diff --git a/Makefile b/Makefile
index 895c62f84def0..1fcc68ffb040d 100644
--- a/Makefile
+++ b/Makefile
@@ -1,8 +1,8 @@
 # Define the default target now so that it is always the first target
 BUILD_TARGETS = \
 	main quantize quantize-stats perplexity imatrix embedding vdot q8dot train-text-from-scratch convert-llama2c-to-ggml \
-	simple batched batched-bench save-load-state server gguf gguf-split eval-callback llama-bench libllava.a llava-cli baby-llama \
-	retrieval speculative infill tokenize benchmark-matmult parallel finetune export-lora lookahead lookup passkey gritlm tests/test-c.o
+	simple batched batched-bench save-load-state server gguf gguf-split eval-callback llama-bench libllava.a llava-cli baby-llama beam-search  \
+	retrieval speculative infill tokenize benchmark-matmult parallel finetune export-lora lookahead lookup passkey gritlm tests/test-c.o llamacheck
 
 # Binaries only useful for tests
 TEST_TARGETS = \
@@ -24,7 +24,7 @@ TEST_TARGETS = \
 	tests/test-tokenizer-1-bpe \
 	tests/test-tokenizer-1-spm
 
-# Code coverage output files
+# Code coverage output files;
 COV_TARGETS = *.gcno tests/*.gcno *.gcda tests/*.gcda *.gcov tests/*.gcov lcov-report gcovr-report
 
 ifndef UNAME_S
@@ -57,8 +57,6 @@ ifeq ($(UNAME_S),Darwin)
 		LLAMA_METAL := 1
 	endif
 
-	LLAMA_NO_OPENMP := 1
-
 	ifneq ($(UNAME_P),arm)
 		SYSCTL_M := $(shell sysctl -n hw.optional.arm64 2>/dev/null)
 		ifeq ($(SYSCTL_M),1)
@@ -69,10 +67,6 @@ ifeq ($(UNAME_S),Darwin)
 	endif
 endif
 
-ifdef LLAMA_RPC
-	BUILD_TARGETS += rpc-server
-endif
-
 default: $(BUILD_TARGETS)
 
 test: $(TEST_TARGETS)
@@ -141,16 +135,12 @@ MK_NVCCFLAGS = -std=c++11
 ifdef LLAMA_FAST
 MK_CFLAGS     += -Ofast
 HOST_CXXFLAGS += -Ofast
-ifndef LLAMA_DEBUG
 MK_NVCCFLAGS  += -O3
-endif # LLAMA_DEBUG
 else
 MK_CFLAGS     += -O3
 MK_CXXFLAGS   += -O3
-ifndef LLAMA_DEBUG
 MK_NVCCFLAGS  += -O3
-endif # LLAMA_DEBUG
-endif # LLAMA_FAST
+endif
 
 ifndef LLAMA_NO_CCACHE
 CCACHE := $(shell which ccache)
@@ -211,10 +201,9 @@ ifdef LLAMA_SCHED_MAX_COPIES
 endif
 
 ifdef LLAMA_DEBUG
-	MK_CFLAGS    += -O0 -g
-	MK_CXXFLAGS  += -O0 -g
-	MK_LDFLAGS   += -g
-	MK_NVCCFLAGS += -O0 -g
+	MK_CFLAGS   += -O0 -g
+	MK_CXXFLAGS += -O0 -g
+	MK_LDFLAGS  += -g
 
 	ifeq ($(UNAME_S),Linux)
 		MK_CPPFLAGS += -D_GLIBCXX_ASSERTIONS
@@ -411,12 +400,6 @@ ifndef LLAMA_NO_ACCELERATE
 	endif
 endif # LLAMA_NO_ACCELERATE
 
-ifndef LLAMA_NO_OPENMP
-	MK_CPPFLAGS += -DGGML_USE_OPENMP
-	MK_CFLAGS   += -fopenmp
-	MK_CXXFLAGS += -fopenmp
-endif # LLAMA_NO_OPENMP
-
 ifdef LLAMA_OPENBLAS
 	MK_CPPFLAGS += -DGGML_USE_OPENBLAS $(shell pkg-config --cflags-only-I openblas)
 	MK_CFLAGS   += $(shell pkg-config --cflags-only-other openblas)
@@ -433,26 +416,11 @@ ifdef LLAMA_BLIS
 	MK_LDFLAGS  += -lblis -L/usr/local/lib
 endif # LLAMA_BLIS
 
-ifdef LLAMA_RPC
-	MK_CPPFLAGS   += -DGGML_USE_RPC
-	OBJS          += ggml-rpc.o
-endif # LLAMA_RPC
-
 ifdef LLAMA_CUBLAS
 # LLAMA_CUBLAS is deprecated and will be removed in the future
 	LLAMA_CUDA := 1
 endif
 
-OBJS_CUDA_TEMP_INST      = $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-wmma*.cu))
-OBJS_CUDA_TEMP_INST     += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/mmq*.cu))
-ifdef LLAMA_CUDA_FA_ALL_QUANTS
-	OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*.cu))
-else
-	OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu))
-	OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu))
-	OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*f16-f16.cu))
-endif # LLAMA_CUDA_FA_ALL_QUANTS
-
 ifdef LLAMA_CUDA
 	ifneq ('', '$(wildcard /opt/cuda)')
 		CUDA_PATH ?= /opt/cuda
@@ -463,7 +431,6 @@ ifdef LLAMA_CUDA
 	MK_LDFLAGS   += -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L$(CUDA_PATH)/lib64 -L/usr/lib64 -L$(CUDA_PATH)/targets/$(UNAME_M)-linux/lib -L/usr/lib/wsl/lib
 	OBJS         += ggml-cuda.o
 	OBJS         += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/*.cu))
-	OBJS         += $(OBJS_CUDA_TEMP_INST)
 	MK_NVCCFLAGS += -use_fast_math
 ifdef LLAMA_FATAL_WARNINGS
 	MK_NVCCFLAGS += -Werror all-warnings
@@ -526,10 +493,7 @@ ifdef LLAMA_CUDA_NO_PEER_COPY
 endif # LLAMA_CUDA_NO_PEER_COPY
 ifdef LLAMA_CUDA_CCBIN
 	MK_NVCCFLAGS += -ccbin $(LLAMA_CUDA_CCBIN)
-endif # LLAMA_CUDA_CCBIN
-ifdef LLAMA_CUDA_FA_ALL_QUANTS
-	MK_NVCCFLAGS += -DGGML_CUDA_FA_ALL_QUANTS
-endif # LLAMA_CUDA_FA_ALL_QUANTS
+endif
 
 ifdef JETSON_EOL_MODULE_DETECT
 define NVCC_COMPILE
@@ -541,13 +505,30 @@ define NVCC_COMPILE
 endef # NVCC_COMPILE
 endif # JETSON_EOL_MODULE_DETECT
 
-ggml-cuda/%.o: ggml-cuda/%.cu ggml.h ggml-common.h ggml-cuda/common.cuh
+ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/common.cuh
 	$(NVCC_COMPILE)
 
 ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
 	$(NVCC_COMPILE)
 endif # LLAMA_CUDA
 
+ifdef LLAMA_CLBLAST
+	MK_CPPFLAGS += -DGGML_USE_CLBLAST $(shell pkg-config --cflags-only-I clblast OpenCL)
+	MK_CFLAGS   += $(shell pkg-config --cflags-only-other clblast OpenCL)
+	MK_CXXFLAGS += $(shell pkg-config --cflags-only-other clblast OpenCL)
+
+	# Mac provides OpenCL as a framework
+	ifeq ($(UNAME_S),Darwin)
+		MK_LDFLAGS += -lclblast -framework OpenCL
+	else
+		MK_LDFLAGS += $(shell pkg-config --libs clblast OpenCL)
+	endif
+	OBJS    += ggml-opencl.o
+
+ggml-opencl.o: ggml-opencl.cpp ggml-opencl.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+endif # LLAMA_CLBLAST
+
 ifdef LLAMA_VULKAN
 	MK_CPPFLAGS  += -DGGML_USE_VULKAN
 	MK_LDFLAGS += -lvulkan
@@ -590,7 +571,6 @@ ifdef LLAMA_HIP_UMA
 	MK_CPPFLAGS += -DGGML_HIP_UMA
 endif # LLAMA_HIP_UMA
 	MK_LDFLAGS  += -L$(ROCM_PATH)/lib -Wl,-rpath=$(ROCM_PATH)/lib
-	MK_LDFLAGS  += -L$(ROCM_PATH)/lib64 -Wl,-rpath=$(ROCM_PATH)/lib64
 	MK_LDFLAGS	+= -lhipblas -lamdhip64 -lrocblas
 	HIPFLAGS    += $(addprefix --offload-arch=,$(AMDGPU_TARGETS))
 	HIPFLAGS    += -DGGML_CUDA_DMMV_X=$(LLAMA_CUDA_DMMV_X)
@@ -604,12 +584,11 @@ ifdef LLAMA_CUDA_NO_PEER_COPY
 endif # LLAMA_CUDA_NO_PEER_COPY
 	OBJS        += ggml-cuda.o
 	OBJS        += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/*.cu))
-	OBJS        += $(OBJS_CUDA_TEMP_INST)
 
 ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
 	$(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
 
-ggml-cuda/%.o: ggml-cuda/%.cu ggml.h ggml-common.h ggml-cuda/common.cuh
+ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/common.cuh
 	$(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
 
 endif # LLAMA_HIPBLAS
@@ -647,26 +626,11 @@ ggml-metal-embed.o: ggml-metal.metal ggml-common.h
 endif
 endif # LLAMA_METAL
 
-OBJS += ggml-alloc.o ggml-backend.o ggml-quants.o unicode.o unicode-data.o
-COMMON_H_DEPS = common/common.h common/sampling.h common/log.h llama.h
-COMMON_DEPS   = common.o sampling.o grammar-parser.o build-info.o json-schema-to-grammar.o
-
 ifndef LLAMA_NO_LLAMAFILE
 sgemm.o: sgemm.cpp sgemm.h ggml.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 endif
 
-ifdef LLAMA_RPC
-ggml-rpc.o: ggml-rpc.cpp ggml-rpc.h
-	$(CXX) $(CXXFLAGS) -c $< -o $@
-
-rpc-server.o: examples/rpc/rpc-server.cpp ggml-rpc.h
-	$(CXX) $(CXXFLAGS) -c $< -o $@
-
-rpc-server: rpc-server.o ggml.o llama.o $(COMMON_DEPS) $(OBJS)
-	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
-endif # LLAMA_RPC
-
 GF_CC := $(CC)
 include scripts/get-flags.mk
 
@@ -746,9 +710,14 @@ unicode.o: unicode.cpp unicode.h
 unicode-data.o: unicode-data.cpp unicode-data.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
+OBJS += ggml-alloc.o ggml-backend.o ggml-quants.o unicode.o unicode-data.o
+
 llama.o: llama.cpp unicode.h ggml.h ggml-alloc.h ggml-backend.h ggml-cuda.h ggml-metal.h llama.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
+COMMON_H_DEPS = common/common.h common/sampling.h common/log.h llama.h
+COMMON_DEPS   = common.o sampling.o grammar-parser.o build-info.o json-schema-to-grammar.o
+
 common.o: common/common.cpp $(COMMON_H_DEPS)
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
@@ -779,7 +748,6 @@ libllama.a: llama.o ggml.o $(OBJS) $(COMMON_DEPS)
 clean:
 	rm -vrf *.o tests/*.o *.so *.a *.dll benchmark-matmult lookup-create lookup-merge lookup-stats common/build-info.cpp *.dot $(COV_TARGETS) $(BUILD_TARGETS) $(TEST_TARGETS)
 	rm -vrf ggml-cuda/*.o
-	rm -vrf ggml-cuda/template-instances/*.o
 	find examples pocs -type f -name "*.o" -delete
 
 #
@@ -848,7 +816,7 @@ save-load-state: examples/save-load-state/save-load-state.cpp ggml.o llama.o $(C
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
-server: examples/server/server.cpp examples/server/utils.hpp examples/server/httplib.h common/json.hpp examples/server/colorthemes.css.hpp examples/server/style.css.hpp examples/server/theme-beeninorder.css.hpp examples/server/theme-ketivah.css.hpp examples/server/theme-mangotango.css.hpp examples/server/theme-playground.css.hpp examples/server/theme-polarnight.css.hpp examples/server/theme-snowstorm.css.hpp examples/server/index.html.hpp examples/server/index-new.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp examples/server/system-prompts.js.hpp examples/server/prompt-formats.js.hpp examples/server/json-schema-to-grammar.mjs.hpp common/stb_image.h ggml.o llama.o $(COMMON_DEPS) grammar-parser.o $(OBJS)
+server: examples/server/server.cpp examples/server/utils.hpp examples/server/httplib.h common/json.hpp examples/server/index.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp examples/server/json-schema-to-grammar.mjs.hpp common/stb_image.h ggml.o llama.o $(COMMON_DEPS) grammar-parser.o $(OBJS)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h %.hpp $<,$^) -Iexamples/server $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS) $(LWINSOCK2)
 
@@ -898,6 +866,10 @@ baby-llama: examples/baby-llama/baby-llama.cpp ggml.o llama.o $(COMMON_DEPS) tra
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
+beam-search: examples/beam-search/beam-search.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS)
+	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
+
 finetune: examples/finetune/finetune.cpp ggml.o llama.o $(COMMON_DEPS) train.o $(OBJS)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
@@ -940,6 +912,10 @@ gbnf-validator: examples/gbnf-validator/gbnf-validator.cpp ggml.o llama.o $(COMM
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
+llamacheck: examples/llamacheck/llamacheck.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS)
+	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
+
 ifeq ($(UNAME_S),Darwin)
 swift: examples/batched.swift
 	(cd examples/batched.swift; make build)
diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt
index 53002f8e1ce76..ad7fd393eb433 100644
--- a/examples/CMakeLists.txt
+++ b/examples/CMakeLists.txt
@@ -12,6 +12,7 @@ include_directories(${CMAKE_CURRENT_SOURCE_DIR})
 
 if (EMSCRIPTEN)
 else()
+    add_subdirectory(llamacheck)
     add_subdirectory(baby-llama)
     add_subdirectory(batched)
     add_subdirectory(batched-bench)
@@ -44,6 +45,7 @@ else()
     add_subdirectory(gguf)
     add_subdirectory(train-text-from-scratch)
     add_subdirectory(imatrix)
+
     if (LLAMA_BUILD_SERVER)
         add_subdirectory(server)
     endif()
diff --git a/examples/llamacheck/CMakeLists.txt b/examples/llamacheck/CMakeLists.txt
new file mode 100644
index 0000000000000..e07bcdfcbcd19
--- /dev/null
+++ b/examples/llamacheck/CMakeLists.txt
@@ -0,0 +1,5 @@
+set(TARGET llamacheck)
+add_executable(${TARGET} llamacheck.cpp)
+install(TARGETS ${TARGET} RUNTIME)
+target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PRIVATE cxx_std_11)
diff --git a/examples/llamacheck/README.md b/examples/llamacheck/README.md
new file mode 100644
index 0000000000000..b5b6ef426beef
--- /dev/null
+++ b/examples/llamacheck/README.md
@@ -0,0 +1,15 @@
+# Llamacheck: Basic Spellcheck and Grammarcheck using Llama
+
+
+The attached file provides a basic implementation of LLama to 
+be used for Spell and Grammar checking.
+We use it as follows:
+```console
+make llamacheck
+./llamacheck <./models/llamacheck.gguf> 
+```
+The weights are quantized. On my machine, it runs with as speed of 7.21 t/s
+
+
+Weights are available at:
+https://huggingface.co/azferruolo/llamacheck
\ No newline at end of file
diff --git a/examples/llamacheck/llamacheck.cpp b/examples/llamacheck/llamacheck.cpp
new file mode 100644
index 0000000000000..5d1e706d99a17
--- /dev/null
+++ b/examples/llamacheck/llamacheck.cpp
@@ -0,0 +1,187 @@
+#include "common.h"
+#include <cstdio>
+#include <string>
+#include <vector>
+
+int main(int argc, char ** argv) {
+    gpt_params params;
+
+    if (argc == 1 || argv[1][0] == '-') {
+        printf("usage: %s MODEL_PATH [PROMPT]\n" , argv[0]);
+        return 1 ;
+    }
+
+    if (argc >= 2) {
+        params.model = argv[1];
+    }
+
+    params.prompt = "";
+
+    // total length of the sequence including the prompt
+    const int n_len = 150;
+
+    // init LLM
+
+    llama_backend_init();
+    llama_numa_init(params.numa);
+
+    // initialize the model
+
+    llama_model_params model_params = llama_model_default_params();
+
+    // model_params.n_gpu_layers = 99; // offload all layers to the GPU
+    llama_model * model = llama_load_model_from_file(params.model.c_str(), model_params);
+
+    if (model == NULL) {
+        fprintf(stderr , "%s: error: unable to load model\n" , __func__);
+        return 1;
+    }
+
+    // initialize the context
+    llama_context_params ctx_params = llama_context_default_params();
+
+    ctx_params.seed  = 1234;
+    ctx_params.n_ctx = 2048;
+    ctx_params.n_threads = params.n_threads;
+    ctx_params.n_threads_batch = params.n_threads_batch == -1 ? params.n_threads : params.n_threads_batch;
+
+    llama_context * ctx = llama_new_context_with_model(model, ctx_params);
+
+    if (ctx == NULL) {
+        fprintf(stderr , "%s: error: failed to create the llama_context\n" , __func__);
+        return 1;
+    }
+
+    // main loop
+
+    std::string prompt_template = "You will see two sentences. The first is marked INCORRECT and has a plethora of spelling and grammatical issues, the second is marked CORRECT and shows the fixed version of the prior sentence. INCORRECT:";
+    std::string prompt_suffix = " CORRECT: ";
+    std::string input_string = "";
+    LOG_TEE(">>>\n");
+    while (std::getline(std::cin, input_string, '\n')) {
+        if (input_string == "q") {
+            break;
+        }
+
+        // tokenize the prompt
+
+        params.prompt = prompt_template;
+        params.prompt += input_string;
+        params.prompt += prompt_suffix;
+
+        std::vector<llama_token> tokens_list;
+        tokens_list = ::llama_tokenize(ctx, params.prompt, true);
+
+        const int n_ctx    = llama_n_ctx(ctx);
+        const int n_kv_req = tokens_list.size() + (n_len - tokens_list.size());
+
+        LOG_TEE("\n%s: n_len = %d, n_ctx = %d, n_kv_req = %d\n", __func__, n_len, n_ctx, n_kv_req);
+
+        // make sure the KV cache is big enough to hold all the prompt and generated tokens
+        if (n_kv_req > n_ctx) {
+            LOG_TEE("%s: error: n_kv_req > n_ctx, the required KV cache size is not big enough\n", __func__);
+            LOG_TEE("%s:        either reduce n_len or increase n_ctx\n", __func__);
+            return 1;
+        }
+
+        // print the prompt token-by-token
+
+        fprintf(stderr, "\n");
+
+        for (auto id : tokens_list) {
+            fprintf(stderr, "%s", llama_token_to_piece(ctx, id).c_str());
+        }
+
+        fflush(stderr);
+
+        // create a llama_batch with size 512
+        // we use this object to submit token data for decoding
+
+        llama_batch batch = llama_batch_init(512, 0, 1);
+
+        // evaluate the initial prompt
+        for (size_t i = 0; i < tokens_list.size(); i++) {
+            llama_batch_add(batch, tokens_list[i], i, { 0 }, false);
+        }
+
+        // llama_decode will output logits only for the last token of the prompt
+        batch.logits[batch.n_tokens - 1] = true;
+
+        if (llama_decode(ctx, batch) != 0) {
+            LOG_TEE("%s: llama_decode() failed\n", __func__);
+            return 1;
+        }
+
+        int n_cur    = batch.n_tokens;
+        int n_decode = 0;
+
+        const auto t_main_start = ggml_time_us();
+
+        while (n_cur <= n_len) {
+            // sample the next token
+            {
+                auto   n_vocab = llama_n_vocab(model);
+                auto * logits  = llama_get_logits_ith(ctx, batch.n_tokens - 1);
+
+                std::vector<llama_token_data> candidates;
+                candidates.reserve(n_vocab);
+
+                for (llama_token token_id = 0; token_id < n_vocab; token_id++) {
+                    candidates.emplace_back(llama_token_data{ token_id, logits[token_id], 0.0f });
+                }
+
+                llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
+
+                // sample the most likely token
+                const llama_token new_token_id = llama_sample_token_greedy(ctx, &candidates_p);
+
+                // is it an end of generation?
+                if (llama_token_is_eog(model, new_token_id) || n_cur >= n_len) {
+                    LOG_TEE("\n");
+
+                    break;
+                }
+
+                LOG_TEE("%s", llama_token_to_piece(ctx, new_token_id).c_str());
+                fflush(stdout);
+
+                // prepare the next batch
+                llama_batch_clear(batch);
+
+                // push this new token for next evaluation
+                llama_batch_add(batch, new_token_id, n_cur, { 0 }, true);
+
+                n_decode += 1;
+            }
+
+            n_cur += 1;
+
+            // evaluate the current batch with the transformer model
+            if (llama_decode(ctx, batch)) {
+                fprintf(stderr, "%s : failed to eval, return code %d\n", __func__, 1);
+                return 1;
+            }
+        }
+        LOG_TEE("\n");
+
+        const auto t_main_end = ggml_time_us();
+
+        LOG_TEE("%s: decoded %d tokens in %.2f s, speed: %.2f t/s\n",
+                __func__, n_decode, (t_main_end - t_main_start) / 1000000.0f, n_decode / ((t_main_end - t_main_start) / 1000000.0f));
+
+        llama_print_timings(ctx);
+
+        fprintf(stderr, "\n");
+
+        llama_batch_free(batch);
+        LOG_TEE(">>>\n");
+    }
+
+
+    llama_free(ctx);
+    llama_free_model(model);
+
+    llama_backend_free();
+
+    return 0;
+}
\ No newline at end of file