GGUF: C++ refactor, backend support, misc fixes (llama/11030)

* GGUF: C++ refactor, backend support, misc fixes

remove ggml_tensor.backend

update CODEOWNERS [no ci]

remove gguf_get_data from API

revise GGUF API data types

ggml/CMakeLists.txt

@@ -243,7 +243,8 @@ set(GGML_PUBLIC_HEADERS
     include/ggml-metal.h
     include/ggml-rpc.h
     include/ggml-sycl.h
-    include/ggml-vulkan.h)
+    include/ggml-vulkan.h
+    include/gguf.h)
 
 set_target_properties(ggml PROPERTIES PUBLIC_HEADER "${GGML_PUBLIC_HEADERS}")
 #if (GGML_METAL)

ggml/include/ggml-cpp.h

@@ -7,6 +7,7 @@
 #include "ggml.h"
 #include "ggml-alloc.h"
 #include "ggml-backend.h"
+#include "gguf.h"
 #include <memory>
 
 // Smart pointers for ggml types

ggml/include/ggml.h

@@ -241,12 +241,6 @@
 #define GGML_ROPE_TYPE_MROPE  8
 #define GGML_ROPE_TYPE_VISION 24
 
-#define GGUF_MAGIC "GGUF"
-
-#define GGUF_VERSION 3
-
-#define GGUF_DEFAULT_ALIGNMENT 32
-
 #define GGML_UNUSED(x) (void)(x)
 
 #define GGML_PAD(x, n) (((x) + (n) - 1) & ~((n) - 1))
@@ -403,12 +397,6 @@ extern "C" {
         GGML_PREC_F32,
     };
 
-    enum ggml_backend_type {
-        GGML_BACKEND_TYPE_CPU = 0,
-        GGML_BACKEND_TYPE_GPU = 10,
-        GGML_BACKEND_TYPE_GPU_SPLIT = 20,
-    };
-
     // model file types
     enum ggml_ftype {
         GGML_FTYPE_UNKNOWN        = -1,
@@ -587,8 +575,6 @@ extern "C" {
     struct ggml_tensor {
         enum ggml_type type;
 
-        GGML_DEPRECATED(enum ggml_backend_type backend, "use the buffer type to find the storage location of the tensor");
-
         struct ggml_backend_buffer * buffer;
 
         int64_t ne[GGML_MAX_DIMS]; // number of elements
@@ -2111,132 +2097,6 @@ extern "C" {
                    int64_t   n_per_row,
                const float * imatrix);
 
-    //
-    // gguf
-    //
-
-    enum gguf_type {
-        GGUF_TYPE_UINT8   = 0,
-        GGUF_TYPE_INT8    = 1,
-        GGUF_TYPE_UINT16  = 2,
-        GGUF_TYPE_INT16   = 3,
-        GGUF_TYPE_UINT32  = 4,
-        GGUF_TYPE_INT32   = 5,
-        GGUF_TYPE_FLOAT32 = 6,
-        GGUF_TYPE_BOOL    = 7,
-        GGUF_TYPE_STRING  = 8,
-        GGUF_TYPE_ARRAY   = 9,
-        GGUF_TYPE_UINT64  = 10,
-        GGUF_TYPE_INT64   = 11,
-        GGUF_TYPE_FLOAT64 = 12,
-        GGUF_TYPE_COUNT,       // marks the end of the enum
-    };
-
-    struct gguf_context;
-
-    struct gguf_init_params {
-        bool no_alloc;
-
-        // if not NULL, create a ggml_context and allocate the tensor data in it
-        struct ggml_context ** ctx;
-    };
-
-    GGML_API struct gguf_context * gguf_init_empty(void);
-    GGML_API struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params);
-    //GGML_API struct gguf_context * gguf_init_from_buffer(..);
-
-    GGML_API void gguf_free(struct gguf_context * ctx);
-
-    GGML_API const char * gguf_type_name(enum gguf_type type);
-
-    GGML_API int    gguf_get_version    (const struct gguf_context * ctx);
-    GGML_API size_t gguf_get_alignment  (const struct gguf_context * ctx);
-    GGML_API size_t gguf_get_data_offset(const struct gguf_context * ctx);
-    GGML_API void * gguf_get_data       (const struct gguf_context * ctx);
-
-    GGML_API int          gguf_get_n_kv(const struct gguf_context * ctx);
-    GGML_API int          gguf_find_key(const struct gguf_context * ctx, const char * key);
-    GGML_API const char * gguf_get_key (const struct gguf_context * ctx, int key_id);
-
-    GGML_API enum gguf_type gguf_get_kv_type (const struct gguf_context * ctx, int key_id);
-    GGML_API enum gguf_type gguf_get_arr_type(const struct gguf_context * ctx, int key_id);
-
-    // will abort if the wrong type is used for the key
-    GGML_API uint8_t      gguf_get_val_u8  (const struct gguf_context * ctx, int key_id);
-    GGML_API int8_t       gguf_get_val_i8  (const struct gguf_context * ctx, int key_id);
-    GGML_API uint16_t     gguf_get_val_u16 (const struct gguf_context * ctx, int key_id);
-    GGML_API int16_t      gguf_get_val_i16 (const struct gguf_context * ctx, int key_id);
-    GGML_API uint32_t     gguf_get_val_u32 (const struct gguf_context * ctx, int key_id);
-    GGML_API int32_t      gguf_get_val_i32 (const struct gguf_context * ctx, int key_id);
-    GGML_API float        gguf_get_val_f32 (const struct gguf_context * ctx, int key_id);
-    GGML_API uint64_t     gguf_get_val_u64 (const struct gguf_context * ctx, int key_id);
-    GGML_API int64_t      gguf_get_val_i64 (const struct gguf_context * ctx, int key_id);
-    GGML_API double       gguf_get_val_f64 (const struct gguf_context * ctx, int key_id);
-    GGML_API bool         gguf_get_val_bool(const struct gguf_context * ctx, int key_id);
-    GGML_API const char * gguf_get_val_str (const struct gguf_context * ctx, int key_id);
-    GGML_API const void * gguf_get_val_data(const struct gguf_context * ctx, int key_id);
-    GGML_API int          gguf_get_arr_n   (const struct gguf_context * ctx, int key_id);
-    GGML_API const void * gguf_get_arr_data(const struct gguf_context * ctx, int key_id);
-    GGML_API const char * gguf_get_arr_str (const struct gguf_context * ctx, int key_id, int i);
-
-    GGML_API int            gguf_get_n_tensors    (const struct gguf_context * ctx);
-    GGML_API int            gguf_find_tensor      (const struct gguf_context * ctx, const char * name);
-    GGML_API size_t         gguf_get_tensor_offset(const struct gguf_context * ctx, int i);
-    GGML_API char *         gguf_get_tensor_name  (const struct gguf_context * ctx, int i);
-    GGML_API enum ggml_type gguf_get_tensor_type  (const struct gguf_context * ctx, int i);
-
-    // removes key if it exists
-    GGML_API void gguf_remove_key(struct gguf_context * ctx, const char * key);
-
-    // overrides existing values or adds a new one
-    GGML_API void gguf_set_val_u8  (struct gguf_context * ctx, const char * key, uint8_t  val);
-    GGML_API void gguf_set_val_i8  (struct gguf_context * ctx, const char * key, int8_t   val);
-    GGML_API void gguf_set_val_u16 (struct gguf_context * ctx, const char * key, uint16_t val);
-    GGML_API void gguf_set_val_i16 (struct gguf_context * ctx, const char * key, int16_t  val);
-    GGML_API void gguf_set_val_u32 (struct gguf_context * ctx, const char * key, uint32_t val);
-    GGML_API void gguf_set_val_i32 (struct gguf_context * ctx, const char * key, int32_t  val);
-    GGML_API void gguf_set_val_f32 (struct gguf_context * ctx, const char * key, float    val);
-    GGML_API void gguf_set_val_u64 (struct gguf_context * ctx, const char * key, uint64_t val);
-    GGML_API void gguf_set_val_i64 (struct gguf_context * ctx, const char * key, int64_t  val);
-    GGML_API void gguf_set_val_f64 (struct gguf_context * ctx, const char * key, double   val);
-    GGML_API void gguf_set_val_bool(struct gguf_context * ctx, const char * key, bool     val);
-    GGML_API void gguf_set_val_str (struct gguf_context * ctx, const char * key, const char * val);
-    GGML_API void gguf_set_arr_data(struct gguf_context * ctx, const char * key, enum gguf_type type, const void * data, int n);
-    GGML_API void gguf_set_arr_str (struct gguf_context * ctx, const char * key, const char ** data, int n);
-
-    // set or add KV pairs from another context
-    GGML_API void gguf_set_kv(struct gguf_context * ctx, struct gguf_context * src);
-
-    // manage tensor info
-    GGML_API void gguf_add_tensor(struct gguf_context * ctx, const struct ggml_tensor * tensor);
-    GGML_API void gguf_set_tensor_type(struct gguf_context * ctx, const char * name, enum ggml_type type);
-    GGML_API void gguf_set_tensor_data(struct gguf_context * ctx, const char * name, const void * data, size_t size);
-
-    // writing gguf files can be done in 2 ways:
-    //
-    // - write the entire gguf_context to a binary file in a single pass:
-    //
-    //   gguf_write_to_file(ctx, fname);
-    //
-    // - first prepare a file with a placeholder for the meta data, write the tensor data, then write the meta data:
-    //
-    //   FILE * f = fopen(fname, "wb");
-    //   fseek(f, gguf_get_meta_size(ctx), SEEK_SET);
-    //   fwrite(f, ...);
-    //   void * data = gguf_meta_get_meta_data(ctx);
-    //   fseek(f, 0, SEEK_SET);
-    //   fwrite(f, data, gguf_get_meta_size(ctx));
-    //   free(data);
-    //   fclose(f);
-    //
-
-    // write the entire context to a binary file
-    GGML_API void gguf_write_to_file(const struct gguf_context * ctx, const char * fname, bool only_meta);
-
-    // get the size in bytes of the meta data (header, kv pairs, tensor info) including padding
-    GGML_API size_t gguf_get_meta_size(const struct gguf_context * ctx);
-    GGML_API void   gguf_get_meta_data(const struct gguf_context * ctx, void * data);
-
 #ifdef __cplusplus
     // restrict not standard in C++
 #    if defined(__GNUC__)

ggml/src/CMakeLists.txt

@@ -208,6 +208,7 @@ add_library(ggml-base
             ../include/ggml-backend.h
             ../include/ggml-cpp.h
             ../include/ggml-opt.h
+            ../include/gguf.h
             ggml.c
             ggml-alloc.c
             ggml-backend.cpp
@@ -215,7 +216,8 @@ add_library(ggml-base
             ggml-threading.cpp
             ggml-threading.h
             ggml-quants.c
-            ggml-quants.h)
+            ggml-quants.h
+            gguf.cpp)
 
 target_include_directories(ggml-base PRIVATE .)
 

ggml/src/ggml-impl.h

@@ -3,6 +3,8 @@
 // GGML internal header
 
 #include "ggml.h"
+#include "gguf.h"
+
 #include <assert.h>
 #include <math.h>
 #include <stdlib.h> // load `stdlib.h` before other headers to work around MinGW bug: https://sourceforge.net/p/mingw-w64/bugs/192/
@@ -551,22 +553,15 @@ static inline ggml_bf16_t ggml_compute_fp32_to_bf16(float s) {
 #define GGML_FP32_TO_BF16(x) ggml_compute_fp32_to_bf16(x)
 #define GGML_BF16_TO_FP32(x) ggml_compute_bf16_to_fp32(x)
 
-// expose GGUF internals for test code
-
-GGML_API size_t gguf_type_size(enum gguf_type type);
-
-GGML_API struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_params params);
-
-struct gguf_buf {
-    void * data;
-    size_t size;
-    size_t offset;
-};
-GGML_API struct gguf_buf gguf_buf_init(size_t size);
-GGML_API void gguf_buf_free(struct gguf_buf buf);
-
-GGML_API void gguf_write_to_buf(const struct gguf_context * ctx, struct gguf_buf * buf, bool only_meta);
-
 #ifdef __cplusplus
 }
 #endif
+
+#ifdef __cplusplus
+#include <vector>
+
+// expose GGUF internals for test code
+GGML_API size_t gguf_type_size(enum gguf_type type);
+GGML_API struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_params params);
+GGML_API void gguf_write_to_buf(const struct gguf_context * ctx, std::vector<int8_t> & buf, bool only_meta);
+#endif // __cplusplus

ggml/src/ggml.c

@@ -1588,15 +1588,8 @@ static struct ggml_tensor * ggml_new_tensor_impl(
 
     struct ggml_tensor * const result = (struct ggml_tensor *)((char *)ctx->mem_buffer + obj_new->offs);
 
-#ifdef __clang__
-    // temporary until ggml_tensor::backend is removed
-    #pragma clang diagnostic push
-    #pragma clang diagnostic ignored "-Wdeprecated-declarations"
-#endif
-
     *result = (struct ggml_tensor) {
         /*.type         =*/ type,
-        /*.backend      =*/ GGML_BACKEND_TYPE_CPU,
         /*.buffer       =*/ NULL,
         /*.ne           =*/ { 1, 1, 1, 1 },
         /*.nb           =*/ { 0, 0, 0, 0 },
@@ -1612,10 +1605,6 @@ static struct ggml_tensor * ggml_new_tensor_impl(
         /*.padding      =*/ { 0 },
     };
 
-#ifdef __clang__
-    #pragma clang diagnostic pop
-#endif
-
     // TODO: this should not be needed as long as we don't rely on aligned SIMD loads
     //GGML_ASSERT_ALIGNED(result->data);
 
@@ -6417,1271 +6406,6 @@ size_t ggml_quantize_chunk(
 
 ////////////////////////////////////////////////////////////////////////////////
 
-struct gguf_str {
-    uint64_t n;  // GGUFv2
-    char * data;
-};
-
-static const size_t GGUF_TYPE_SIZE[GGUF_TYPE_COUNT] = {
-    [GGUF_TYPE_UINT8]   = sizeof(uint8_t),
-    [GGUF_TYPE_INT8]    = sizeof(int8_t),
-    [GGUF_TYPE_UINT16]  = sizeof(uint16_t),
-    [GGUF_TYPE_INT16]   = sizeof(int16_t),
-    [GGUF_TYPE_UINT32]  = sizeof(uint32_t),
-    [GGUF_TYPE_INT32]   = sizeof(int32_t),
-    [GGUF_TYPE_FLOAT32] = sizeof(float),
-    [GGUF_TYPE_BOOL]    = sizeof(bool),
-    [GGUF_TYPE_STRING]  = sizeof(struct gguf_str),
-    [GGUF_TYPE_UINT64]  = sizeof(uint64_t),
-    [GGUF_TYPE_INT64]   = sizeof(int64_t),
-    [GGUF_TYPE_FLOAT64] = sizeof(double),
-    [GGUF_TYPE_ARRAY]   = 0, // undefined
-};
-static_assert(GGUF_TYPE_COUNT == 13, "GGUF_TYPE_COUNT != 13");
-
-static const char * GGUF_TYPE_NAME[GGUF_TYPE_COUNT] = {
-    [GGUF_TYPE_UINT8]   = "u8",
-    [GGUF_TYPE_INT8]    = "i8",
-    [GGUF_TYPE_UINT16]  = "u16",
-    [GGUF_TYPE_INT16]   = "i16",
-    [GGUF_TYPE_UINT32]  = "u32",
-    [GGUF_TYPE_INT32]   = "i32",
-    [GGUF_TYPE_FLOAT32] = "f32",
-    [GGUF_TYPE_BOOL]    = "bool",
-    [GGUF_TYPE_STRING]  = "str",
-    [GGUF_TYPE_ARRAY]   = "arr",
-    [GGUF_TYPE_UINT64]  = "u64",
-    [GGUF_TYPE_INT64]   = "i64",
-    [GGUF_TYPE_FLOAT64] = "f64",
-};
-static_assert(GGUF_TYPE_COUNT == 13, "GGUF_TYPE_COUNT != 13");
-
-union gguf_value {
-    uint8_t  uint8;
-    int8_t   int8;
-    uint16_t uint16;
-    int16_t  int16;
-    uint32_t uint32;
-    int32_t  int32;
-    float    float32;
-    uint64_t uint64;
-    int64_t  int64;
-    double   float64;
-    bool     bool_;
-
-    struct gguf_str str;
-
-    struct {
-        enum gguf_type type;
-
-        uint64_t n;  // GGUFv2
-        void * data;
-    } arr;
-};
-
-struct gguf_kv {
-    struct gguf_str key;
-
-    enum  gguf_type  type;
-    union gguf_value value;
-};
-
-struct gguf_header {
-    char magic[4];
-
-    uint32_t version;
-    uint64_t n_tensors; // GGUFv2
-    uint64_t n_kv;      // GGUFv2
-};
-
-struct gguf_tensor_info {
-    struct gguf_str name;
-
-    uint32_t n_dims;
-    uint64_t ne[GGML_MAX_DIMS];
-
-    enum ggml_type type;
-
-    uint64_t offset; // offset from start of `data`, must be a multiple of `ALIGNMENT`
-
-    // for writing API
-    const void * data;
-    size_t size;
-};
-
-struct gguf_context {
-    struct gguf_header header;
-
-    struct gguf_kv          * kv;
-    struct gguf_tensor_info * infos;
-
-    size_t alignment;
-    size_t offset;    // offset of `data` from beginning of file
-    size_t size;      // size of `data` in bytes
-
-    //uint8_t * padding;
-    void * data;
-};
-
-size_t gguf_type_size(enum gguf_type type) {
-    GGML_ASSERT(0 <= type && type < GGUF_TYPE_COUNT);
-    return GGUF_TYPE_SIZE[type];
-}
-
-static bool gguf_tensor_info_sanitize(struct gguf_tensor_info * info) {
-    if (info->n_dims > GGML_MAX_DIMS) {
-        fprintf(stderr, "%s: invalid number of dimensions (%" PRIu32 ")\n", __func__, info->n_dims);
-        return false;
-    }
-
-    if (info->type < 0 || info->type >= GGML_TYPE_COUNT) {
-        fprintf(stderr, "%s: invalid type (%d)\n", __func__, info->type);
-        return false;
-    }
-
-    if (strlen(info->name.data) >= GGML_MAX_NAME) {
-        fprintf(stderr, "%s: tensor '%s' name is too long\n", __func__, info->name.data);
-        return false;
-    }
-
-    for (uint32_t i = 0; i < info->n_dims; ++i) {
-        if (info->ne[i] <= 0) {
-            fprintf(stderr, "%s: invalid number of elements (%" PRIu64 ")\n", __func__, info->ne[i]);
-            return false;
-        }
-    }
-
-    // prevent overflow for total number of elements
-    if (INT64_MAX/info->ne[1] <= info->ne[0]) {
-        fprintf(stderr, "%s: invalid number of elements (%" PRIu64 ")\n", __func__, info->ne[1]);
-        return false;
-    }
-
-    if (INT64_MAX/info->ne[2] <= info->ne[0]*info->ne[1]) {
-        fprintf(stderr, "%s: invalid number of elements (%" PRIu64 ")\n", __func__, info->ne[2]);
-        return false;
-    }
-
-    if (INT64_MAX/info->ne[3] <= info->ne[0]*info->ne[1]*info->ne[2]) {
-        fprintf(stderr, "%s: invalid number of elements (%" PRIu64 ")\n", __func__, info->ne[3]);
-        return false;
-    }
-
-    return true;
-}
-
-static bool gguf_fread_el(FILE * file, void * dst, size_t size, size_t * offset) {
-    const size_t n = fread(dst, 1, size, file);
-    *offset += n;
-    return n == size;
-}
-
-static bool gguf_fread_str(FILE * file, struct gguf_str * p, size_t * offset) {
-    p->n    = 0;
-    p->data = NULL;
-
-    bool ok = true;
-
-    ok = ok && gguf_fread_el(file, &p->n, sizeof(p->n), offset);
-
-    // early exit if string length is invalid, prevents from integer overflow
-    if (p->n == SIZE_MAX) {
-        fprintf(stderr, "%s: invalid string length (%" PRIu64 ")\n", __func__, p->n);
-        return false;
-    }
-
-    p->data = calloc(p->n + 1, 1);
-    if (!p->data) {
-        fprintf(stderr, "%s: failed to allocate memory for string of length %" PRIu64 "\n", __func__, p->n);
-        return false;
-    }
-
-    ok = ok && gguf_fread_el(file,  p->data, p->n, offset);
-
-    return ok;
-}
-
-static void gguf_free_kv(struct gguf_kv * kv) {
-    if (kv->key.data) {
-        GGML_FREE(kv->key.data);
-    }
-
-    if (kv->type == GGUF_TYPE_STRING) {
-        if (kv->value.str.data) {
-            GGML_FREE(kv->value.str.data);
-        }
-    }
-
-    if (kv->type == GGUF_TYPE_ARRAY) {
-        if (kv->value.arr.data) {
-            if (kv->value.arr.type == GGUF_TYPE_STRING) {
-                for (uint64_t j = 0; j < kv->value.arr.n; ++j) {
-                    struct gguf_str * str = &((struct gguf_str *) kv->value.arr.data)[j];
-                    if (str->data) {
-                        GGML_FREE(str->data);
-                    }
-                }
-            }
-            GGML_FREE(kv->value.arr.data);
-        }
-    }
-}
-
-struct gguf_context * gguf_init_empty(void) {
-    struct gguf_context * ctx = calloc(1, sizeof(struct gguf_context));
-    if (!ctx) {
-        fprintf(stderr, "%s: failed to allocate memory for context\n", __func__);
-        return NULL;
-    }
-
-    memcpy(ctx->header.magic, GGUF_MAGIC, sizeof(ctx->header.magic));
-    ctx->header.version   = GGUF_VERSION;
-    ctx->header.n_tensors = 0;
-    ctx->header.n_kv      = 0;
-
-    ctx->kv    = NULL;
-    ctx->infos = NULL;
-
-    ctx->alignment = GGUF_DEFAULT_ALIGNMENT;
-    ctx->offset    = 0;
-    ctx->size      = 0;
-
-    ctx->data = NULL;
-
-    return ctx;
-}
-
-struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_params params) {
-    // offset from start of file
-    size_t offset = 0;
-
-    char magic[4];
-
-    // check the magic before making allocations
-    {
-        gguf_fread_el(file, &magic, sizeof(magic), &offset);
-
-        for (uint32_t i = 0; i < sizeof(magic); i++) {
-            if (magic[i] != GGUF_MAGIC[i]) {
-                fprintf(stderr, "%s: invalid magic characters '%c%c%c%c'\n", __func__, magic[0], magic[1], magic[2], magic[3]);
-                return NULL;
-            }
-        }
-    }
-
-    bool ok = true;
-
-    struct gguf_context * ctx = calloc(1, sizeof(struct gguf_context));
-    if (!ctx) {
-        fprintf(stderr, "%s: failed to allocate memory for context\n", __func__);
-        return NULL;
-    }
-
-    // read the header
-    {
-        strncpy(ctx->header.magic, magic, 4);
-
-        ctx->kv    = NULL;
-        ctx->infos = NULL;
-        ctx->data  = NULL;
-
-        ok = ok && gguf_fread_el(file, &ctx->header.version,   sizeof(ctx->header.version),   &offset);
-        ok = ok && gguf_fread_el(file, &ctx->header.n_tensors, sizeof(ctx->header.n_tensors), &offset);
-        ok = ok && gguf_fread_el(file, &ctx->header.n_kv,      sizeof(ctx->header.n_kv),      &offset);
-
-        if (ctx->header.version == 1) {
-            fprintf(stderr, "%s: GGUFv1 is no longer supported. please use a more up-to-date version\n", __func__);
-            gguf_free(ctx);
-            return NULL;
-        }
-
-        // sanity-checks to prevent from integer/buffer overflows
-
-        ok = ok && (ctx->header.n_tensors < (SIZE_MAX/2)/sizeof(struct gguf_tensor_info));
-        ok = ok && (ctx->header.n_tensors < (SIZE_MAX/2)/ggml_tensor_overhead());
-        ok = ok && (ctx->header.n_kv      < (SIZE_MAX/2)/sizeof(struct gguf_kv));
-
-        if (!ok) {
-            fprintf(stderr, "%s: failed to read header\n", __func__);
-            gguf_free(ctx);
-            return NULL;
-        }
-    }
-
-    // read the kv pairs
-    {
-        const uint64_t n_kv = ctx->header.n_kv;
-
-        if (n_kv > 0) {
-            ctx->kv = calloc(n_kv, sizeof(struct gguf_kv));
-            if (!ctx->kv) {
-                fprintf(stderr, "%s: failed to allocate memory for kv pairs\n", __func__);
-                gguf_free(ctx);
-                return NULL;
-            }
-        }
-
-        for (uint64_t i = 0; i < n_kv; ++i) {
-            struct gguf_kv * kv = &ctx->kv[i];
-
-            //fprintf(stderr, "%s: reading kv %d\n", __func__, i);
-
-            ok = ok && gguf_fread_str(file, &kv->key,                    &offset);
-            ok = ok && gguf_fread_el (file, &kv->type, sizeof(kv->type), &offset);
-
-            //fprintf(stderr, "%s: reading kv with key %s\n", __func__, kv->key.data);
-
-            switch (kv->type) {
-                case GGUF_TYPE_UINT8:   ok = ok && gguf_fread_el (file, &kv->value.uint8,   sizeof(kv->value.uint8),   &offset); break;
-                case GGUF_TYPE_INT8:    ok = ok && gguf_fread_el (file, &kv->value.int8,    sizeof(kv->value.int8),    &offset); break;
-                case GGUF_TYPE_UINT16:  ok = ok && gguf_fread_el (file, &kv->value.uint16,  sizeof(kv->value.uint16),  &offset); break;
-                case GGUF_TYPE_INT16:   ok = ok && gguf_fread_el (file, &kv->value.int16,   sizeof(kv->value.int16),   &offset); break;
-                case GGUF_TYPE_UINT32:  ok = ok && gguf_fread_el (file, &kv->value.uint32,  sizeof(kv->value.uint32),  &offset); break;
-                case GGUF_TYPE_INT32:   ok = ok && gguf_fread_el (file, &kv->value.int32,   sizeof(kv->value.int32),   &offset); break;
-                case GGUF_TYPE_FLOAT32: ok = ok && gguf_fread_el (file, &kv->value.float32, sizeof(kv->value.float32), &offset); break;
-                case GGUF_TYPE_UINT64:  ok = ok && gguf_fread_el (file, &kv->value.uint64,  sizeof(kv->value.uint64),  &offset); break;
-                case GGUF_TYPE_INT64:   ok = ok && gguf_fread_el (file, &kv->value.int64,   sizeof(kv->value.int64),   &offset); break;
-                case GGUF_TYPE_FLOAT64: ok = ok && gguf_fread_el (file, &kv->value.float64, sizeof(kv->value.float64), &offset); break;
-                case GGUF_TYPE_BOOL:    ok = ok && gguf_fread_el (file, &kv->value.bool_,   sizeof(kv->value.bool_),   &offset); break;
-                case GGUF_TYPE_STRING:  ok = ok && gguf_fread_str(file, &kv->value.str,                                &offset); break;
-                case GGUF_TYPE_ARRAY:
-                    {
-                        ok = ok && gguf_fread_el(file, &kv->value.arr.type, sizeof(kv->value.arr.type), &offset);
-                        ok = ok && gguf_fread_el(file, &kv->value.arr.n,    sizeof(kv->value.arr.n),    &offset);
-
-                        switch (kv->value.arr.type) {
-                            case GGUF_TYPE_UINT8:
-                            case GGUF_TYPE_INT8:
-                            case GGUF_TYPE_UINT16:
-                            case GGUF_TYPE_INT16:
-                            case GGUF_TYPE_UINT32:
-                            case GGUF_TYPE_INT32:
-                            case GGUF_TYPE_FLOAT32:
-                            case GGUF_TYPE_UINT64:
-                            case GGUF_TYPE_INT64:
-                            case GGUF_TYPE_FLOAT64:
-                            case GGUF_TYPE_BOOL:
-                                {
-                                    // prevent from integer overflow in the malloc below
-                                    if (kv->value.arr.n >= SIZE_MAX/gguf_type_size(kv->value.arr.type)) {
-                                        fprintf(stderr, "%s: array size is too large (%" PRIu64 ")\n", __func__, kv->value.arr.n);
-                                        gguf_free(ctx);
-                                        return NULL;
-                                    }
-
-                                    kv->value.arr.data = calloc(kv->value.arr.n, gguf_type_size(kv->value.arr.type));
-                                    if (!kv->value.arr.data) {
-                                        fprintf(stderr, "%s: failed to allocate memory for array\n", __func__);
-                                        gguf_free(ctx);
-                                        return NULL;
-                                    }
-
-                                    ok = ok && gguf_fread_el(file, kv->value.arr.data, kv->value.arr.n * gguf_type_size(kv->value.arr.type), &offset);
-                                } break;
-                            case GGUF_TYPE_STRING:
-                                {
-                                    // prevent from integer overflow in the malloc below
-                                    if (kv->value.arr.n >= SIZE_MAX/sizeof(struct gguf_str)) {
-                                        fprintf(stderr, "%s: array size is too large (%" PRIu64 ")\n", __func__, kv->value.arr.n);
-                                        gguf_free(ctx);
-                                        return NULL;
-                                    }
-
-                                    kv->value.arr.data = calloc(kv->value.arr.n, sizeof(struct gguf_str));
-                                    if (!kv->value.arr.data) {
-                                        fprintf(stderr, "%s: failed to allocate memory for array\n", __func__);
-                                        gguf_free(ctx);
-                                        return NULL;
-                                    }
-
-                                    for (uint64_t j = 0; j < kv->value.arr.n; ++j) {
-                                        ok = ok && gguf_fread_str(file, &((struct gguf_str *) kv->value.arr.data)[j], &offset);
-                                    }
-                                } break;
-                            case GGUF_TYPE_ARRAY:
-                            default:
-                                {
-                                    fprintf(stderr, "%s: invalid array type %d\n", __func__, kv->value.arr.type);
-                                    ok = false;
-                                } break;
-                        }
-                    } break;
-                default:
-                    {
-                        fprintf(stderr, "%s: invalid type %d\n", __func__, kv->type);
-                        ok = false;
-                    } break;
-            }
-
-            if (!ok) {
-                break;
-            }
-        }
-
-        if (!ok) {
-            fprintf(stderr, "%s: failed to read key-value pairs\n", __func__);
-            gguf_free(ctx);
-            return NULL;
-        }
-    }
-
-    // read the tensor infos
-    if (ctx->header.n_tensors > 0) {
-        ctx->infos = calloc(ctx->header.n_tensors, sizeof(struct gguf_tensor_info));
-        if (!ctx->infos) {
-            fprintf(stderr, "%s: failed to allocate memory for tensor infos\n", __func__);
-            gguf_free(ctx);
-            return NULL;
-        }
-
-        for (uint64_t i = 0; i < ctx->header.n_tensors; ++i) {
-            struct gguf_tensor_info * info = &ctx->infos[i];
-
-            for (int j = 0; j < GGML_MAX_DIMS; ++j) {
-                info->ne[j] = 1;
-            }
-
-            ok = ok && gguf_fread_str(file, &info->name,                          &offset);
-            ok = ok && gguf_fread_el (file, &info->n_dims, sizeof(info->n_dims),  &offset);
-
-            ok = ok && (info->n_dims <= GGML_MAX_DIMS);
-
-            for (uint32_t j = 0; j < info->n_dims; ++j) {
-                ok = ok && gguf_fread_el(file, &info->ne[j], sizeof(info->ne[j]), &offset);
-            }
-
-            ok = ok && gguf_fread_el (file, &info->type,   sizeof(info->type),    &offset);
-            ok = ok && gguf_fread_el (file, &info->offset, sizeof(info->offset),  &offset);
-
-            ok = ok && gguf_tensor_info_sanitize(info);
-
-            // make sure there is no duplicated tensor names
-            for (uint64_t j = 0; j < i && ok; ++j) {
-                if (strcmp(info->name.data, ctx->infos[j].name.data) == 0) {
-                    fprintf(stderr, "%s: duplicated tensor name %s\n", __func__, info->name.data);
-                    ok = false;
-                }
-            }
-
-            if (!ok) {
-                fprintf(stderr, "%s: failed to read tensor info\n", __func__);
-                gguf_free(ctx);
-                return NULL;
-            }
-        }
-    }
-
-    ctx->alignment = GGUF_DEFAULT_ALIGNMENT;
-
-    int alignment_idx = gguf_find_key(ctx, "general.alignment");
-    if (alignment_idx != -1) {
-        ctx->alignment = gguf_get_val_u32(ctx, alignment_idx);
-    }
-
-    // we require the data section to be aligned, so take into account any padding
-    {
-        const size_t offset_pad = offset % ctx->alignment;
-
-        if (offset_pad != 0) {
-            offset += ctx->alignment - offset_pad;
-            fseek(file, offset, SEEK_SET);
-        }
-    }
-
-    // store the current file offset - this is where the data section starts
-    ctx->offset = offset;
-
-    // compute the total size of the data section, taking into account the alignment
-    {
-        ctx->size = 0;
-        for (uint64_t i = 0; i < ctx->header.n_tensors; ++i) {
-            struct gguf_tensor_info * info = &ctx->infos[i];
-
-            const int64_t ne =
-                (int64_t) info->ne[0] *
-                (int64_t) info->ne[1] *
-                (int64_t) info->ne[2] *
-                (int64_t) info->ne[3];
-
-            if (ggml_blck_size(info->type) == 0 ) {
-                // this tensor type support have been removed:
-                fprintf(stderr, "%s: tensor '%s' of type %d: %s\n",
-                        __func__, info->name.data, (int) info->type, ggml_type_name(info->type));
-                gguf_free(ctx);
-                return NULL;
-            }
-
-            if (ne % ggml_blck_size(info->type) != 0) {
-                fprintf(stderr, "%s: tensor '%s' of type %d (%s) number of elements (%" PRId64 ") is not a multiple of block size (%" PRId64 ")\n",
-                        __func__, info->name.data, (int) info->type, ggml_type_name(info->type), ne, ggml_blck_size(info->type));
-                gguf_free(ctx);
-                return NULL;
-            }
-
-            const size_t size_cur = ggml_row_size(info->type, ne);
-
-            ctx->size += GGML_PAD(size_cur, ctx->alignment);
-        }
-    }
-
-    // load the tensor data only if requested
-    if (params.ctx != NULL) {
-        // if the provided gguf_context is no_alloc, then we create "empty" tensors and do not read the binary blob
-        // otherwise, we load the binary blob into the created ggml_context as well, and point the "data" members of
-        // the ggml_tensor structs to the appropriate locations in the binary blob
-
-        // compute the exact size needed for the new ggml_context
-        const size_t mem_size =
-            params.no_alloc ?
-            (ctx->header.n_tensors    )*ggml_tensor_overhead() :
-            (ctx->header.n_tensors + 1)*ggml_tensor_overhead() + ctx->size;
-
-        struct ggml_init_params pdata = {
-            .mem_size   = mem_size,
-            .mem_buffer = NULL,
-            .no_alloc   = params.no_alloc,
-        };
-
-        *params.ctx = ggml_init(pdata);
-        if (*params.ctx == NULL) {
-            fprintf(stderr, "%s: failed to initialize context\n", __func__);
-            gguf_free(ctx);
-            return NULL;
-        }
-
-        struct ggml_context * ctx_data = *params.ctx;
-
-        struct ggml_tensor * data = NULL;
-
-        if (!params.no_alloc) {
-            data = ggml_new_tensor_1d(ctx_data, GGML_TYPE_I8, ctx->size);
-
-            ok = ok && data != NULL;
-
-            // read the binary blob with the tensor data
-            ok = ok && gguf_fread_el(file, data->data, ctx->size, &offset);
-
-            if (!ok) {
-                fprintf(stderr, "%s: failed to read tensor data\n", __func__);
-                ggml_free(ctx_data);
-                gguf_free(ctx);
-                return NULL;
-            }
-
-            ctx->data = data->data;
-        }
-
-        ggml_set_no_alloc(ctx_data, true);
-
-        // create the tensors
-        for (uint64_t i = 0; i < ctx->header.n_tensors; ++i) {
-            const int64_t ne[GGML_MAX_DIMS] = {
-                ctx->infos[i].ne[0],
-                ctx->infos[i].ne[1],
-                ctx->infos[i].ne[2],
-                ctx->infos[i].ne[3],
-            };
-
-            struct ggml_tensor * cur = ggml_new_tensor(ctx_data, ctx->infos[i].type, ctx->infos[i].n_dims, ne);
-
-            ok = ok && cur != NULL;
-
-            if (!ok) {
-                break;
-            }
-
-            ggml_set_name(cur, ctx->infos[i].name.data);
-
-            // point the data member to the appropriate location in the binary blob using the tensor infos
-            if (!params.no_alloc) {
-              //cur->data = (char *) data->data + ctx->infos[i].offset - ctx->offset; // offset from start of file
-                cur->data = (char *) data->data + ctx->infos[i].offset;               // offset from data
-            }
-        }
-
-        if (!ok) {
-            fprintf(stderr, "%s: failed to read the tensor data\n", __func__);
-            ggml_free(ctx_data);
-            gguf_free(ctx);
-            return NULL;
-        }
-
-        ggml_set_no_alloc(ctx_data, params.no_alloc);
-    }
-
-    return ctx;
-}
-
-struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params) {
-    FILE * file = ggml_fopen(fname, "rb");
-    if (!file) {
-        fprintf(stderr, "%s: failed to open '%s': '%s'\n", __func__, fname, strerror(errno));
-        return NULL;
-    }
-
-    struct gguf_context * result = gguf_init_from_file_impl(file, params);
-    fclose(file);
-    return result;
-}
-
-void gguf_free(struct gguf_context * ctx) {
-    if (ctx == NULL) {
-        return;
-    }
-
-    if (ctx->kv) {
-        // free string memory - not great..
-        for (uint64_t i = 0; i < ctx->header.n_kv; ++i) {
-            gguf_free_kv(&ctx->kv[i]);
-        }
-
-        GGML_FREE(ctx->kv);
-    }
-
-    if (ctx->infos) {
-        for (uint64_t i = 0; i < ctx->header.n_tensors; ++i) {
-            struct gguf_tensor_info * info = &ctx->infos[i];
-
-            if (info->name.data) {
-                GGML_FREE(info->name.data);
-            }
-        }
-
-        GGML_FREE(ctx->infos);
-    }
-
-    GGML_FREE(ctx);
-}
-
-const char * gguf_type_name(enum gguf_type type) {
-    return GGUF_TYPE_NAME[type];
-}
-
-int gguf_get_version(const struct gguf_context * ctx) {
-    return ctx->header.version;
-}
-
-size_t gguf_get_alignment(const struct gguf_context * ctx) {
-    return ctx->alignment;
-}
-
-size_t gguf_get_data_offset(const struct gguf_context * ctx) {
-    return ctx->offset;
-}
-
-void * gguf_get_data(const struct gguf_context * ctx) {
-    return ctx->data;
-}
-
-int gguf_get_n_kv(const struct gguf_context * ctx) {
-    return ctx->header.n_kv;
-}
-
-int gguf_find_key(const struct gguf_context * ctx, const char * key) {
-    // return -1 if key not found
-    int keyfound = -1;
-
-    const int n_kv = gguf_get_n_kv(ctx);
-
-    for (int i = 0; i < n_kv; ++i) {
-        if (strcmp(key, gguf_get_key(ctx, i)) == 0) {
-            keyfound = i;
-            break;
-        }
-    }
-
-    return keyfound;
-}
-
-const char * gguf_get_key(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    return ctx->kv[key_id].key.data;
-}
-
-enum gguf_type gguf_get_kv_type(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    return ctx->kv[key_id].type;
-}
-
-enum gguf_type gguf_get_arr_type(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_ARRAY);
-    return ctx->kv[key_id].value.arr.type;
-}
-
-const void * gguf_get_arr_data(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_ARRAY);
-    return ctx->kv[key_id].value.arr.data;
-}
-
-const char * gguf_get_arr_str(const struct gguf_context * ctx, int key_id, int i) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_ARRAY);
-    struct gguf_kv * kv = &ctx->kv[key_id];
-    struct gguf_str * str = &((struct gguf_str *) kv->value.arr.data)[i];
-    return str->data;
-}
-
-int gguf_get_arr_n(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_ARRAY);
-    return ctx->kv[key_id].value.arr.n;
-}
-
-uint8_t gguf_get_val_u8(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_UINT8);
-    return ctx->kv[key_id].value.uint8;
-}
-
-int8_t gguf_get_val_i8(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_INT8);
-    return ctx->kv[key_id].value.int8;
-}
-
-uint16_t gguf_get_val_u16(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_UINT16);
-    return ctx->kv[key_id].value.uint16;
-}
-
-int16_t gguf_get_val_i16(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_INT16);
-    return ctx->kv[key_id].value.int16;
-}
-
-uint32_t gguf_get_val_u32(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_UINT32);
-    return ctx->kv[key_id].value.uint32;
-}
-
-int32_t gguf_get_val_i32(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_INT32);
-    return ctx->kv[key_id].value.int32;
-}
-
-float gguf_get_val_f32(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_FLOAT32);
-    return ctx->kv[key_id].value.float32;
-}
-
-uint64_t gguf_get_val_u64(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_UINT64);
-    return ctx->kv[key_id].value.uint64;
-}
-
-int64_t gguf_get_val_i64(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_INT64);
-    return ctx->kv[key_id].value.int64;
-}
-
-double gguf_get_val_f64(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_FLOAT64);
-    return ctx->kv[key_id].value.float64;
-}
-
-bool gguf_get_val_bool(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_BOOL);
-    return ctx->kv[key_id].value.bool_;
-}
-
-const char * gguf_get_val_str(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type == GGUF_TYPE_STRING);
-    return ctx->kv[key_id].value.str.data;
-}
-
-const void * gguf_get_val_data(const struct gguf_context * ctx, int key_id) {
-    GGML_ASSERT(key_id >= 0 && key_id < gguf_get_n_kv(ctx));
-    GGML_ASSERT(ctx->kv[key_id].type != GGUF_TYPE_ARRAY);
-    GGML_ASSERT(ctx->kv[key_id].type != GGUF_TYPE_STRING);
-    return &ctx->kv[key_id].value;
-}
-
-int gguf_get_n_tensors(const struct gguf_context * ctx) {
-    return ctx->header.n_tensors;
-}
-
-int gguf_find_tensor(const struct gguf_context * ctx, const char * name) {
-    // return -1 if tensor not found
-    int tensorfound = -1;
-
-    const int n_tensors = gguf_get_n_tensors(ctx);
-
-    for (int i = 0; i < n_tensors; ++i) {
-        if (strcmp(name, gguf_get_tensor_name(ctx, i)) == 0) {
-            tensorfound = i;
-            break;
-        }
-    }
-
-    return tensorfound;
-}
-
-size_t gguf_get_tensor_offset(const struct gguf_context * ctx, int i) {
-    return ctx->infos[i].offset;
-}
-
-char * gguf_get_tensor_name(const struct gguf_context * ctx, int i) {
-    return ctx->infos[i].name.data;
-}
-
-enum ggml_type gguf_get_tensor_type(const struct gguf_context * ctx, int i) {
-    return ctx->infos[i].type;
-}
-
-// returns the index
-static int gguf_get_or_add_key(struct gguf_context * ctx, const char * key) {
-    const int idx = gguf_find_key(ctx, key);
-    if (idx >= 0) {
-        return idx;
-    }
-
-    const int n_kv = gguf_get_n_kv(ctx);
-
-    ctx->kv = realloc(ctx->kv, (n_kv + 1) * sizeof(struct gguf_kv));
-    ctx->kv[n_kv].key.n    = strlen(key);
-    ctx->kv[n_kv].key.data = strdup(key);
-    ctx->header.n_kv++;
-
-    return n_kv;
-}
-
-void gguf_remove_key(struct gguf_context * ctx, const char * key) {
-    const int idx = gguf_find_key(ctx, key);
-    if (idx >= 0) {
-        const int n_kv = gguf_get_n_kv(ctx);
-        gguf_free_kv(&ctx->kv[idx]);
-        for (int i = idx; i < n_kv-1; ++i) {
-            ctx->kv[i] = ctx->kv[i+1];
-        }
-        ctx->kv = realloc(ctx->kv, (n_kv - 1) * sizeof(struct gguf_kv));
-        ctx->header.n_kv--;
-    }
-}
-
-void gguf_set_val_u8(struct gguf_context * ctx, const char * key, uint8_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type        = GGUF_TYPE_UINT8;
-    ctx->kv[idx].value.uint8 = val;
-}
-
-void gguf_set_val_i8(struct gguf_context * ctx, const char * key, int8_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type       = GGUF_TYPE_INT8;
-    ctx->kv[idx].value.int8 = val;
-}
-
-void gguf_set_val_u16(struct gguf_context * ctx, const char * key, uint16_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type         = GGUF_TYPE_UINT16;
-    ctx->kv[idx].value.uint16 = val;
-}
-
-void gguf_set_val_i16(struct gguf_context * ctx, const char * key, int16_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type        = GGUF_TYPE_INT16;
-    ctx->kv[idx].value.int16 = val;
-}
-
-void gguf_set_val_u32(struct gguf_context * ctx, const char * key, uint32_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type         = GGUF_TYPE_UINT32;
-    ctx->kv[idx].value.uint32 = val;
-}
-
-void gguf_set_val_i32(struct gguf_context * ctx, const char * key, int32_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type        = GGUF_TYPE_INT32;
-    ctx->kv[idx].value.int32 = val;
-}
-
-void gguf_set_val_f32(struct gguf_context * ctx, const char * key, float val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type          = GGUF_TYPE_FLOAT32;
-    ctx->kv[idx].value.float32 = val;
-}
-
-void gguf_set_val_u64(struct gguf_context * ctx, const char * key, uint64_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type         = GGUF_TYPE_UINT64;
-    ctx->kv[idx].value.uint64 = val;
-}
-
-void gguf_set_val_i64(struct gguf_context * ctx, const char * key, int64_t val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type        = GGUF_TYPE_INT64;
-    ctx->kv[idx].value.int64 = val;
-}
-
-void gguf_set_val_f64(struct gguf_context * ctx, const char * key, double val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type          = GGUF_TYPE_FLOAT64;
-    ctx->kv[idx].value.float64 = val;
-}
-
-void gguf_set_val_bool(struct gguf_context * ctx, const char * key, bool val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type        = GGUF_TYPE_BOOL;
-    ctx->kv[idx].value.bool_ = val;
-}
-
-void gguf_set_val_str(struct gguf_context * ctx, const char * key, const char * val) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type           = GGUF_TYPE_STRING;
-    ctx->kv[idx].value.str.n    = strlen(val);
-    ctx->kv[idx].value.str.data = strdup(val);
-}
-
-void gguf_set_arr_data(struct gguf_context * ctx, const char * key, enum gguf_type type, const void * data, int n) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type           = GGUF_TYPE_ARRAY;
-    ctx->kv[idx].value.arr.type = type;
-    ctx->kv[idx].value.arr.n    = n;
-    ctx->kv[idx].value.arr.data = GGML_CALLOC(n, gguf_type_size(type));
-    memcpy(ctx->kv[idx].value.arr.data, data, n*gguf_type_size(type));
-}
-
-void gguf_set_arr_str(struct gguf_context * ctx, const char * key, const char ** data, int n) {
-    const int idx = gguf_get_or_add_key(ctx, key);
-
-    ctx->kv[idx].type           = GGUF_TYPE_ARRAY;
-    ctx->kv[idx].value.arr.type = GGUF_TYPE_STRING;
-    ctx->kv[idx].value.arr.n    = n;
-    ctx->kv[idx].value.arr.data = GGML_CALLOC(n, sizeof(struct gguf_str));
-    for (int i = 0; i < n; i++) {
-        struct gguf_str * str = &((struct gguf_str *)ctx->kv[idx].value.arr.data)[i];
-        str->n    = strlen(data[i]);
-        str->data = strdup(data[i]);
-    }
-}
-
-// set or add KV pairs from another context
-void gguf_set_kv(struct gguf_context * ctx, struct gguf_context * src) {
-    for (uint32_t i = 0; i < src->header.n_kv; i++) {
-        switch (src->kv[i].type) {
-            case GGUF_TYPE_UINT8:   gguf_set_val_u8  (ctx, src->kv[i].key.data, src->kv[i].value.uint8);    break;
-            case GGUF_TYPE_INT8:    gguf_set_val_i8  (ctx, src->kv[i].key.data, src->kv[i].value.int8);     break;
-            case GGUF_TYPE_UINT16:  gguf_set_val_u16 (ctx, src->kv[i].key.data, src->kv[i].value.uint16);   break;
-            case GGUF_TYPE_INT16:   gguf_set_val_i16 (ctx, src->kv[i].key.data, src->kv[i].value.int16);    break;
-            case GGUF_TYPE_UINT32:  gguf_set_val_u32 (ctx, src->kv[i].key.data, src->kv[i].value.uint32);   break;
-            case GGUF_TYPE_INT32:   gguf_set_val_i32 (ctx, src->kv[i].key.data, src->kv[i].value.int32);    break;
-            case GGUF_TYPE_FLOAT32: gguf_set_val_f32 (ctx, src->kv[i].key.data, src->kv[i].value.float32);  break;
-            case GGUF_TYPE_UINT64:  gguf_set_val_u64 (ctx, src->kv[i].key.data, src->kv[i].value.uint64);   break;
-            case GGUF_TYPE_INT64:   gguf_set_val_i64 (ctx, src->kv[i].key.data, src->kv[i].value.int64);    break;
-            case GGUF_TYPE_FLOAT64: gguf_set_val_f64 (ctx, src->kv[i].key.data, src->kv[i].value.float64);  break;
-            case GGUF_TYPE_BOOL:    gguf_set_val_bool(ctx, src->kv[i].key.data, src->kv[i].value.bool_);    break;
-            case GGUF_TYPE_STRING:  gguf_set_val_str (ctx, src->kv[i].key.data, src->kv[i].value.str.data); break;
-            case GGUF_TYPE_ARRAY:
-                {
-                    if (src->kv[i].value.arr.type == GGUF_TYPE_STRING) {
-                        const char ** data = GGML_CALLOC(src->kv[i].value.arr.n, sizeof(char *));
-                        for (uint32_t j = 0; j < src->kv[i].value.arr.n; j++) {
-                            data[j] = ((struct gguf_str *)src->kv[i].value.arr.data)[j].data;
-                        }
-                        gguf_set_arr_str(ctx, src->kv[i].key.data, data, src->kv[i].value.arr.n);
-                        GGML_FREE((void *)data);
-                    } else if (src->kv[i].value.arr.type == GGUF_TYPE_ARRAY) {
-                        GGML_ABORT("nested arrays not supported");
-                    } else {
-                        gguf_set_arr_data(ctx, src->kv[i].key.data, src->kv[i].value.arr.type, src->kv[i].value.arr.data, src->kv[i].value.arr.n);
-                    }
-                } break;
-            default: GGML_ABORT("invalid type");
-        }
-    }
-}
-
-void gguf_add_tensor(
-             struct gguf_context * ctx,
-        const struct ggml_tensor * tensor) {
-    GGML_ASSERT(tensor);
-    if (gguf_find_tensor(ctx, tensor->name) != -1) {
-        GGML_ABORT("duplicated tensor name");
-    }
-
-    const int idx = ctx->header.n_tensors;
-    ctx->infos = realloc(ctx->infos, (idx + 1)*sizeof(struct gguf_tensor_info));
-
-    ctx->infos[idx].name.n    = strlen(tensor->name);
-    ctx->infos[idx].name.data = strdup(tensor->name);
-
-    for (int i = 0; i < GGML_MAX_DIMS; ++i) {
-        ctx->infos[idx].ne[i] = 1;
-    }
-
-    ctx->infos[idx].n_dims = ggml_n_dims(tensor);
-    for (uint32_t i = 0; i < ctx->infos[idx].n_dims; i++) {
-        ctx->infos[idx].ne[i] = tensor->ne[i];
-    }
-
-    ctx->infos[idx].type   = tensor->type;
-    ctx->infos[idx].offset = 0;
-    ctx->infos[idx].data   = tensor->data;
-    ctx->infos[idx].size   = ggml_nbytes(tensor);
-
-    if (ctx->header.n_tensors > 0) {
-        ctx->infos[idx].offset = ctx->infos[idx - 1].offset + GGML_PAD(ctx->infos[idx - 1].size, ctx->alignment);
-    }
-
-    ctx->header.n_tensors++;
-}
-
-void gguf_set_tensor_type(struct gguf_context * ctx, const char * name, enum ggml_type type) {
-    const int idx = gguf_find_tensor(ctx, name);
-    if (idx < 0) {
-        GGML_ABORT("tensor not found");
-    }
-
-    ctx->infos[idx].type = type;
-}
-
-void gguf_set_tensor_data(struct gguf_context * ctx, const char * name, const void * data, size_t size) {
-    const int idx = gguf_find_tensor(ctx, name);
-    if (idx < 0) {
-        GGML_ABORT("tensor not found");
-    }
-
-    ctx->infos[idx].data = data;
-    ctx->infos[idx].size = size;
-
-    // update offsets
-    for (uint32_t i = idx + 1; i < ctx->header.n_tensors; ++i) {
-        ctx->infos[i].offset = ctx->infos[i - 1].offset + GGML_PAD(ctx->infos[i - 1].size, ctx->alignment);
-    }
-}
-
-//static void gguf_fwrite_str(FILE * file, const struct gguf_str * val) {
-//    fwrite(&val->n,   sizeof(val->n),    1, file);
-//    fwrite(val->data, sizeof(char), val->n, file);
-//}
-//
-//static void gguf_fwrite_el(FILE * file, const void * val, size_t size) {
-//    fwrite(val, sizeof(char), size, file);
-//}
-
-struct gguf_buf gguf_buf_init(size_t size) {
-    struct gguf_buf buf = {
-        /*buf.data   =*/ size == 0 ? NULL : GGML_CALLOC(1, size),
-        /*buf.size   =*/ size,
-        /*buf.offset =*/ 0,
-    };
-
-    return buf;
-}
-
-void gguf_buf_free(struct gguf_buf buf) {
-    if (buf.data) {
-        GGML_FREE(buf.data);
-    }
-}
-
-static void gguf_buf_grow(struct gguf_buf * buf, size_t size) {
-    if (buf->offset + size > buf->size) {
-        buf->size = 1.5*(buf->offset + size);
-        if (buf->data) {
-            buf->data = realloc(buf->data, buf->size);
-        }
-    }
-}
-
-static void gguf_bwrite_str(struct gguf_buf * buf, const struct gguf_str * val) {
-    gguf_buf_grow(buf, sizeof(val->n) + val->n);
-
-    if (buf->data) {
-        memcpy((char *) buf->data + buf->offset, &val->n, sizeof(val->n));
-    }
-    buf->offset += sizeof(val->n);
-
-    if (buf->data) {
-        memcpy((char *) buf->data + buf->offset, val->data, val->n);
-    }
-    buf->offset += val->n;
-}
-
-static void gguf_bwrite_el(struct gguf_buf * buf, const void * val, size_t el_size) {
-    gguf_buf_grow(buf, el_size);
-
-    if (buf->data) {
-        memcpy((char *) buf->data + buf->offset, val, el_size);
-    }
-    buf->offset += el_size;
-}
-
-void gguf_write_to_buf(const struct gguf_context * ctx, struct gguf_buf * buf, bool only_meta) {
-    // write header
-    gguf_bwrite_el(buf, &ctx->header.magic,     sizeof(ctx->header.magic));
-    gguf_bwrite_el(buf, &ctx->header.version,   sizeof(ctx->header.version));
-    gguf_bwrite_el(buf, &ctx->header.n_tensors, sizeof(ctx->header.n_tensors));
-    gguf_bwrite_el(buf, &ctx->header.n_kv,      sizeof(ctx->header.n_kv));
-
-    // write key-value pairs
-    for (uint32_t i = 0; i < ctx->header.n_kv; ++i) {
-        struct gguf_kv * kv = &ctx->kv[i];
-
-        gguf_bwrite_str(buf, &kv->key);
-        gguf_bwrite_el (buf, &kv->type, sizeof(kv->type));
-
-        switch (kv->type) {
-            case GGUF_TYPE_UINT8:   gguf_bwrite_el( buf, &kv->value.uint8,   sizeof(kv->value.uint8)  ); break;
-            case GGUF_TYPE_INT8:    gguf_bwrite_el (buf, &kv->value.int8,    sizeof(kv->value.int8)   ); break;
-            case GGUF_TYPE_UINT16:  gguf_bwrite_el (buf, &kv->value.uint16,  sizeof(kv->value.uint16) ); break;
-            case GGUF_TYPE_INT16:   gguf_bwrite_el (buf, &kv->value.int16,   sizeof(kv->value.int16)  ); break;
-            case GGUF_TYPE_UINT32:  gguf_bwrite_el (buf, &kv->value.uint32,  sizeof(kv->value.uint32) ); break;
-            case GGUF_TYPE_INT32:   gguf_bwrite_el (buf, &kv->value.int32,   sizeof(kv->value.int32)  ); break;
-            case GGUF_TYPE_FLOAT32: gguf_bwrite_el (buf, &kv->value.float32, sizeof(kv->value.float32)); break;
-            case GGUF_TYPE_UINT64:  gguf_bwrite_el (buf, &kv->value.uint64,  sizeof(kv->value.uint64) ); break;
-            case GGUF_TYPE_INT64:   gguf_bwrite_el (buf, &kv->value.int64,   sizeof(kv->value.int64)  ); break;
-            case GGUF_TYPE_FLOAT64: gguf_bwrite_el (buf, &kv->value.float64, sizeof(kv->value.float64)); break;
-            case GGUF_TYPE_BOOL:    gguf_bwrite_el (buf, &kv->value.bool_,   sizeof(kv->value.bool_)  ); break;
-            case GGUF_TYPE_STRING:  gguf_bwrite_str(buf, &kv->value.str                               ); break;
-            case GGUF_TYPE_ARRAY:
-                {
-                    gguf_bwrite_el(buf, &kv->value.arr.type, sizeof(kv->value.arr.type));
-                    gguf_bwrite_el(buf, &kv->value.arr.n,    sizeof(kv->value.arr.n)   );
-
-                    switch (kv->value.arr.type) {
-                        case GGUF_TYPE_UINT8:
-                        case GGUF_TYPE_INT8:
-                        case GGUF_TYPE_UINT16:
-                        case GGUF_TYPE_INT16:
-                        case GGUF_TYPE_UINT32:
-                        case GGUF_TYPE_INT32:
-                        case GGUF_TYPE_FLOAT32:
-                        case GGUF_TYPE_UINT64:
-                        case GGUF_TYPE_INT64:
-                        case GGUF_TYPE_FLOAT64:
-                        case GGUF_TYPE_BOOL:
-                            {
-                                gguf_bwrite_el(buf, kv->value.arr.data, kv->value.arr.n * gguf_type_size(kv->value.arr.type));
-                            } break;
-                        case GGUF_TYPE_STRING:
-                            {
-                                for (uint32_t j = 0; j < kv->value.arr.n; ++j) {
-                                    gguf_bwrite_str(buf, &((struct gguf_str *) kv->value.arr.data)[j]);
-                                }
-                            } break;
-                        case GGUF_TYPE_ARRAY:
-                        default: GGML_ABORT("invalid type");
-                    }
-                } break;
-            default: GGML_ABORT("invalid type");
-        }
-    }
-
-    // write tensor infos
-    for (uint32_t i = 0; i < ctx->header.n_tensors; ++i) {
-        struct gguf_tensor_info * info = &ctx->infos[i];
-
-        gguf_bwrite_str(buf, &info->name);
-        gguf_bwrite_el (buf, &info->n_dims, sizeof(info->n_dims));
-        for (uint32_t j = 0; j < info->n_dims; ++j) {
-            gguf_bwrite_el(buf, &info->ne[j], sizeof(info->ne[j]));
-        }
-        gguf_bwrite_el(buf, &info->type,   sizeof(info->type));
-        gguf_bwrite_el(buf, &info->offset, sizeof(info->offset));
-    }
-
-    // we require the data section to be aligned, so take into account any padding
-    {
-        const size_t offset     = buf->offset;
-        const size_t offset_pad = GGML_PAD(offset, ctx->alignment);
-
-        if (offset_pad != offset) {
-            uint8_t pad = 0;
-            for (size_t i = 0; i < offset_pad - offset; ++i) {
-                gguf_bwrite_el(buf, &pad, sizeof(pad));
-            }
-        }
-    }
-
-    if (only_meta) {
-        return;
-    }
-
-    size_t offset = 0;
-
-    // write tensor data
-    for (uint32_t i = 0; i < ctx->header.n_tensors; ++i) {
-        struct gguf_tensor_info * info = &ctx->infos[i];
-
-        const size_t size     = info->size;
-        const size_t size_pad = GGML_PAD(size, ctx->alignment);
-
-        gguf_bwrite_el(buf, info->data, size);
-
-        if (size_pad != size) {
-            uint8_t pad = 0;
-            for (size_t j = 0; j < size_pad - size; ++j) {
-                gguf_bwrite_el(buf, &pad, sizeof(pad));
-            }
-        }
-
-        GGML_ASSERT(offset == info->offset);
-
-        offset += size_pad;
-    }
-}
-
-void gguf_write_to_file(const struct gguf_context * ctx, const char * fname, bool only_meta) {
-    FILE * file = ggml_fopen(fname, "wb");
-    if (!file) {
-        GGML_ABORT("failed to open file for writing");
-    }
-
-    struct gguf_buf buf = gguf_buf_init(16*1024);
-
-    gguf_write_to_buf(ctx, &buf, only_meta);
-
-    fwrite(buf.data, 1, buf.offset, file);
-
-    gguf_buf_free(buf);
-
-    fclose(file);
-}
-
-size_t gguf_get_meta_size(const struct gguf_context * ctx) {
-    // no allocs - only compute size
-    struct gguf_buf buf = gguf_buf_init(0);
-
-    gguf_write_to_buf(ctx, &buf, true);
-
-    return buf.offset;
-}
-
-void gguf_get_meta_data(const struct gguf_context * ctx, void * data) {
-    struct gguf_buf buf = gguf_buf_init(16*1024);
-
-    gguf_write_to_buf(ctx, &buf, true);
-
-    memcpy(data, buf.data, buf.offset);
-
-    gguf_buf_free(buf);
-}
-
 void ggml_log_set(ggml_log_callback log_callback, void * user_data) {
     g_logger_state.log_callback = log_callback ? log_callback : ggml_log_callback_default;
     g_logger_state.log_callback_user_data = user_data;