diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt
index 9210e9fea5eb3..b2df4f2fb6cbc 100644
--- a/examples/CMakeLists.txt
+++ b/examples/CMakeLists.txt
@@ -20,7 +20,12 @@ else()
     add_subdirectory(batched)
     add_subdirectory(embedding)
     add_subdirectory(eval-callback)
-    add_subdirectory(gbnf-validator)
+
+    if (NOT WIN32)
+        # disabled on Windows because it uses internal functions not exported with LLAMA_API
+        add_subdirectory(gbnf-validator)
+    endif()
+
     add_subdirectory(gguf-hash)
     add_subdirectory(gguf-split)
     add_subdirectory(gguf)
@@ -51,7 +56,10 @@ else()
         add_subdirectory(convert-llama2c-to-ggml)
         add_subdirectory(cvector-generator)
         add_subdirectory(export-lora)
-        add_subdirectory(quantize-stats)
+        if (NOT WIN32)
+            # disabled on Windows because it uses internal functions not exported with LLAMA_API
+            add_subdirectory(quantize-stats)
+        endif()
         add_subdirectory(llava)
         if (GGML_RPC)
             add_subdirectory(rpc)
diff --git a/examples/gguf-split/gguf-split.cpp b/examples/gguf-split/gguf-split.cpp
index 7e62657e118a4..75f63f93891f5 100644
--- a/examples/gguf-split/gguf-split.cpp
+++ b/examples/gguf-split/gguf-split.cpp
@@ -287,7 +287,7 @@ struct split_strategy {
     }
 
     void print_info() {
-        printf("n_split: %ld\n", ctx_outs.size());
+        printf("n_split: %zu\n", ctx_outs.size());
         int i_split = 0;
         for (auto & ctx_out : ctx_outs) {
             // re-calculate the real gguf size for each split (= metadata size + total size of all tensors)
@@ -297,7 +297,7 @@ struct split_strategy {
                 total_size += ggml_nbytes(t);
             }
             total_size = total_size / 1000 / 1000; // convert to megabytes
-            printf("split %05d: n_tensors = %d, total_size = %ldM\n", i_split + 1, gguf_get_n_tensors(ctx_out), total_size);
+            printf("split %05d: n_tensors = %d, total_size = %zuM\n", i_split + 1, gguf_get_n_tensors(ctx_out), total_size);
             i_split++;
         }
     }
diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp
index bac606f471639..2338ad1067dde 100644
--- a/examples/llama-bench/llama-bench.cpp
+++ b/examples/llama-bench/llama-bench.cpp
@@ -1521,7 +1521,7 @@ int main(int argc, char ** argv) {
     for (const auto & inst : params_instances) {
         params_idx++;
         if (params.progress) {
-            fprintf(stderr, "llama-bench: benchmark %d/%ld: starting\n", params_idx, params_count);
+            fprintf(stderr, "llama-bench: benchmark %d/%zu: starting\n", params_idx, params_count);
         }
         // keep the same model between tests when possible
         if (!lmodel || !prev_inst || !inst.equal_mparams(*prev_inst)) {
@@ -1573,14 +1573,14 @@ int main(int argc, char ** argv) {
         // warmup run
         if (t.n_prompt > 0) {
             if (params.progress) {
-                fprintf(stderr, "llama-bench: benchmark %d/%ld: warmup prompt run\n", params_idx, params_count);
+                fprintf(stderr, "llama-bench: benchmark %d/%zu: warmup prompt run\n", params_idx, params_count);
             }
             //test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads);
             test_prompt(ctx, t.n_prompt, t.n_batch, t.n_threads);
         }
         if (t.n_gen > 0) {
             if (params.progress) {
-                fprintf(stderr, "llama-bench: benchmark %d/%ld: warmup generation run\n", params_idx, params_count);
+                fprintf(stderr, "llama-bench: benchmark %d/%zu: warmup generation run\n", params_idx, params_count);
             }
             test_gen(ctx, 1, t.n_threads);
         }
@@ -1592,14 +1592,14 @@ int main(int argc, char ** argv) {
 
             if (t.n_prompt > 0) {
                 if (params.progress) {
-                    fprintf(stderr, "llama-bench: benchmark %d/%ld: prompt run %d/%d\n", params_idx, params_count,
+                    fprintf(stderr, "llama-bench: benchmark %d/%zu: prompt run %d/%d\n", params_idx, params_count,
                             i + 1, params.reps);
                 }
                 test_prompt(ctx, t.n_prompt, t.n_batch, t.n_threads);
             }
             if (t.n_gen > 0) {
                 if (params.progress) {
-                    fprintf(stderr, "llama-bench: benchmark %d/%ld: generation run %d/%d\n", params_idx, params_count,
+                    fprintf(stderr, "llama-bench: benchmark %d/%zu: generation run %d/%d\n", params_idx, params_count,
                             i + 1, params.reps);
                 }
                 test_gen(ctx, t.n_gen, t.n_threads);
diff --git a/examples/retrieval/retrieval.cpp b/examples/retrieval/retrieval.cpp
index e78a8596d8cfe..23ff4db27a420 100644
--- a/examples/retrieval/retrieval.cpp
+++ b/examples/retrieval/retrieval.cpp
@@ -143,7 +143,7 @@ int main(int argc, char ** argv) {
         std::vector<chunk> file_chunk = chunk_file(context_file, params.chunk_size, params.chunk_separator);
         chunks.insert(chunks.end(), file_chunk.begin(), file_chunk.end());
     }
-    LOG_INF("Number of chunks: %ld\n", chunks.size());
+    LOG_INF("Number of chunks: %zu\n", chunks.size());
 
     llama_backend_init();
     llama_numa_init(params.numa);
diff --git a/examples/tokenize/tokenize.cpp b/examples/tokenize/tokenize.cpp
index 12ad542565128..c97e227242c42 100644
--- a/examples/tokenize/tokenize.cpp
+++ b/examples/tokenize/tokenize.cpp
@@ -394,7 +394,7 @@ int main(int raw_argc, char ** raw_argv) {
     }
 
     if (show_token_count) {
-        printf("Total number of tokens: %ld\n", tokens.size());
+        printf("Total number of tokens: %zu\n", tokens.size());
     }
     // silence valgrind
     llama_free(ctx);
diff --git a/ggml/CMakeLists.txt b/ggml/CMakeLists.txt
index 1b3d9896781f1..c91e931634255 100644
--- a/ggml/CMakeLists.txt
+++ b/ggml/CMakeLists.txt
@@ -32,6 +32,13 @@ else()
     endif()
 endif()
 
+# remove the lib prefix on win32 mingw
+if (WIN32)
+    set(CMAKE_STATIC_LIBRARY_PREFIX "")
+    set(CMAKE_SHARED_LIBRARY_PREFIX "")
+    set(CMAKE_SHARED_MODULE_PREFIX  "")
+endif()
+
 option(BUILD_SHARED_LIBS "ggml: build shared libraries" ${BUILD_SHARED_LIBS_DEFAULT})
 option(GGML_BACKEND_DL   "ggml: build backends as dynamic libraries (requires BUILD_SHARED_LIBS)" OFF)
 
diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt
index a267a8b596bd1..349f4c57fdf51 100644
--- a/ggml/src/CMakeLists.txt
+++ b/ggml/src/CMakeLists.txt
@@ -194,11 +194,6 @@ endif()
 
 if (WIN32)
     add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
-
-    if (BUILD_SHARED_LIBS)
-        # TODO: should not use this
-        set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
-    endif()
 endif()
 
 # ggml
diff --git a/ggml/src/ggml-cpu/amx/amx.cpp b/ggml/src/ggml-cpu/amx/amx.cpp
index b9074cb3aca55..5ec5263ceb4ba 100644
--- a/ggml/src/ggml-cpu/amx/amx.cpp
+++ b/ggml/src/ggml-cpu/amx/amx.cpp
@@ -122,7 +122,7 @@ static const char * ggml_backend_amx_buffer_type_get_name(ggml_backend_buffer_ty
 }
 
 static ggml_backend_buffer_t ggml_backend_amx_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
-    void * data = aligned_alloc(TENSOR_ALIGNMENT, size);
+    void * data = ggml_aligned_malloc(size);
     if (data == NULL) {
         fprintf(stderr, "%s: failed to allocate buffer of size %zu\n", __func__, size);
         return NULL;
diff --git a/ggml/src/ggml-cpu/ggml-cpu.c b/ggml/src/ggml-cpu/ggml-cpu.c
index ea17d6077e7cf..52881737c06fa 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.c
+++ b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -126,8 +126,7 @@ struct ggml_arm_arch_features_type {
 #endif
 #include <windows.h>
 
-
-#if !defined(__clang__)
+#if defined(_MSC_VER) && !defined(__clang__)
 #define GGML_CACHE_ALIGN __declspec(align(GGML_CACHE_LINE))
 
 typedef volatile LONG atomic_int;
@@ -12945,7 +12944,7 @@ static thread_ret_t ggml_graph_compute_secondary_thread(void* data);
 #include "windows.h"
 
 // TODO: support > 64 CPUs
-bool ggml_thread_apply_affinity(bool * mask) {
+static bool ggml_thread_apply_affinity(bool * mask) {
     HANDLE    h = GetCurrentThread();
     uint64_t  bitmask = 0ULL;
 
diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
index 00a1546a7d5d0..f961134edd735 100644
--- a/ggml/src/ggml-impl.h
+++ b/ggml/src/ggml-impl.h
@@ -74,8 +74,8 @@ static inline int ggml_up(int n, int m) {
 //
 
 GGML_ATTRIBUTE_FORMAT(2, 3)
-void ggml_log_internal        (enum ggml_log_level level, const char * format, ...);
-void ggml_log_callback_default(enum ggml_log_level level, const char * text, void * user_data);
+GGML_API void ggml_log_internal        (enum ggml_log_level level, const char * format, ...);
+GGML_API void ggml_log_callback_default(enum ggml_log_level level, const char * text, void * user_data);
 
 #define GGML_LOG(...)       ggml_log_internal(GGML_LOG_LEVEL_NONE , __VA_ARGS__)
 #define GGML_LOG_INFO(...)  ggml_log_internal(GGML_LOG_LEVEL_INFO , __VA_ARGS__)
@@ -304,8 +304,8 @@ struct ggml_cgraph ggml_graph_view(struct ggml_cgraph * cgraph, int i0, int i1);
 
 // Memory allocation
 
-void * ggml_aligned_malloc(size_t size);
-void ggml_aligned_free(void * ptr, size_t size);
+GGML_API void * ggml_aligned_malloc(size_t size);
+GGML_API void ggml_aligned_free(void * ptr, size_t size);
 
 // FP16 to FP32 conversion
 
diff --git a/ggml/src/ggml-threading.h b/ggml/src/ggml-threading.h
index ce975d880a70f..dec2c8840aa36 100644
--- a/ggml/src/ggml-threading.h
+++ b/ggml/src/ggml-threading.h
@@ -1,11 +1,13 @@
 #pragma once
 
+#include "ggml.h"
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-void ggml_critical_section_start(void);
-void ggml_critical_section_end(void);
+GGML_API void ggml_critical_section_start(void);
+GGML_API void ggml_critical_section_end(void);
 
 #ifdef __cplusplus
 }
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index f3b3908b11222..2d3ea09945790 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -1,10 +1,3 @@
-# TODO: should not use this
-if (WIN32)
-    if (BUILD_SHARED_LIBS)
-        set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
-    endif()
-endif()
-
 llama_add_compile_flags()
 
 #
diff --git a/src/llama.cpp b/src/llama.cpp
index cae3f76ad1677..49ef5b78a515c 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -1794,7 +1794,7 @@ struct llama_file {
         DWORD bufLen = FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS,
                                     NULL, error_code, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), (LPSTR)&lpMsgBuf, 0, NULL);
         if (!bufLen) {
-            ret = format("Win32 error code: %s", error_code);
+            ret = format("Win32 error code: %lx", error_code);
         } else {
             ret = lpMsgBuf;
             LocalFree(lpMsgBuf);
@@ -2132,7 +2132,7 @@ struct llama_mmap {
             HMODULE hKernel32 = GetModuleHandleW(L"kernel32.dll");
 
             // may fail on pre-Windows 8 systems
-            pPrefetchVirtualMemory = reinterpret_cast<decltype(pPrefetchVirtualMemory)> (GetProcAddress(hKernel32, "PrefetchVirtualMemory"));
+            pPrefetchVirtualMemory = (decltype(pPrefetchVirtualMemory))(void *) GetProcAddress(hKernel32, "PrefetchVirtualMemory");
 
             if (pPrefetchVirtualMemory) {
                 // advise the kernel to preload the mapped memory
@@ -21577,7 +21577,7 @@ float * llama_get_embeddings_ith(struct llama_context * ctx, int32_t i) {
                 throw std::runtime_error(format("negative index out of range [0, %d)", ctx->n_outputs));
             }
         } else if ((size_t) i >= ctx->output_ids.size()) {
-            throw std::runtime_error(format("out of range [0, %lu)", ctx->output_ids.size()));
+            throw std::runtime_error(format("out of range [0, %zu)", ctx->output_ids.size()));
         } else {
             j = ctx->output_ids[i];
         }
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
index 82373ff4e1862..daeed4564c1d1 100644
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -84,38 +84,50 @@ llama_test(test-tokenizer-0 NAME test-tokenizer-0-qwen2             ARGS ${CMAKE
 llama_test(test-tokenizer-0 NAME test-tokenizer-0-refact            ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-refact.gguf)
 llama_test(test-tokenizer-0 NAME test-tokenizer-0-starcoder         ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-starcoder.gguf)
 
-# build test-tokenizer-1-bpe target once and add many tests
-add_executable(test-tokenizer-1-bpe test-tokenizer-1-bpe.cpp)
-target_link_libraries(test-tokenizer-1-bpe PRIVATE common)
-install(TARGETS test-tokenizer-1-bpe RUNTIME)
-
-# TODO: disabled due to slowness
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-aquila    ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-aquila.gguf)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-falcon    ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-gpt-2     ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-gpt-2.gguf)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-gpt-neox  ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-gpt-neox.gguf)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-llama-bpe ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama-bpe.gguf --ignore-merges)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-mpt       ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-mpt.gguf)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-refact    ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-refact.gguf)
-#llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-starcoder ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-starcoder.gguf)
-
-# build test-tokenizer-1-spm target once and add many tests
-add_executable(test-tokenizer-1-spm test-tokenizer-1-spm.cpp)
-target_link_libraries(test-tokenizer-1-spm PRIVATE common)
-install(TARGETS test-tokenizer-1-spm RUNTIME)
-
-llama_test(test-tokenizer-1-spm  NAME test-tokenizer-1-llama-spm ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama-spm.gguf)
-#llama_test(test-tokenizer-1-spm  NAME test-tokenizer-1-baichuan  ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-baichuan.gguf)
-
-# llama_target_and_test(test-double-float.cpp) # SLOW
+
+if (NOT WIN32)
+    # these tests are disabled on Windows because they use internal functions not exported with LLAMA_API
+    llama_target_and_test(test-sampling.cpp)
+    llama_target_and_test(test-grammar-parser.cpp)
+    llama_target_and_test(test-grammar-integration.cpp)
+    llama_target_and_test(test-llama-grammar.cpp)
+    # TODO: disabled on loongarch64 because the ggml-ci node lacks Python 3.8
+    if (NOT ${CMAKE_SYSTEM_PROCESSOR} MATCHES "loongarch64")
+        llama_target_and_test(test-json-schema-to-grammar.cpp   WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/..)
+        target_include_directories(test-json-schema-to-grammar PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../examples/server)
+    endif()
+
+
+    # build test-tokenizer-1-bpe target once and add many tests
+    add_executable(test-tokenizer-1-bpe test-tokenizer-1-bpe.cpp)
+    target_link_libraries(test-tokenizer-1-bpe PRIVATE common)
+    install(TARGETS test-tokenizer-1-bpe RUNTIME)
+
+    # TODO: disabled due to slowness
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-aquila    ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-aquila.gguf)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-falcon    ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-gpt-2     ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-gpt-2.gguf)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-gpt-neox  ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-gpt-neox.gguf)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-llama-bpe ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama-bpe.gguf --ignore-merges)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-mpt       ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-mpt.gguf)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-refact    ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-refact.gguf)
+    #llama_test(test-tokenizer-1-bpe NAME test-tokenizer-1-starcoder ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-starcoder.gguf)
+
+    # build test-tokenizer-1-spm target once and add many tests
+    add_executable(test-tokenizer-1-spm test-tokenizer-1-spm.cpp)
+    target_link_libraries(test-tokenizer-1-spm PRIVATE common)
+    install(TARGETS test-tokenizer-1-spm RUNTIME)
+
+    llama_test(test-tokenizer-1-spm  NAME test-tokenizer-1-llama-spm ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama-spm.gguf)
+    #llama_test(test-tokenizer-1-spm  NAME test-tokenizer-1-baichuan  ARGS ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-baichuan.gguf)
+
+    # llama_target_and_test(test-double-float.cpp) # SLOW
+endif()
+
 llama_target_and_test(test-log.cpp)
 llama_target_and_test(test-arg-parser.cpp)
-llama_target_and_test(test-sampling.cpp)
 llama_target_and_test(test-chat-template.cpp)
 
-llama_target_and_test(test-grammar-parser.cpp)
-llama_target_and_test(test-grammar-integration.cpp)
-llama_target_and_test(test-llama-grammar.cpp)
 # llama_target_and_test(test-opt.cpp) # SLOW
 llama_target_and_test(test-backend-ops.cpp)
 
@@ -130,11 +142,6 @@ if (NOT GGML_BACKEND_DL)
     llama_target_and_test(test-rope.cpp)
 endif()
 
-# TODO: disabled on loongarch64 because the ggml-ci node lacks Python 3.8
-if (NOT ${CMAKE_SYSTEM_PROCESSOR} MATCHES "loongarch64")
-    llama_target_and_test(test-json-schema-to-grammar.cpp   WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/..)
-    target_include_directories(test-json-schema-to-grammar PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../examples/server)
-endif()
 
 # dummy executable - not installed
 get_filename_component(TEST_TARGET test-c.c NAME_WE)