EXTRA_LIBS_PRIVATE "/usr/lib/llvm-10/lib/libomp.so")
+            set(GGML_EXTRA_INCLUDES ${GGML_EXTRA_INCLUDES} "/usr/lib/llvm-10/include/openmp")
+            set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} "/usr/lib/llvm-10/lib/libomp.so")
         endif()
     else()
         message(WARNING "OpenMP not found")
@@ -247,8 +244,8 @@ if (GGML_BLAS)
         set(GGML_HEADERS_BLAS ../include/ggml-blas.h)
         set(GGML_SOURCES_BLAS ggml-blas.cpp)
 
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE ${BLAS_LIBRARIES})
-        list(APPEND GGML_EXTRA_INCLUDES     ${BLAS_INCLUDE_DIRS})
+        set(GGML_EXTRA_LIBS     ${GGML_EXTRA_LIBS}     ${BLAS_LIBRARIES})
+        set(GGML_EXTRA_INCLUDES ${GGML_EXTRA_INCLUDES} ${BLAS_INCLUDE_DIRS})
     else()
         message(WARNING "BLAS not found, please refer to "
         "https://cmake.org/cmake/help/latest/module/FindBLAS.html#blas-lapack-vendors"
@@ -371,19 +368,19 @@ if (GGML_CUDA)
         if (GGML_STATIC)
             if (WIN32)
                 # As of 12.3.1 CUDA Toolkit for Windows does not offer a static cublas library
-                list(APPEND GGML_EXTRA_LIBS_PRIVATE CUDA::cudart_static CUDA::cublas CUDA::cublasLt)
+                set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas CUDA::cublasLt)
             else ()
                 if (GGML_MUSA)
-                    list(APPEND GGML_EXTRA_LIBS_PRIVATE MUSA::musart_static MUSA::mublas_static)
+                    set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} MUSA::musart_static MUSA::mublas_static)
                 else()
-                    list(APPEND GGML_EXTRA_LIBS_PRIVATE CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
+                    set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
                 endif()
             endif()
         else()
             if (GGML_MUSA)
-                list(APPEND GGML_EXTRA_LIBS_PRIVATE MUSA::musart MUSA::mublas)
+                set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} MUSA::musart MUSA::mublas)
             else()
-                list(APPEND GGML_EXTRA_LIBS_PRIVATE CUDA::cudart CUDA::cublas CUDA::cublasLt)
+                set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt)
             endif()
         endif()
 
@@ -391,9 +388,9 @@ if (GGML_CUDA)
             # No VMM requested, no need to link directly with the cuda driver lib (libcuda.so)
         else()
             if (GGML_MUSA)
-                list(APPEND GGML_EXTRA_LIBS_PRIVATE MUSA::musa_driver) # required by muDeviceGetAttribute(), muMemGetAllocationGranularity(...), ...
+                set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} MUSA::musa_driver) # required by muDeviceGetAttribute(), muMemGetAllocationGranularity(...), ...
             else()
-                list(APPEND GGML_EXTRA_LIBS_PRIVATE CUDA::cuda_driver) # required by cuDeviceGetAttribute(), cuMemGetAllocationGranularity(...), ...
+                set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} CUDA::cuda_driver) # required by cuDeviceGetAttribute(), cuMemGetAllocationGranularity(...), ...
             endif()
         endif()
     else()
@@ -498,7 +495,7 @@ if (GGML_HIPBLAS)
 
     if (CXX_IS_HIPCC)
         set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE CXX)
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE hip::device)
+        set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} hip::device)
     else()
         set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE HIP)
     endif()
@@ -507,7 +504,7 @@ if (GGML_HIPBLAS)
         message(FATAL_ERROR "Static linking not supported for HIP/ROCm")
     endif()
 
-    list(APPEND GGML_EXTRA_LIBS_PUBLIC hip::host roc::rocblas roc::hipblas)
+    set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} PUBLIC hip::host roc::rocblas roc::hipblas)
 endif()
 
 if (GGML_SYCL)
@@ -516,8 +513,7 @@ if (GGML_SYCL)
     endif()
 
     check_cxx_compiler_flag("-fsycl" SUPPORTS_SYCL)
-
-    if (DEFINED ENV{ONEAPI_ROOT})
+    if ( DEFINED ENV{ONEAPI_ROOT})
         message(STATUS "Using oneAPI Release SYCL compiler (icpx).")
     elseif(SUPPORTS_SYCL)
         message(WARNING "Using open-source SYCL compiler (clang++). Didn't detect ENV {ONEAPI_ROOT}.
@@ -555,31 +551,26 @@ if (GGML_SYCL)
 
     find_package(DNNL)
     message("-- DNNL found:" ${DNNL_FOUND})
-
     if (GGML_SYCL_TARGET STREQUAL "INTEL")
         add_compile_definitions(GGML_SYCL_DNNL=${DNNL_FOUND})
     else()
         add_compile_definitions(GGML_SYCL_DNNL=0)
     endif()
-
-    if (${DNNL_FOUND} AND GGML_SYCL_TARGET STREQUAL "INTEL")
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE DNNL::dnnl)
-    endif()
-
     if (WIN32)
         find_package(IntelSYCL REQUIRED)
         find_package(MKL REQUIRED)
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE IntelSYCL::SYCL_CXX MKL::MKL MKL::MKL_SYCL)
+        set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} IntelSYCL::SYCL_CXX MKL::MKL MKL::MKL_SYCL)
     else()
         if (GGML_SYCL_TARGET STREQUAL "INTEL")
-            set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsycl")
-            list(APPEND GGML_EXTRA_LIBS_PRIVATE OpenCL mkl_core pthread m dl mkl_sycl_blas mkl_intel_ilp64 mkl_tbb_thread)
+            set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} -fsycl OpenCL mkl_core pthread m dl mkl_sycl_blas mkl_intel_ilp64 mkl_tbb_thread)
         elseif (GGML_SYCL_TARGET STREQUAL "NVIDIA")
-            set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsycl")
-            set(CMAKE_CXX_FLAGS        "${CMAKE_CXX_FLAGS} -fsycl-targets=nvptx64-nvidia-cuda")
-            list(APPEND GGML_EXTRA_LIBS_PRIVATE pthread m dl onemkl)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl-targets=nvptx64-nvidia-cuda")
+            set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} -fsycl pthread m dl onemkl)
         endif()
     endif()
+    if (${DNNL_FOUND} AND GGML_SYCL_TARGET STREQUAL "INTEL")
+        list(APPEND GGML_EXTRA_LIBS DNNL::dnnl)
+    endif()
 endif()
 
 if (GGML_RPC)
@@ -588,7 +579,7 @@ if (GGML_RPC)
     list(APPEND GGML_CDEF_PUBLIC GGML_USE_RPC)
 
     if (WIN32)
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE ws2_32)
+        set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS} ws2_32)
     endif()
 
     set(GGML_HEADERS_RPC ../include/ggml-rpc.h)
@@ -666,8 +657,8 @@ if (GGML_VULKAN)
         set(GGML_HEADERS_VULKAN ${CMAKE_CURRENT_SOURCE_DIR}/../include/ggml-vulkan.h ${_ggml_vk_header})
         set(GGML_SOURCES_VULKAN ggml-vulkan.cpp ${_ggml_vk_source})
 
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE Vulkan::Vulkan)
-        list(APPEND GGML_EXTRA_INCLUDES     ${CMAKE_CURRENT_BINARY_DIR})
+        set(GGML_EXTRA_LIBS     ${GGML_EXTRA_LIBS} Vulkan::Vulkan)
+        set(GGML_EXTRA_INCLUDES ${GGML_EXTRA_INCLUDES} ${CMAKE_CURRENT_BINARY_DIR})
     else()
         message(WARNING "Vulkan not found")
     endif()
@@ -826,8 +817,8 @@ if (GGML_KOMPUTE)
 
         list(APPEND GGML_CDEF_PUBLIC GGML_USE_KOMPUTE)
 
-        list(APPEND GGML_EXTRA_LIBS_PRIVATE kompute)
-        list(APPEND GGML_EXTRA_INCLUDES     ${CMAKE_CURRENT_BINARY_DIR})
+        set(GGML_EXTRA_LIBS     ${GGML_EXTRA_LIBS}     kompute)
+        set(GGML_EXTRA_INCLUDES ${GGML_EXTRA_INCLUDES} ${CMAKE_CURRENT_BINARY_DIR})
     else()
         message(WARNING "Kompute not found")
     endif()
@@ -892,10 +883,9 @@ if (GGML_CANN)
             message(STATUS "CANN: CANN_INCLUDE_DIRS =  ${CANN_INCLUDE_DIRS}")
             message(STATUS "CANN: CANN_LIBRARIES =  ${CANN_LIBRARIES}")
 
-            list(APPEND GGML_EXTRA_LIBS_PRIVATE ${CANN_LIBRARIES} )
-            list(APPEND GGML_EXTRA_INCLUDES     ${CANN_INCLUDE_DIRS})
-            list(APPEND GGML_EXTRA_LIBDIRS      ${CANN_INSTALL_DIR}/lib64)
-
+            set(GGML_EXTRA_LIBS     ${GGML_EXTRA_LIBS}     ${CANN_LIBRARIES} )
+            set(GGML_EXTRA_INCLUDES ${GGML_EXTRA_INCLUDES} ${CANN_INCLUDE_DIRS})
+            set(GGML_EXTRA_LIBDIRS  ${GGML_EXTRA_LIBDIRS}  ${CANN_INSTALL_DIR}/lib64)
             list(APPEND GGML_CDEF_PUBLIC GGML_USE_CANN)
         endif()
     else()
@@ -1332,13 +1322,13 @@ if (EMSCRIPTEN)
     set_target_properties(ggml PROPERTIES COMPILE_FLAGS "-msimd128")
 endif()
 
-target_compile_definitions(ggml PUBLIC    ${GGML_CDEF_PUBLIC})
-target_include_directories(ggml PUBLIC  ../include)
+target_compile_definitions(ggml PUBLIC  ${GGML_CDEF_PUBLIC})
+target_include_directories(ggml PUBLIC ../include)
 target_include_directories(ggml PRIVATE . ${GGML_EXTRA_INCLUDES})
-target_link_directories   (ggml PRIVATE   ${GGML_EXTRA_LIBDIRS})
+target_link_directories(ggml PRIVATE ${GGML_EXTRA_LIBDIRS})
 target_compile_features   (ggml PRIVATE c_std_11) # don't bump
 
-list(APPEND GGML_EXTRA_LIBS_PRIVATE Threads::Threads)
+target_link_libraries(ggml PRIVATE Threads::Threads ${GGML_EXTRA_LIBS})
 
 find_library(MATH_LIBRARY m)
 if (MATH_LIBRARY)
@@ -1347,10 +1337,6 @@ if (MATH_LIBRARY)
     endif()
 endif()
 
-list(REMOVE_DUPLICATES GGML_EXTRA_LIBS_PRIVATE)
-list(REMOVE_DUPLICATES GGML_EXTRA_LIBS_PUBLIC)
-target_link_libraries(ggml PRIVATE ${GGML_EXTRA_LIBS_PRIVATE} PUBLIC ${GGML_EXTRA_LIBS_PUBLIC})
-
 if (BUILD_SHARED_LIBS)
     set_target_properties(ggml PROPERTIES POSITION_INDEPENDENT_CODE ON)
     target_compile_definitions(ggml PRIVATE GGML_SHARED GGML_BUILD)

ggml/src/ggml.c

@@ -3399,7 +3399,7 @@ double ggml_type_sizef(enum ggml_type type) {
 }
 
 GGML_CALL const char * ggml_type_name(enum ggml_type type) {
-    return type < GGML_TYPE_COUNT ? type_traits[type].type_name : "NONE";
+    return type_traits[type].type_name;
 }
 
 GGML_CALL bool ggml_is_quantized(enum ggml_type type) {

src/llama.cpp

@@ -15826,7 +15826,7 @@ static void llama_set_inputs(llama_context & lctx, const llama_ubatch & batch) {
 
             // clear unused states
             for (int i = 0; i < n_kv; ++i) {
-                const uint32_t  cell_id = i + kv_self.head;
+                uint32_t        cell_id = i + kv_self.head;
                 llama_kv_cell & kv_cell = lctx.kv_self.cells[cell_id];
 
                 data[i] = (float) (kv_cell.src >= 0);