whisper.rn 0.4.0-rc.8 → 0.4.0

package/cpp/gguf.cpp

@@ -0,0 +1,1347 @@
+#include "ggml.h"
+#include "ggml-backend.h"
+#include "ggml-impl.h"
+#include "gguf.h"
+
+#include <cinttypes>
+#include <cstddef>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <map>
+#include <new>
+#include <stdexcept>
+#include <string>
+#include <vector>
+
+template <typename T>
+struct type_to_wsp_gguf_type;
+
+template <>
+struct type_to_wsp_gguf_type<uint8_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_UINT8;
+};
+
+template <>
+struct type_to_wsp_gguf_type<int8_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_INT8;
+};
+
+template <>
+struct type_to_wsp_gguf_type<uint16_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_UINT16;
+};
+
+template <>
+struct type_to_wsp_gguf_type<int16_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_INT16;
+};
+
+template <>
+struct type_to_wsp_gguf_type<uint32_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_UINT32;
+};
+
+template <>
+struct type_to_wsp_gguf_type<int32_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_INT32;
+};
+
+template <>
+struct type_to_wsp_gguf_type<float> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_FLOAT32;
+};
+
+template <>
+struct type_to_wsp_gguf_type<bool> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_BOOL;
+};
+
+template <>
+struct type_to_wsp_gguf_type<std::string> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_STRING;
+};
+
+template <>
+struct type_to_wsp_gguf_type<uint64_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_UINT64;
+};
+
+template <>
+struct type_to_wsp_gguf_type<int64_t> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_INT64;
+};
+
+template <>
+struct type_to_wsp_gguf_type<double> {
+    static constexpr enum wsp_gguf_type value = WSP_GGUF_TYPE_FLOAT64;
+};
+
+static const std::map<wsp_gguf_type, size_t> WSP_GGUF_TYPE_SIZE = {
+    {WSP_GGUF_TYPE_UINT8,   sizeof(uint8_t)},
+    {WSP_GGUF_TYPE_INT8,    sizeof(int8_t)},
+    {WSP_GGUF_TYPE_UINT16,  sizeof(uint16_t)},
+    {WSP_GGUF_TYPE_INT16,   sizeof(int16_t)},
+    {WSP_GGUF_TYPE_UINT32,  sizeof(uint32_t)},
+    {WSP_GGUF_TYPE_INT32,   sizeof(int32_t)},
+    {WSP_GGUF_TYPE_FLOAT32, sizeof(float)},
+    {WSP_GGUF_TYPE_BOOL,    sizeof(int8_t)},
+    {WSP_GGUF_TYPE_STRING,  0}, // undefined
+    {WSP_GGUF_TYPE_ARRAY,   0}, // undefined
+    {WSP_GGUF_TYPE_UINT64,  sizeof(uint64_t)},
+    {WSP_GGUF_TYPE_INT64,   sizeof(int64_t)},
+    {WSP_GGUF_TYPE_FLOAT64, sizeof(double)},
+};
+static_assert(WSP_GGUF_TYPE_COUNT == 13, "WSP_GGUF_TYPE_COUNT != 13");
+
+static const std::map<wsp_gguf_type, const char *> WSP_GGUF_TYPE_NAME = {
+    {WSP_GGUF_TYPE_UINT8,   "u8"},
+    {WSP_GGUF_TYPE_INT8,    "i8"},
+    {WSP_GGUF_TYPE_UINT16,  "u16"},
+    {WSP_GGUF_TYPE_INT16,   "i16"},
+    {WSP_GGUF_TYPE_UINT32,  "u32"},
+    {WSP_GGUF_TYPE_INT32,   "i32"},
+    {WSP_GGUF_TYPE_FLOAT32, "f32"},
+    {WSP_GGUF_TYPE_BOOL,    "bool"},
+    {WSP_GGUF_TYPE_STRING,  "str"},
+    {WSP_GGUF_TYPE_ARRAY,   "arr"},
+    {WSP_GGUF_TYPE_UINT64,  "u64"},
+    {WSP_GGUF_TYPE_INT64,   "i64"},
+    {WSP_GGUF_TYPE_FLOAT64, "f64"},
+};
+static_assert(WSP_GGUF_TYPE_COUNT == 13, "WSP_GGUF_TYPE_COUNT != 13");
+
+size_t wsp_gguf_type_size(enum wsp_gguf_type type) {
+    auto it = WSP_GGUF_TYPE_SIZE.find(type);
+    return it == WSP_GGUF_TYPE_SIZE.end() ? 0 : it->second;
+}
+
+struct wsp_gguf_kv {
+    std::string key;
+
+    bool is_array;
+    enum wsp_gguf_type type;
+
+    std::vector<int8_t>      data;
+    std::vector<std::string> data_string;
+
+    template <typename T>
+    wsp_gguf_kv(const std::string & key, const T value)
+            : key(key), is_array(false), type(type_to_wsp_gguf_type<T>::value) {
+        WSP_GGML_ASSERT(!key.empty());
+        data.resize(sizeof(T));
+        memcpy(data.data(), &value, sizeof(T));
+    }
+
+    template <typename T>
+    wsp_gguf_kv(const std::string & key, const std::vector<T> & value)
+            : key(key), is_array(true), type(type_to_wsp_gguf_type<T>::value) {
+        WSP_GGML_ASSERT(!key.empty());
+        data.resize(value.size()*sizeof(T));
+        for (size_t i = 0; i < value.size(); ++i) {
+            const T tmp = value[i];
+            memcpy(data.data() + i*sizeof(T), &tmp, sizeof(T));
+        }
+    }
+
+    wsp_gguf_kv(const std::string & key, const std::string & value)
+            : key(key), is_array(false), type(WSP_GGUF_TYPE_STRING) {
+        WSP_GGML_ASSERT(!key.empty());
+        data_string.push_back(value);
+    }
+
+    wsp_gguf_kv(const std::string & key, const std::vector<std::string> & value)
+            : key(key), is_array(true), type(WSP_GGUF_TYPE_STRING) {
+        WSP_GGML_ASSERT(!key.empty());
+        data_string = value;
+    }
+
+    const std::string & get_key() const {
+        return key;
+    }
+
+    const enum wsp_gguf_type & get_type() const {
+        return type;
+    }
+
+    size_t get_ne() const {
+        if (type == WSP_GGUF_TYPE_STRING) {
+            const size_t ne = data_string.size();
+            WSP_GGML_ASSERT(is_array || ne == 1);
+            return ne;
+        }
+        const size_t type_size = wsp_gguf_type_size(type);
+        WSP_GGML_ASSERT(data.size() % type_size == 0);
+        const size_t ne = data.size() / type_size;
+        WSP_GGML_ASSERT(is_array || ne == 1);
+        return ne;
+    }
+
+    template <typename T>
+    const T & get_val(const size_t i = 0) const {
+        WSP_GGML_ASSERT(type_to_wsp_gguf_type<T>::value == type);
+        if constexpr (std::is_same<T, std::string>::value) {
+            WSP_GGML_ASSERT(data_string.size() >= i+1);
+            return data_string[i];
+        }
+        const size_t type_size = wsp_gguf_type_size(type);
+        WSP_GGML_ASSERT(data.size() % type_size == 0);
+        WSP_GGML_ASSERT(data.size() >= (i+1)*type_size);
+        return reinterpret_cast<const T *>(data.data())[i];
+    }
+
+    void cast(const enum wsp_gguf_type new_type) {
+        const size_t new_type_size = wsp_gguf_type_size(new_type);
+        WSP_GGML_ASSERT(data.size() % new_type_size == 0);
+        type = new_type;
+    }
+};
+
+struct wsp_gguf_tensor_info {
+    struct wsp_ggml_tensor t; // for holding the equivalent info
+    uint64_t offset;      // offset from start of `data`, must be a multiple of `ALIGNMENT`
+};
+
+struct wsp_gguf_context {
+    uint32_t version = WSP_GGUF_VERSION;
+
+    std::vector<struct wsp_gguf_kv> kv;
+    std::vector<struct wsp_gguf_tensor_info> info;
+
+    size_t alignment = WSP_GGUF_DEFAULT_ALIGNMENT;
+    size_t offset    = 0; // offset of `data` from beginning of file
+    size_t size      = 0; // size of `data` in bytes
+
+    void * data = nullptr;
+};
+
+struct wsp_gguf_reader {
+    FILE * file;
+
+    wsp_gguf_reader(FILE * file) : file(file) {}
+
+    template <typename T>
+    bool read(T & dst) const {
+        return fread(&dst, 1, sizeof(dst), file) == sizeof(dst);
+    }
+
+    template <typename T>
+    bool read(std::vector<T> & dst, const size_t n) const {
+        dst.resize(n);
+        for (size_t i = 0; i < dst.size(); ++i) {
+            if constexpr (std::is_same<T, bool>::value) {
+                bool tmp;
+                if (!read(tmp)) {
+                    return false;
+                }
+                dst[i] = tmp;
+            } else {
+                if (!read(dst[i])) {
+                    return false;
+                }
+            }
+        }
+        return true;
+    }
+
+    bool read(bool & dst) const {
+        int8_t tmp = -1;
+        if (!read(tmp)) {
+            return false;
+        }
+        dst = tmp != 0;
+        return true;
+    }
+
+    bool read(enum wsp_ggml_type & dst) const {
+        int32_t tmp = -1;
+        if (!read(tmp)) {
+            return false;
+        }
+        dst = wsp_ggml_type(tmp);
+        return true;
+    }
+
+    bool read(enum wsp_gguf_type & dst) const {
+        int32_t tmp = -1;
+        if (!read(tmp)) {
+            return false;
+        }
+        dst = wsp_gguf_type(tmp);
+        return true;
+    }
+
+    bool read(std::string & dst) const {
+        uint64_t size = -1;
+        if (!read(size)) {
+            return false;
+        }
+        dst.resize(size);
+        return fread(dst.data(), 1, dst.length(), file) == dst.length();
+    }
+
+    bool read(void * dst, const size_t size) const {
+        return fread(dst, 1, size, file) == size;
+    }
+};
+
+struct wsp_gguf_context * wsp_gguf_init_empty(void) {
+    return new wsp_gguf_context;
+}
+
+template<typename T>
+bool wsp_gguf_read_emplace_helper(const struct wsp_gguf_reader & gr, std::vector<struct wsp_gguf_kv> & kv, const std::string & key, const bool is_array, const size_t n) {
+    if (is_array) {
+        std::vector<T> value;
+        try {
+            if (!gr.read(value, n)) {
+                return false;
+            }
+        } catch (std::length_error &) {
+            WSP_GGML_LOG_ERROR("%s: encountered length_error while reading value for key '%s'\n", __func__, key.c_str());
+            return false;
+        } catch (std::bad_alloc &) {
+            WSP_GGML_LOG_ERROR("%s: encountered bad_alloc error while reading value for key '%s'\n", __func__, key.c_str());
+            return false;
+        }
+        kv.emplace_back(key, value);
+    } else {
+        T value;
+        if (!gr.read(value)) {
+            return false;
+        }
+        kv.emplace_back(key, value);
+    }
+    return true;
+}
+
+struct wsp_gguf_context * wsp_gguf_init_from_file_impl(FILE * file, struct wsp_gguf_init_params params) {
+    const struct wsp_gguf_reader gr(file);
+    struct wsp_gguf_context * ctx = new wsp_gguf_context;
+
+    bool ok = true;
+
+    // file magic
+    {
+        std::vector<char> magic;
+        ok = ok && gr.read(magic, 4);
+
+        if (!ok) {
+            WSP_GGML_LOG_ERROR("%s: failed to read magic\n", __func__);
+            wsp_gguf_free(ctx);
+            return nullptr;
+        }
+
+        for (uint32_t i = 0; i < magic.size(); i++) {
+            if (magic[i] != WSP_GGUF_MAGIC[i]) {
+                WSP_GGML_LOG_ERROR("%s: invalid magic characters: '%c%c%c%c', expected 'GGUF'\n", __func__, magic[0], magic[1], magic[2], magic[3]);
+                wsp_gguf_free(ctx);
+                return nullptr;
+            }
+        }
+    }
+
+    // header
+    int64_t n_kv      = 0;
+    int64_t n_tensors = 0;
+
+    if (ok && gr.read(ctx->version)) {
+        if (ok && ctx->version == 0) {
+            WSP_GGML_LOG_ERROR("%s: bad GGUF version: %" PRIu32 "\n", __func__, ctx->version);
+            ok = false;
+        }
+
+        /*
+         * bit layout is different when reading non-native endian models.
+         * assuming that the GGUF version is 3, the non-native endian model
+         * would read it as 0x30000000. we can use the AND operation against
+         * the last 4 hexadecimal digits to check if the model is the same
+         * endianness as the host system.
+        */
+        if (ok && (ctx->version & 0x0000FFFF) == 0x00000000) {
+            WSP_GGML_LOG_ERROR("%s: failed to load model: this GGUF file version %" PRIu32 " is extremely large, is there a mismatch between the host and model endianness?\n", __func__, ctx->version);
+            ok = false;
+        }
+
+        if (ok && ctx->version == 1) {
+            WSP_GGML_LOG_ERROR("%s: GGUFv1 is no longer supported, please use a more up-to-date version\n", __func__);
+            ok = false;
+        }
+        if (ok && ctx->version > WSP_GGUF_VERSION) {
+            WSP_GGML_LOG_ERROR("%s: this GGUF file is version %" PRIu32 " but this software only supports up to version %d\n",
+                __func__, ctx->version, WSP_GGUF_VERSION);
+            ok = false;
+        }
+    } else {
+        ok = false;
+    }
+
+    if (ok && gr.read(n_tensors)) {
+        static_assert(sizeof(size_t) <= 8 && sizeof(wsp_gguf_tensor_info) >= 2, "int64_t insufficient for indexing");
+        if (n_tensors < 0 || n_tensors > int64_t(SIZE_MAX/sizeof(wsp_gguf_tensor_info))) {
+            WSP_GGML_LOG_ERROR("%s: number of tensors is %" PRIi64 " but must be in [0, %zu]\n",
+                __func__, n_tensors, SIZE_MAX/sizeof(wsp_gguf_tensor_info));
+            ok = false;
+        }
+    } else {
+        ok = false;
+    }
+
+    if (ok && gr.read(n_kv)) {
+        static_assert(sizeof(size_t) <= 8 && sizeof(wsp_gguf_tensor_info) >= 2, "int64_t insufficient for indexing");
+        if (n_kv < 0 || n_kv > int64_t(SIZE_MAX/sizeof(wsp_gguf_kv))) {
+            WSP_GGML_LOG_ERROR("%s: number of key value pairs is %" PRIi64 " but must be in [0, %zu]\n",
+                    __func__, n_kv, SIZE_MAX/sizeof(wsp_gguf_kv));
+            ok = false;
+        }
+    } else {
+        ok = false;
+    }
+
+    if (!ok) {
+        WSP_GGML_LOG_ERROR("%s: failed to read header\n", __func__);
+        wsp_gguf_free(ctx);
+        return nullptr;
+    }
+
+    // KV pairs
+    {
+        for (int64_t i = 0; ok && i < n_kv; ++i) {
+            std::string key;
+            wsp_gguf_type   type     = wsp_gguf_type(-1);
+            bool        is_array = false;
+            uint64_t    n        = 1;
+
+            try {
+                ok = ok && gr.read(key);
+            } catch (std::length_error &) {
+                WSP_GGML_LOG_ERROR("%s: encountered length_error while reading key %" PRIi64 "\n", __func__, i);
+                ok = false;
+            } catch (std::bad_alloc &) {
+                WSP_GGML_LOG_ERROR("%s: encountered bad_alloc error while reading key %" PRIi64 "\n", __func__, i);
+                ok = false;
+            }
+            for (size_t j = 0; ok && j < ctx->kv.size(); ++j) {
+                if (key == ctx->kv[j].key) {
+                    WSP_GGML_LOG_ERROR("%s: duplicate key '%s' for tensors %zu and %" PRIi64 " \n", __func__, key.c_str(), j, i);
+                    ok = false;
+                }
+            }
+            if (!ok) {
+                break;
+            }
+
+            ok = ok && gr.read(type);
+            if (type == WSP_GGUF_TYPE_ARRAY) {
+                is_array = true;
+                ok = ok && gr.read(type);
+                ok = ok && gr.read(n);
+            }
+            if (!ok) {
+                break;
+            }
+
+            switch (type) {
+                case WSP_GGUF_TYPE_UINT8:   ok = ok && wsp_gguf_read_emplace_helper<uint8_t>    (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_INT8:    ok = ok && wsp_gguf_read_emplace_helper<int8_t>     (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_UINT16:  ok = ok && wsp_gguf_read_emplace_helper<uint16_t>   (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_INT16:   ok = ok && wsp_gguf_read_emplace_helper<int16_t>    (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_UINT32:  ok = ok && wsp_gguf_read_emplace_helper<uint32_t>   (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_INT32:   ok = ok && wsp_gguf_read_emplace_helper<int32_t>    (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_FLOAT32: ok = ok && wsp_gguf_read_emplace_helper<float>      (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_BOOL:    ok = ok && wsp_gguf_read_emplace_helper<bool>       (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_STRING:  ok = ok && wsp_gguf_read_emplace_helper<std::string>(gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_UINT64:  ok = ok && wsp_gguf_read_emplace_helper<uint64_t>   (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_INT64:   ok = ok && wsp_gguf_read_emplace_helper<int64_t>    (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_FLOAT64: ok = ok && wsp_gguf_read_emplace_helper<double>     (gr, ctx->kv, key, is_array, n); break;
+                case WSP_GGUF_TYPE_ARRAY:
+                default:
+                    {
+                        WSP_GGML_LOG_ERROR("%s: key '%s' has invalid GGUF type %d\n", __func__, key.c_str(), type);
+                        ok = false;
+                    } break;
+            }
+        }
+
+        if (!ok) {
+            WSP_GGML_LOG_ERROR("%s: failed to read key-value pairs\n", __func__);
+            wsp_gguf_free(ctx);
+            return nullptr;
+        }
+        WSP_GGML_ASSERT(int64_t(ctx->kv.size()) == n_kv);
+
+        const int alignment_idx = wsp_gguf_find_key(ctx, WSP_GGUF_KEY_GENERAL_ALIGNMENT);
+        ctx->alignment = alignment_idx == -1 ? WSP_GGUF_DEFAULT_ALIGNMENT : wsp_gguf_get_val_u32(ctx, alignment_idx);
+
+        if (ctx->alignment == 0 || (ctx->alignment & (ctx->alignment - 1)) != 0) {
+            WSP_GGML_LOG_ERROR("%s: alignment %zu is not a power of 2\n", __func__, ctx->alignment);
+            wsp_gguf_free(ctx);
+            return nullptr;
+        }
+    }
+
+    // read the tensor info
+    for (int64_t i = 0; ok && i < n_tensors; ++i) {
+        struct wsp_gguf_tensor_info info;
+
+        // tensor name
+        {
+            std::string name;
+            try {
+                ok = ok && gr.read(name);
+            } catch (std::length_error &) {
+                WSP_GGML_LOG_ERROR("%s: encountered length_error while reading tensor name %" PRIi64 "\n", __func__, i);
+                ok = false;
+            } catch (std::bad_alloc &) {
+                WSP_GGML_LOG_ERROR("%s: encountered bad_alloc error while reading tensor name %" PRIi64 "\n", __func__, i);
+                ok = false;
+            }
+            if (name.length() >= WSP_GGML_MAX_NAME) {
+                WSP_GGML_LOG_ERROR("%s: tensor name %" PRIi64 " is too long: %zu >= %d\n", __func__, i, name.length(), WSP_GGML_MAX_NAME);
+                ok = false;
+                break;
+            }
+            wsp_ggml_set_name(&info.t, name.c_str());
+
+            // make sure there are no duplicate tensor names
+            for (int64_t j = 0; ok && j < i; ++j) {
+                if (strcmp(info.t.name, ctx->info[j].t.name) == 0) {
+                    WSP_GGML_LOG_ERROR("%s: duplicate tensor name '%s' for tensors %" PRIi64 " and %" PRIi64 "\n", __func__, info.t.name, j, i);
+                    ok = false;
+                    break;
+                }
+            }
+        }
+        if (!ok) {
+            break;
+        }
+
+        // tensor shape
+        {
+            uint32_t n_dims = -1;
+            ok = ok && gr.read(n_dims);
+            if (n_dims > WSP_GGML_MAX_DIMS) {
+                WSP_GGML_LOG_ERROR("%s: tensor '%s' has invalid number of dimensions: %" PRIu32 " > %" PRIu32 "\n",
+                    __func__, info.t.name, n_dims, WSP_GGML_MAX_DIMS);
+                ok = false;
+                break;
+            }
+            for (uint32_t j = 0; ok && j < WSP_GGML_MAX_DIMS; ++j) {
+                info.t.ne[j] = 1;
+                if (j < n_dims) {
+                    ok = ok && gr.read(info.t.ne[j]);
+                }
+
+                // check that all ne are non-negative
+                if (info.t.ne[j] < 0) {
+                    WSP_GGML_LOG_ERROR("%s: tensor '%s' dimension %" PRIu32 " has invalid number of elements: %" PRIi64 " < 0\n",
+                        __func__, info.t.name, j, info.t.ne[j]);
+                    ok = false;
+                    break;
+                }
+            }
+
+            // check that the total number of elements is representable
+            if (ok && ((INT64_MAX/info.t.ne[1] <= info.t.ne[0]) ||
+                       (INT64_MAX/info.t.ne[2] <= info.t.ne[0]*info.t.ne[1]) ||
+                       (INT64_MAX/info.t.ne[3] <= info.t.ne[0]*info.t.ne[1]*info.t.ne[2]))) {
+
+                WSP_GGML_LOG_ERROR("%s: total number of elements in tensor '%s' with shape "
+                    "(%" PRIi64 ", %" PRIi64 ", %" PRIi64 ", %" PRIi64 ") is >= %" PRIi64 "\n",
+                    __func__, info.t.name, info.t.ne[0], info.t.ne[1], info.t.ne[2], info.t.ne[3], INT64_MAX);
+                ok = false;
+                break;
+            }
+        }
+        if (!ok) {
+            break;
+        }
+
+        // tensor type
+        {
+            ok = ok && gr.read(info.t.type);
+
+            // check that tensor type is within defined range
+            if (info.t.type < 0 || info.t.type >= WSP_GGML_TYPE_COUNT) {
+                WSP_GGML_LOG_ERROR("%s: tensor '%s' has invalid ggml type %d (%s)\n",
+                    __func__, info.t.name, info.t.type, wsp_ggml_type_name(info.t.type));
+                ok = false;
+                break;
+            }
+            const size_t  type_size = wsp_ggml_type_size(info.t.type);
+            const int64_t blck_size = wsp_ggml_blck_size(info.t.type);
+
+            // check that row size is divisible by block size
+            if (blck_size == 0 || info.t.ne[0] % blck_size != 0) {
+                WSP_GGML_LOG_ERROR("%s: tensor '%s' of type %d (%s) has %" PRId64 " elements per row, "
+                    "not a multiple of block size (%" PRId64 ")\n",
+                    __func__, info.t.name, (int) info.t.type, wsp_ggml_type_name(info.t.type), info.t.ne[0], blck_size);
+                ok = false;
+                break;
+            }
+
+            // calculate byte offsets given the tensor shape and type
+            info.t.nb[0] = type_size;
+            info.t.nb[1] = info.t.nb[0]*(info.t.ne[0]/blck_size);
+            for (int j = 2; j < WSP_GGML_MAX_DIMS; ++j) {
+                info.t.nb[j] = info.t.nb[j - 1]*info.t.ne[j - 1];
+            }
+        }
+        if (!ok) {
+            break;
+        }
+
+        // tensor data offset within buffer
+        ok = ok && gr.read(info.offset);
+
+        ctx->info.push_back(info);
+    }
+
+    if (!ok) {
+        WSP_GGML_LOG_ERROR("%s: failed to read tensor info\n", __func__);
+        wsp_gguf_free(ctx);
+        return nullptr;
+    }
+    WSP_GGML_ASSERT(int64_t(ctx->info.size()) == n_tensors);
+
+    // we require the data section to be aligned, so take into account any padding
+    if (fseek(file, WSP_GGML_PAD(ftell(file), ctx->alignment), SEEK_SET) != 0) {
+        WSP_GGML_LOG_ERROR("%s: failed to seek to beginning of data section\n", __func__);
+        wsp_gguf_free(ctx);
+        return nullptr;
+    }
+
+    // store the current file offset - this is where the data section starts
+    ctx->offset = ftell(file);
+
+    // compute the total size of the data section, taking into account the alignment
+    {
+        ctx->size = 0;
+        for (size_t i = 0; i < ctx->info.size(); ++i) {
+            const wsp_gguf_tensor_info & ti = ctx->info[i];
+            if (ti.offset != ctx->size) {
+                WSP_GGML_LOG_ERROR("%s: tensor '%s' has offset %" PRIu64 ", expected %zu\n",
+                    __func__, ti.t.name, ti.offset, ctx->size);
+                WSP_GGML_LOG_ERROR("%s: failed to read tensor data\n", __func__);
+                wsp_gguf_free(ctx);
+                return nullptr;
+            }
+            ctx->size += WSP_GGML_PAD(wsp_ggml_nbytes(&ti.t), ctx->alignment);
+        }
+    }
+
+    // load the tensor data only if requested
+    if (params.ctx != nullptr) {
+        // if the provided wsp_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 wsp_ggml_context as well, and point the "data" members of
+        //   the wsp_ggml_tensor structs to the appropriate locations in the binary blob
+
+        // compute the exact size needed for the new wsp_ggml_context
+        const size_t mem_size =
+            params.no_alloc ?
+            (n_tensors    )*wsp_ggml_tensor_overhead() :
+            (n_tensors + 1)*wsp_ggml_tensor_overhead() + ctx->size;
+
+        struct wsp_ggml_init_params pdata = {
+            /*mem_size   =*/ mem_size,
+            /*mem_buffer =*/ nullptr,
+            /*no_alloc   =*/ params.no_alloc,
+        };
+
+        *params.ctx = wsp_ggml_init(pdata);
+        if (*params.ctx == nullptr) {
+            WSP_GGML_LOG_ERROR("%s: failed to initialize ggml context for storing tensors\n", __func__);
+            wsp_gguf_free(ctx);
+            return nullptr;
+        }
+
+        struct wsp_ggml_context * ctx_data = *params.ctx;
+
+        struct wsp_ggml_tensor * data = nullptr;
+
+        if (!params.no_alloc) {
+            data = wsp_ggml_new_tensor_1d(ctx_data, WSP_GGML_TYPE_I8, ctx->size);
+
+            ok = ok && data != nullptr;
+
+            if (ok) {
+                wsp_ggml_set_name(data, "GGUF tensor data binary blob");
+            }
+
+            // read the binary blob with the tensor data
+            ok = ok && gr.read(data->data, ctx->size);
+
+            if (!ok) {
+                WSP_GGML_LOG_ERROR("%s: failed to read tensor data binary blob\n", __func__);
+                wsp_ggml_free(ctx_data);
+                *params.ctx = nullptr;
+                wsp_gguf_free(ctx);
+                return nullptr;
+            }
+
+            ctx->data = data->data;
+        }
+
+        wsp_ggml_set_no_alloc(ctx_data, true);
+
+        // create the tensors
+        for (size_t i = 0; i < ctx->info.size(); ++i) {
+            const struct wsp_gguf_tensor_info & info = ctx->info[i];
+
+            struct wsp_ggml_tensor * cur = wsp_ggml_new_tensor(ctx_data, info.t.type, WSP_GGML_MAX_DIMS, info.t.ne);
+
+            ok = ok && cur != nullptr;
+
+            if (!ok) {
+                break;
+            }
+
+            wsp_ggml_set_name(cur, info.t.name);
+
+            // point the data member to the appropriate location in the binary blob using the tensor info
+            if (!params.no_alloc) {
+                cur->data = (char *) data->data + info.offset;
+            }
+        }
+
+        if (!ok) {
+            WSP_GGML_LOG_ERROR("%s: failed to create tensors\n", __func__);
+            wsp_ggml_free(ctx_data);
+            *params.ctx = nullptr;
+            wsp_gguf_free(ctx);
+            return nullptr;
+        }
+
+        wsp_ggml_set_no_alloc(ctx_data, params.no_alloc);
+    }
+
+    return ctx;
+}
+
+struct wsp_gguf_context * wsp_gguf_init_from_file(const char * fname, struct wsp_gguf_init_params params) {
+    FILE * file = wsp_ggml_fopen(fname, "rb");
+
+    if (!file) {
+        WSP_GGML_LOG_ERROR("%s: failed to open GGUF file '%s'\n", __func__, fname);
+        return nullptr;
+    }
+
+    struct wsp_gguf_context * result = wsp_gguf_init_from_file_impl(file, params);
+    fclose(file);
+    return result;
+}
+
+void wsp_gguf_free(struct wsp_gguf_context * ctx) {
+    if (ctx == nullptr) {
+        return;
+    }
+    delete ctx;
+}
+
+const char * wsp_gguf_type_name(enum wsp_gguf_type type) {
+    auto it = WSP_GGUF_TYPE_NAME.find(type);
+    return it == WSP_GGUF_TYPE_NAME.end() ? nullptr : it->second;
+}
+
+uint32_t wsp_gguf_get_version(const struct wsp_gguf_context * ctx) {
+    return ctx->version;
+}
+
+size_t wsp_gguf_get_alignment(const struct wsp_gguf_context * ctx) {
+    return ctx->alignment;
+}
+
+size_t wsp_gguf_get_data_offset(const struct wsp_gguf_context * ctx) {
+    return ctx->offset;
+}
+
+int64_t wsp_gguf_get_n_kv(const struct wsp_gguf_context * ctx) {
+    return ctx->kv.size();
+}
+
+int64_t wsp_gguf_find_key(const struct wsp_gguf_context * ctx, const char * key) {
+    // return -1 if key not found
+    int64_t keyfound = -1;
+
+    const int64_t n_kv = wsp_gguf_get_n_kv(ctx);
+
+    for (int64_t i = 0; i < n_kv; ++i) {
+        if (strcmp(key, wsp_gguf_get_key(ctx, i)) == 0) {
+            keyfound = i;
+            break;
+        }
+    }
+
+    return keyfound;
+}
+
+const char * wsp_gguf_get_key(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    return ctx->kv[key_id].get_key().c_str();
+}
+
+enum wsp_gguf_type wsp_gguf_get_kv_type(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    return ctx->kv[key_id].is_array ? WSP_GGUF_TYPE_ARRAY : ctx->kv[key_id].get_type();
+}
+
+enum wsp_gguf_type wsp_gguf_get_arr_type(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].is_array);
+    return ctx->kv[key_id].get_type();
+}
+
+const void * wsp_gguf_get_arr_data(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_type() != WSP_GGUF_TYPE_STRING);
+    return ctx->kv[key_id].data.data();
+}
+
+const char * wsp_gguf_get_arr_str(const struct wsp_gguf_context * ctx, int64_t key_id, size_t i) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_type() == WSP_GGUF_TYPE_STRING);
+    return ctx->kv[key_id].data_string[i].c_str();
+}
+
+size_t wsp_gguf_get_arr_n(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+
+    if (ctx->kv[key_id].type == WSP_GGUF_TYPE_STRING) {
+        return ctx->kv[key_id].data_string.size();
+    }
+
+    const size_t type_size = wsp_gguf_type_size(ctx->kv[key_id].type);
+    WSP_GGML_ASSERT(ctx->kv[key_id].data.size() % type_size == 0);
+    return ctx->kv[key_id].data.size() / type_size;
+}
+
+uint8_t wsp_gguf_get_val_u8(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<uint8_t>();
+}
+
+int8_t wsp_gguf_get_val_i8(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<int8_t>();
+}
+
+uint16_t wsp_gguf_get_val_u16(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<uint16_t>();
+}
+
+int16_t wsp_gguf_get_val_i16(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<int16_t>();
+}
+
+uint32_t wsp_gguf_get_val_u32(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<uint32_t>();
+}
+
+int32_t wsp_gguf_get_val_i32(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<int32_t>();
+}
+
+float wsp_gguf_get_val_f32(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<float>();
+}
+
+uint64_t wsp_gguf_get_val_u64(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<uint64_t>();
+}
+
+int64_t wsp_gguf_get_val_i64(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<int64_t>();
+}
+
+double wsp_gguf_get_val_f64(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<double>();
+}
+
+bool wsp_gguf_get_val_bool(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<bool>();
+}
+
+const char * wsp_gguf_get_val_str(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    return ctx->kv[key_id].get_val<std::string>().c_str();
+}
+
+const void * wsp_gguf_get_val_data(const struct wsp_gguf_context * ctx, int64_t key_id) {
+    WSP_GGML_ASSERT(key_id >= 0 && key_id < wsp_gguf_get_n_kv(ctx));
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_ne() == 1);
+    WSP_GGML_ASSERT(ctx->kv[key_id].get_type() != WSP_GGUF_TYPE_STRING);
+    return ctx->kv[key_id].data.data();
+}
+
+int64_t wsp_gguf_get_n_tensors(const struct wsp_gguf_context * ctx) {
+    return ctx->info.size();
+}
+
+int64_t wsp_gguf_find_tensor(const struct wsp_gguf_context * ctx, const char * name) {
+    // return -1 if tensor not found
+    int64_t tensor_id = -1;
+
+    const int64_t n_tensors = wsp_gguf_get_n_tensors(ctx);
+
+    for (int64_t i = 0; i < n_tensors; ++i) {
+        if (strcmp(name, wsp_gguf_get_tensor_name(ctx, i)) == 0) {
+            tensor_id = i;
+            break;
+        }
+    }
+
+    return tensor_id;
+}
+
+size_t wsp_gguf_get_tensor_offset(const struct wsp_gguf_context * ctx, int64_t tensor_id) {
+    WSP_GGML_ASSERT(tensor_id >= 0 && tensor_id < wsp_gguf_get_n_tensors(ctx));
+    return ctx->info[tensor_id].offset;
+}
+
+const char * wsp_gguf_get_tensor_name(const struct wsp_gguf_context * ctx, int64_t tensor_id) {
+    WSP_GGML_ASSERT(tensor_id >= 0 && tensor_id < wsp_gguf_get_n_tensors(ctx));
+    return ctx->info[tensor_id].t.name;
+}
+
+enum wsp_ggml_type wsp_gguf_get_tensor_type(const struct wsp_gguf_context * ctx, int64_t tensor_id) {
+    WSP_GGML_ASSERT(tensor_id >= 0 && tensor_id < wsp_gguf_get_n_tensors(ctx));
+    return ctx->info[tensor_id].t.type;
+}
+
+size_t wsp_gguf_get_tensor_size(const struct wsp_gguf_context * ctx, int64_t tensor_id) {
+    WSP_GGML_ASSERT(tensor_id >= 0 && tensor_id < wsp_gguf_get_n_tensors(ctx));
+    return wsp_ggml_nbytes(&ctx->info[tensor_id].t);
+}
+
+int64_t wsp_gguf_remove_key(struct wsp_gguf_context * ctx, const char * key) {
+    const int64_t key_id = wsp_gguf_find_key(ctx, key);
+    if (key_id >= 0) {
+        ctx->kv.erase(ctx->kv.begin() + key_id);
+    }
+    return key_id;
+}
+
+template<typename T>
+static void wsp_gguf_check_reserved_keys(const std::string & key, const T val) {
+    if (key == WSP_GGUF_KEY_GENERAL_ALIGNMENT) {
+        if constexpr (std::is_same<T, uint32_t>::value) {
+            WSP_GGML_ASSERT(val > 0 && (val & (val - 1)) == 0 && WSP_GGUF_KEY_GENERAL_ALIGNMENT " must be power of 2");
+        } else {
+            WSP_GGML_UNUSED(val);
+            WSP_GGML_ABORT(WSP_GGUF_KEY_GENERAL_ALIGNMENT " must be type u32");
+        }
+    }
+}
+
+void wsp_gguf_set_val_u8(struct wsp_gguf_context * ctx, const char * key, uint8_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_i8(struct wsp_gguf_context * ctx, const char * key, int8_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_u16(struct wsp_gguf_context * ctx, const char * key, uint16_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_i16(struct wsp_gguf_context * ctx, const char * key, int16_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_u32(struct wsp_gguf_context * ctx, const char * key, uint32_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_i32(struct wsp_gguf_context * ctx, const char * key, int32_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_f32(struct wsp_gguf_context * ctx, const char * key, float val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_u64(struct wsp_gguf_context * ctx, const char * key, uint64_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_i64(struct wsp_gguf_context * ctx, const char * key, int64_t val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_f64(struct wsp_gguf_context * ctx, const char * key, double val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_bool(struct wsp_gguf_context * ctx, const char * key, bool val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, val);
+}
+
+void wsp_gguf_set_val_str(struct wsp_gguf_context * ctx, const char * key, const char * val) {
+    wsp_gguf_check_reserved_keys(key, val);
+    wsp_gguf_remove_key(ctx, key);
+    ctx->kv.emplace_back(key, std::string(val));
+}
+
+void wsp_gguf_set_arr_data(struct wsp_gguf_context * ctx, const char * key, enum wsp_gguf_type type, const void * data, size_t n) {
+    wsp_gguf_check_reserved_keys(key, data);
+    wsp_gguf_remove_key(ctx, key);
+
+    const size_t nbytes = n*wsp_gguf_type_size(type);
+    std::vector<int8_t> tmp(nbytes);
+    if (!tmp.empty()) {
+        memcpy(tmp.data(), data, nbytes);
+    }
+    ctx->kv.emplace_back(key, tmp);
+    ctx->kv.back().cast(type);
+}
+
+void wsp_gguf_set_arr_str(struct wsp_gguf_context * ctx, const char * key, const char ** data, size_t n) {
+    wsp_gguf_check_reserved_keys(key, data);
+    wsp_gguf_remove_key(ctx, key);
+
+    std::vector<std::string> tmp(n);
+    for (size_t i = 0; i < n; ++i) {
+        tmp[i] = data[i];
+    }
+    ctx->kv.emplace_back(key, tmp);
+}
+
+// set or add KV pairs from another context
+void wsp_gguf_set_kv(struct wsp_gguf_context * ctx, const struct wsp_gguf_context * src) {
+    const int64_t n_kv = wsp_gguf_get_n_kv(src);
+    for (int64_t i = 0; i < n_kv; ++i) {
+        const struct wsp_gguf_kv & kv = src->kv[i];
+
+        if (!kv.is_array) {
+            switch (kv.get_type()) {
+                case WSP_GGUF_TYPE_UINT8:   wsp_gguf_set_val_u8  (ctx, kv.get_key().c_str(), kv.get_val<uint8_t>());             break;
+                case WSP_GGUF_TYPE_INT8:    wsp_gguf_set_val_i8  (ctx, kv.get_key().c_str(), kv.get_val<int8_t>());              break;
+                case WSP_GGUF_TYPE_UINT16:  wsp_gguf_set_val_u16 (ctx, kv.get_key().c_str(), kv.get_val<uint16_t>());            break;
+                case WSP_GGUF_TYPE_INT16:   wsp_gguf_set_val_i16 (ctx, kv.get_key().c_str(), kv.get_val<int16_t>());             break;
+                case WSP_GGUF_TYPE_UINT32:  wsp_gguf_set_val_u32 (ctx, kv.get_key().c_str(), kv.get_val<uint32_t>());            break;
+                case WSP_GGUF_TYPE_INT32:   wsp_gguf_set_val_i32 (ctx, kv.get_key().c_str(), kv.get_val<int32_t>());             break;
+                case WSP_GGUF_TYPE_FLOAT32: wsp_gguf_set_val_f32 (ctx, kv.get_key().c_str(), kv.get_val<float>());               break;
+                case WSP_GGUF_TYPE_UINT64:  wsp_gguf_set_val_u64 (ctx, kv.get_key().c_str(), kv.get_val<uint64_t>());            break;
+                case WSP_GGUF_TYPE_INT64:   wsp_gguf_set_val_i64 (ctx, kv.get_key().c_str(), kv.get_val<int64_t>());             break;
+                case WSP_GGUF_TYPE_FLOAT64: wsp_gguf_set_val_f64 (ctx, kv.get_key().c_str(), kv.get_val<double>());              break;
+                case WSP_GGUF_TYPE_BOOL:    wsp_gguf_set_val_bool(ctx, kv.get_key().c_str(), kv.get_val<bool>());                break;
+                case WSP_GGUF_TYPE_STRING:  wsp_gguf_set_val_str (ctx, kv.get_key().c_str(), kv.get_val<std::string>().c_str()); break;
+                case WSP_GGUF_TYPE_ARRAY:
+                default: WSP_GGML_ABORT("invalid type");
+            }
+            continue;
+        }
+
+        const size_t ne = kv.get_ne();
+
+        switch (kv.get_type()) {
+            case WSP_GGUF_TYPE_UINT8:
+            case WSP_GGUF_TYPE_INT8:
+            case WSP_GGUF_TYPE_UINT16:
+            case WSP_GGUF_TYPE_INT16:
+            case WSP_GGUF_TYPE_UINT32:
+            case WSP_GGUF_TYPE_INT32:
+            case WSP_GGUF_TYPE_FLOAT32:
+            case WSP_GGUF_TYPE_UINT64:
+            case WSP_GGUF_TYPE_INT64:
+            case WSP_GGUF_TYPE_FLOAT64:
+            case WSP_GGUF_TYPE_BOOL: {
+                wsp_gguf_set_arr_data(ctx, kv.get_key().c_str(), kv.get_type(), kv.data.data(), ne);
+            } break;
+            case WSP_GGUF_TYPE_STRING: {
+                std::vector<const char *> tmp(ne);
+                for (size_t j = 0; j < ne; ++j) {
+                    tmp[j] = kv.data_string[j].c_str();
+                }
+                wsp_gguf_set_arr_str(ctx, kv.get_key().c_str(), tmp.data(), ne);
+            } break;
+            case WSP_GGUF_TYPE_ARRAY:
+            default: WSP_GGML_ABORT("invalid type");
+        }
+    }
+}
+
+void wsp_gguf_add_tensor(
+             struct wsp_gguf_context * ctx,
+        const struct wsp_ggml_tensor * tensor) {
+    WSP_GGML_ASSERT(tensor);
+    if (wsp_gguf_find_tensor(ctx, tensor->name) != -1) {
+        WSP_GGML_ABORT("duplicate tensor name: %s", tensor->name);
+    }
+
+    struct wsp_gguf_tensor_info ti;
+    ti.t = *tensor;
+    ti.offset = ctx->info.empty() ? 0 :
+        ctx->info.back().offset + WSP_GGML_PAD(wsp_ggml_nbytes(&ctx->info.back().t), ctx->alignment);
+    ctx->info.push_back(ti);
+}
+
+void wsp_gguf_set_tensor_type(struct wsp_gguf_context * ctx, const char * name, enum wsp_ggml_type type) {
+    const int64_t tensor_id = wsp_gguf_find_tensor(ctx, name);
+    if (tensor_id < 0) {
+        WSP_GGML_ABORT("tensor not found: %s", name);
+    }
+    struct wsp_ggml_tensor * tensor = &ctx->info[tensor_id].t;
+    const size_t  type_size = wsp_ggml_type_size(type);
+    const int64_t blck_size = wsp_ggml_blck_size(type);
+
+    tensor->type = type;
+    WSP_GGML_ASSERT(tensor->ne[0] % blck_size == 0 && "tensor row size not divisible by block size of new type");
+
+    tensor->nb[0] = type_size;
+    tensor->nb[1] = tensor->nb[0]*(tensor->ne[0]/blck_size);
+    for (int i = 2; i < WSP_GGML_MAX_DIMS; i++) {
+        tensor->nb[i] = tensor->nb[i - 1]*tensor->ne[i - 1];
+    }
+
+    // update offsets
+    const int64_t n_tensors = wsp_gguf_get_n_tensors(ctx);
+    for (int64_t i = tensor_id + 1; i < n_tensors; ++i) {
+        ctx->info[i].offset = ctx->info[i - 1].offset + WSP_GGML_PAD(wsp_ggml_nbytes(&ctx->info[i - 1].t), ctx->alignment);
+    }
+}
+
+void wsp_gguf_set_tensor_data(struct wsp_gguf_context * ctx, const char * name, const void * data) {
+    const int64_t tensor_id = wsp_gguf_find_tensor(ctx, name);
+    if (tensor_id < 0) {
+        WSP_GGML_ABORT("tensor not found: %s", name);
+    }
+
+    ctx->info[tensor_id].t.data = (void *)(uintptr_t)data; // double cast suppresses warning about casting away const
+}
+
+struct wsp_gguf_writer {
+    std::vector<int8_t> & buf;
+
+    wsp_gguf_writer(std::vector<int8_t> & buf) : buf(buf) {}
+
+    template <typename T>
+    void write(const T & val) const {
+        for (size_t i = 0; i < sizeof(val); ++i) {
+            buf.push_back(reinterpret_cast<const int8_t *>(&val)[i]);
+        }
+    }
+
+    void write(const std::vector<int8_t> & val) const {
+        buf.insert(buf.end(), val.begin(), val.end());
+    }
+
+    void write(const bool & val) const {
+        const int8_t val8 = val ? 1 : 0;
+        write(val8);
+    }
+
+    void write(const std::string & val) const {
+        {
+            const uint64_t n = val.length();
+            write(n);
+        }
+        for (size_t i = 0; i < val.length(); ++i) {
+            buf.push_back(reinterpret_cast<const int8_t *>(val.data())[i]);
+        }
+    }
+
+    void write(const char * val) const {
+        write(std::string(val));
+    }
+
+    void write(const enum wsp_ggml_type & val) const {
+        write(int32_t(val));
+    }
+
+    void write(const enum wsp_gguf_type & val) const {
+        write(int32_t(val));
+    }
+
+    void write(const struct wsp_gguf_kv & kv) const {
+        const uint64_t ne = kv.get_ne();
+
+        write(kv.get_key());
+
+        if (kv.is_array) {
+            write(WSP_GGUF_TYPE_ARRAY);
+            write(kv.get_type());
+            write(ne);
+        } else {
+            write(kv.get_type());
+        }
+
+        switch (kv.get_type()) {
+            case WSP_GGUF_TYPE_UINT8:
+            case WSP_GGUF_TYPE_INT8:
+            case WSP_GGUF_TYPE_UINT16:
+            case WSP_GGUF_TYPE_INT16:
+            case WSP_GGUF_TYPE_UINT32:
+            case WSP_GGUF_TYPE_INT32:
+            case WSP_GGUF_TYPE_FLOAT32:
+            case WSP_GGUF_TYPE_UINT64:
+            case WSP_GGUF_TYPE_INT64:
+            case WSP_GGUF_TYPE_FLOAT64: {
+                write(kv.data);
+            } break;
+            case WSP_GGUF_TYPE_BOOL: {
+                for (size_t i = 0; i < ne; ++i) {
+                    write(kv.get_val<bool>(i));
+                }
+            } break;
+            case WSP_GGUF_TYPE_STRING: {
+                for (size_t i = 0; i < ne; ++i) {
+                    write(kv.get_val<std::string>(i));
+                }
+            } break;
+            case WSP_GGUF_TYPE_ARRAY:
+            default: WSP_GGML_ABORT("invalid type");
+        }
+    }
+
+    void write_tensor_meta(const struct wsp_gguf_tensor_info & info) const {
+        write(info.t.name);
+
+        const uint32_t n_dims = wsp_ggml_n_dims(&info.t);
+        write(n_dims);
+
+        for (uint32_t j = 0; j < n_dims; ++j) {
+            write(info.t.ne[j]);
+        }
+        write(info.t.type);
+        write(info.offset);
+    }
+
+    void pad(const size_t alignment) const {
+        while (buf.size() % alignment != 0) {
+            const int8_t zero = 0;
+            write(zero);
+        }
+    }
+
+    void write_tensor_data(const struct wsp_gguf_tensor_info & info, const size_t offset_data, const size_t alignment) const {
+        WSP_GGML_ASSERT(buf.size() - offset_data == info.offset);
+
+        WSP_GGML_ASSERT(wsp_ggml_is_contiguous(&info.t));
+        const size_t offset = buf.size();
+        const size_t nbytes = wsp_ggml_nbytes(&info.t);
+
+        buf.resize(offset + nbytes);
+        if (info.t.buffer) {
+            wsp_ggml_backend_tensor_get(&info.t, buf.data() + offset, 0, nbytes);
+        } else {
+            WSP_GGML_ASSERT(info.t.data);
+            memcpy(buf.data() + offset, info.t.data, nbytes);
+        }
+
+        pad(alignment);
+    }
+};
+
+void wsp_gguf_write_to_buf(const struct wsp_gguf_context * ctx, std::vector<int8_t> & buf, bool only_meta) {
+    const struct wsp_gguf_writer gw(buf);
+
+    const int64_t n_kv      = wsp_gguf_get_n_kv(ctx);
+    const int64_t n_tensors = wsp_gguf_get_n_tensors(ctx);
+
+    // write header
+    gw.write(WSP_GGUF_MAGIC[0]);
+    gw.write(WSP_GGUF_MAGIC[1]);
+    gw.write(WSP_GGUF_MAGIC[2]);
+    gw.write(WSP_GGUF_MAGIC[3]);
+    gw.write(ctx->version);
+    gw.write(n_tensors);
+    gw.write(n_kv);
+
+    // write key-value pairs
+    for (int64_t i = 0; i < n_kv; ++i) {
+        gw.write(ctx->kv[i]);
+    }
+
+    // write tensor info
+    for (int64_t i = 0; i < n_tensors; ++i) {
+        gw.write_tensor_meta(ctx->info[i]);
+    }
+
+    // we require the data section to be aligned
+    gw.pad(ctx->alignment);
+
+    if (only_meta) {
+        return;
+    }
+
+    const size_t offset_data = gw.buf.size();
+
+    // write tensor data
+    for (int64_t i = 0; i < n_tensors; ++i) {
+        gw.write_tensor_data(ctx->info[i], offset_data, ctx->alignment);
+    }
+}
+
+bool wsp_gguf_write_to_file(const struct wsp_gguf_context * ctx, const char * fname, bool only_meta) {
+    FILE * file = wsp_ggml_fopen(fname, "wb");
+
+    if (!file) {
+        WSP_GGML_LOG_ERROR("%s: failed to open file '%s' for writing GGUF data\n", __func__, fname);
+        return false;
+    }
+
+    std::vector<int8_t> buf;
+    wsp_gguf_write_to_buf(ctx, buf, only_meta);
+    const bool ok = fwrite(buf.data(), 1, buf.size(), file) == buf.size();
+    fclose(file);
+    return ok;
+}
+
+size_t wsp_gguf_get_meta_size(const struct wsp_gguf_context * ctx) {
+    // only return size
+    std::vector<int8_t> buf;
+    wsp_gguf_write_to_buf(ctx, buf, /*only_meta =*/ true);
+    return buf.size();
+}
+
+void wsp_gguf_get_meta_data(const struct wsp_gguf_context * ctx, void * data) {
+    std::vector<int8_t> buf;
+    wsp_gguf_write_to_buf(ctx, buf, /*only_meta =*/ true);
+    memcpy(data, buf.data(), buf.size());
+}