llama: update vendored code to commit 40c6d79f (#7875)

llama/unicode.cpp

 /**
- * llama.cpp - commit 3f1ae2e32cde00c39b96be6d01c2997c29bae555 - do not edit this file
+ * llama.cpp - commit 40c6d79fb52f995f47507fedfeaae2ac05d9b35c - do not edit this file
  *
  * MIT License
  *
@@ -154,11 +154,11 @@ uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset) {
 static std::vector<codepoint_flags> unicode_cpt_flags_array() {
     std::vector<codepoint_flags> cpt_flags(MAX_CODEPOINTS, codepoint_flags::UNDEFINED);
 
-    assert (unicode_ranges_flags.front().first == 0);
-    assert (unicode_ranges_flags.back().first == MAX_CODEPOINTS);
+    assert (unicode_ranges_flags.begin()[0].first == 0);
+    assert (unicode_ranges_flags.begin()[unicode_ranges_flags.size()-1].first == MAX_CODEPOINTS);
     for (size_t i = 1; i < unicode_ranges_flags.size(); ++i) {
-        const auto range_ini = unicode_ranges_flags[i-1];  // codepoint_ini, flags
-        const auto range_end = unicode_ranges_flags[i];    // codepoint_end, flags
+        const auto range_ini = unicode_ranges_flags.begin()[i-1];  // codepoint_ini, flags
+        const auto range_end = unicode_ranges_flags.begin()[i];    // codepoint_end, flags
         for (uint32_t cpt = range_ini.first; cpt < range_end.first; ++cpt) {
             cpt_flags[cpt] = range_ini.second;
         }
@@ -247,7 +247,19 @@ static inline std::wstring unicode_wstring_from_utf8(const std::string & s) {
     free(wbuf);
     return ret;
 #else
+
+#if defined(__clang__)
+    // disable C++17 deprecation warning for std::codecvt_utf8
+#    pragma clang diagnostic push
+#    pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
+
     std::wstring_convert<std::codecvt_utf8<wchar_t>> conv;
+
+#if defined(__clang__)
+#    pragma clang diagnostic pop
+#endif
+
     return conv.from_bytes(s);
 #endif
 }
@@ -644,7 +656,7 @@ std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & c
     std::vector<uint32_t> result(cpts.size());
     for (size_t i = 0; i < cpts.size(); ++i) {
         const uint32_t cpt = cpts[i];
-        auto it = std::upper_bound(unicode_ranges_nfd.cbegin(), unicode_ranges_nfd.cend(), cpt, comp) - 1;
+        auto it = std::upper_bound(unicode_ranges_nfd.begin(), unicode_ranges_nfd.end(), cpt, comp) - 1;
         result[i] = (it->first <= cpt && cpt <= it->last) ? it->nfd : cpt;
     }
     return result;
@@ -686,8 +698,15 @@ uint8_t unicode_utf8_to_byte(const std::string & utf8) {
 }
 
 uint32_t unicode_tolower(uint32_t cp) {
-    auto it = unicode_map_lowercase.find(cp);
-    return it == unicode_map_lowercase.end() ? cp : it->second;
+    // binary search
+    auto it = std::lower_bound(unicode_map_lowercase.begin(), unicode_map_lowercase.end(), cp,
+        [](const std::pair<uint32_t, uint32_t> & pair, uint32_t value) {
+            return pair.first < value;
+        });
+    if (it != unicode_map_lowercase.end() && it->first == cp) {
+        return it->second;
+    }
+    return cp;  // Return the original code point if no lowercase mapping is found
 }
 
 std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) {

llama/unicode.h

 /**
- * llama.cpp - commit 3f1ae2e32cde00c39b96be6d01c2997c29bae555 - do not edit this file
+ * llama.cpp - commit 40c6d79fb52f995f47507fedfeaae2ac05d9b35c - do not edit this file
  *
  * MIT License
  *

llama/vendoring

-LLAMACPP_BASE_COMMIT=3f1ae2e32cde00c39b96be6d01c2997c29bae555
\ No newline at end of file
+LLAMACPP_BASE_COMMIT=40c6d79fb52f995f47507fedfeaae2ac05d9b35c

llm/server.go

@@ -699,7 +699,6 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
 		"top_k":             req.Options.TopK,
 		"top_p":             req.Options.TopP,
 		"min_p":             req.Options.MinP,
-		"tfs_z":             req.Options.TFSZ,
 		"typical_p":         req.Options.TypicalP,
 		"repeat_last_n":     req.Options.RepeatLastN,
 		"repeat_penalty":    req.Options.RepeatPenalty,

make/Makefile.rocm

@@ -51,7 +51,7 @@ GPU_DIST_LIB_DEPS= $(sort $(addprefix $(ROCM_DIST_DEPS_DIR)/,$(notdir $(GPU_LIBS
 ROCBLAS_DIST_DEP_MANIFEST = $(ROCM_DIST_DEPS_DIR)/rocblas/library/TensileManifest.txt
 
 ifeq ($(OS),linux)
-	GPU_COMPILER_FPIC := -fPIC -Wno-unused-function -std=gnu++11
+	GPU_COMPILER_FPIC := -fPIC -Wno-unused-function -std=gnu++17
 else ifeq ($(OS),windows)
 	GPU_COMPILER_FPIC := -Xclang --dependent-lib=msvcrt
 endif
@@ -69,11 +69,13 @@ GPU_COMPILER_CUFLAGS = \
 	-O3 \
 	-DGGML_USE_CUDA \
 	-DGGML_BUILD=1 \
+	-DGGML_BACKEND_BUILD=1 \
 	-DGGML_SHARED=1 \
+	-DGGML_BACKEND_SHARED=1 \
 	-DGGML_CUDA_DMMV_X=32 \
 	-DGGML_CUDA_MMV_Y=1 \
 	-DGGML_SCHED_MAX_COPIES=4 \
-	-DGGML_USE_HIPBLAS \
+	-DGGML_USE_HIP \
 	-DGGML_USE_LLAMAFILE \
 	-DHIP_FAST_MATH \
 	-D__HIP_PLATFORM_AMD__=1 \

make/Makefile.sync

@@ -86,13 +86,14 @@ LLAMACPP_FILES=\
 	src/llama-sampling.cpp \
 	src/llama-sampling.h \
 	include/llama.h \
-	ggml/src/llamafile/sgemm.cpp \
-	ggml/src/llamafile/sgemm.h
+	ggml/include/ggml-cpu.h \
+	ggml/src/ggml-cpu/llamafile/sgemm.cpp \
+	ggml/src/ggml-cpu/llamafile/sgemm.h
 $(foreach name,$(LLAMACPP_FILES),$(eval $(call vendor_file,$(name),$(DEST_DIR))))
 
 # llama.cpp files -> llama/llamafile
 LLAMAFILE_FILES= \
-	ggml/src/llamafile/sgemm.h
+	ggml/src/ggml-cpu/llamafile/sgemm.h
 $(foreach name,$(LLAMAFILE_FILES),$(eval $(call vendor_file,$(name),$(DEST_DIR)llamafile/)))
 
 # ggml files -> llama/
@@ -101,26 +102,53 @@ GGML_FILES= \
 	ggml/include/ggml.h \
 	ggml/src/ggml-quants.c \
 	ggml/src/ggml-quants.h \
-	ggml/src/ggml-metal.metal \
+	ggml/src/ggml-metal/ggml-metal.metal \
 	ggml/include/ggml-metal.h \
 	ggml/src/ggml-impl.h \
+	ggml/src/ggml-threading.h \
 	ggml/include/ggml-cuda.h \
-	ggml/src/ggml-cuda.cu \
+	ggml/src/ggml-backend-reg.cpp \
+	ggml/src/ggml-metal/ggml-metal-impl.h \
 	ggml/src/ggml-common.h \
 	ggml/include/ggml-backend.h \
-	ggml/src/ggml-backend.c \
+	ggml/src/ggml-backend.cpp \
 	ggml/src/ggml-backend-impl.h \
 	ggml/include/ggml-alloc.h \
 	ggml/src/ggml-alloc.c \
 	ggml/src/ggml-aarch64.h \
 	ggml/src/ggml-aarch64.c \
-	ggml/src/ggml-cpu-impl.h \
 	ggml/include/ggml-blas.h \
-	ggml/src/ggml-blas.cpp
+	ggml/include/ggml-cpp.h \
+	ggml/src/ggml-threading.cpp \
+	ggml/src/ggml-blas/ggml-blas.cpp \
+	ggml/src/ggml-cpu/ggml-cpu.c \
+	ggml/src/ggml-cpu/ggml-cpu-aarch64.c \
+	ggml/src/ggml-cpu/ggml-cpu.cpp \
+	ggml/src/ggml-cpu/ggml-cpu-aarch64.h \
+	ggml/src/ggml-cpu/ggml-cpu-quants.h \
+	ggml/src/ggml-cpu/ggml-cpu-quants.c \
+	ggml/src/ggml-cpu/ggml-cpu-impl.h \
+	ggml/src/ggml-cpu/amx/amx.h \
+	ggml/src/ggml-cpu/amx/amx.cpp \
+	ggml/src/ggml-cpu/amx/mmq.cpp \
+	ggml/src/ggml-cpu/amx/mmq.h
 $(foreach name,$(GGML_FILES),$(eval $(call vendor_file,$(name),$(DEST_DIR))))
 
+$(DEST_DIR)ggml-metal-embed.metal: $(DEST_DIR)ggml-common.h $(DEST_DIR)ggml-metal-impl.h
+	@sed -e '/__embed_ggml-common.h__/r $(DEST_DIR)/ggml-common.h' \
+	     -e '/__embed_ggml-common.h__/d' \
+	     < $(DEST_DIR)/ggml-metal.metal \
+	     > $(DEST_DIR)/ggml-metal-embed.metal.tmp
+	@sed -e '/#include "ggml-metal-impl.h"/r $(DEST_DIR)/ggml-metal-impl.h' \
+	     -e '/#include "ggml-metal-impl.h"/d' \
+	     < $(DEST_DIR)/ggml-metal-embed.metal.tmp \
+	     > $(DEST_DIR)/ggml-metal-embed.metal
+	@rm $(DEST_DIR)/ggml-metal-embed.metal.tmp
+
+VENDORED_FILES += $(DEST_DIR)ggml-metal-embed.metal
+
 # TODO generalize renaming pattern if we have more of these
-$(DEST_DIR)ggml-metal_darwin_arm64.m : $(LLAMACPP_REPO)ggml/src/ggml-metal.m
+$(DEST_DIR)ggml-metal_darwin_arm64.m : $(LLAMACPP_REPO)ggml/src/ggml-metal/ggml-metal.m
 	@echo "vendoring $(subst $(LLAMACPP_REPO),,$<)"; \
 		mkdir -p $(dir $@) && \
 		echo "/**" > $@ && \

make/cuda.make

@@ -41,7 +41,9 @@ GPU_COMPILER_CUFLAGS = \
 	-DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 \
 	-DGGML_USE_CUDA=1 \
 	-DGGML_SHARED=1 \
+	-DGGML_BACKEND_SHARED=1 \
 	-DGGML_BUILD=1 \
+	-DGGML_BACKEND_BUILD=1 \
 	-DGGML_USE_LLAMAFILE \
 	-DK_QUANTS_PER_ITERATION=2 \
 	-DNDEBUG \

make/gpu.make

@@ -15,10 +15,9 @@ DIST_GPU_RUNNER_DEPS_DIR = $(DIST_LIB_DIR)
 GPU_RUNNER_LIBS = $(wildcard $(addsuffix .$(SHARED_EXT).*,$(addprefix $(GPU_LIB_DIR)/$(SHARED_PREFIX),$(GPU_RUNNER_LIBS_SHORT))))
 
 GPU_RUNNER_SRCS := \
-	llama/ggml-cuda.cu \
 	$(filter-out $(wildcard llama/ggml-cuda/fattn*.cu),$(wildcard llama/ggml-cuda/*.cu)) \
 	$(wildcard llama/ggml-cuda/template-instances/mmq*.cu) \
-	llama/ggml.c llama/ggml-backend.c llama/ggml-alloc.c llama/ggml-quants.c llama/sgemm.cpp llama/ggml-aarch64.c
+	llama/ggml.c llama/ggml-backend.cpp llama/ggml-alloc.c llama/ggml-quants.c llama/sgemm.cpp llama/ggml-aarch64.c llama/ggml-threading.cpp
 GPU_RUNNER_HDRS := \
 	$(wildcard llama/ggml-cuda/*.cuh)