llama-cpp-capacitor 0.0.6 → 0.0.7

package/cpp/gguf.h

@@ -0,0 +1,202 @@
+// This file contains functionality related to "GGUF" files, the binary file format used by ggml.
+// GGUF files have the following structure:
+//
+// 1. File magic "GGUF" (4 bytes).
+// 2. File version (uint32_t).
+// 3. Number of ggml tensors in file (int64_t).
+// 4. Number of key-value-pairs in file (int64_t).
+// 5. For each KV pair:
+//   1. The key (string).
+//   2. The value type (lm_gguf_type).
+//   3a. If the value type is LM_GGUF_TYPE_ARRAY:
+//     1. The type of the array (lm_gguf_type).
+//     2. The number of elements in the array (uint64_t).
+//     3. The binary representation of each element in the array.
+//   3b. Otherwise:
+//     1. The binary representation of the value.
+// 6. For each ggml tensor:
+//   1. The tensor name (string).
+//   2. The number of dimensions of the tensor (uint32_t).
+//   3. For each dimension:
+//     1. The size of the tensor in the dimension (int64_t).
+//   4. The tensor data type (lm_ggml_type).
+//   5. The tensor data offset in the tensor data binary blob (uint64_t).
+// 7. The tensor data binary blob (optional, aligned).
+//
+// Strings are serialized as the string length (uint64_t) followed by the C string without the null terminator.
+// All enums are stored as int32_t.
+// All bool values are stored as int8_t.
+// If the special key "general.alignment" (uint32_t) is defined it is used for alignment,
+//   otherwise LM_GGUF_DEFAULT_ALIGNMENT is used.
+//
+// Module maintainer: Johannes Gäßler (@JohannesGaessler, johannesg@5d6.de)
+
+#pragma once
+
+#include "ggml.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+
+#define LM_GGUF_MAGIC   "GGUF"
+#define LM_GGUF_VERSION 3
+
+#define LM_GGUF_KEY_GENERAL_ALIGNMENT "general.alignment"
+
+#define LM_GGUF_DEFAULT_ALIGNMENT 32
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+    // types that can be stored as GGUF KV data
+    enum lm_gguf_type {
+        LM_GGUF_TYPE_UINT8   = 0,
+        LM_GGUF_TYPE_INT8    = 1,
+        LM_GGUF_TYPE_UINT16  = 2,
+        LM_GGUF_TYPE_INT16   = 3,
+        LM_GGUF_TYPE_UINT32  = 4,
+        LM_GGUF_TYPE_INT32   = 5,
+        LM_GGUF_TYPE_FLOAT32 = 6,
+        LM_GGUF_TYPE_BOOL    = 7,
+        LM_GGUF_TYPE_STRING  = 8,
+        LM_GGUF_TYPE_ARRAY   = 9,
+        LM_GGUF_TYPE_UINT64  = 10,
+        LM_GGUF_TYPE_INT64   = 11,
+        LM_GGUF_TYPE_FLOAT64 = 12,
+        LM_GGUF_TYPE_COUNT,       // marks the end of the enum
+    };
+
+    struct lm_gguf_context;
+
+    struct lm_gguf_init_params {
+        bool no_alloc;
+
+        // if not NULL, create a lm_ggml_context and allocate the tensor data in it
+        struct lm_ggml_context ** ctx;
+    };
+
+    LM_GGML_API struct lm_gguf_context * lm_gguf_init_empty(void);
+    LM_GGML_API struct lm_gguf_context * lm_gguf_init_from_file(const char * fname, struct lm_gguf_init_params params);
+    //LM_GGML_API struct lm_gguf_context * lm_gguf_init_from_buffer(..);
+
+    LM_GGML_API void lm_gguf_free(struct lm_gguf_context * ctx);
+
+    LM_GGML_API const char * lm_gguf_type_name(enum lm_gguf_type type);
+
+    LM_GGML_API uint32_t lm_gguf_get_version    (const struct lm_gguf_context * ctx);
+    LM_GGML_API size_t   lm_gguf_get_alignment  (const struct lm_gguf_context * ctx);
+    LM_GGML_API size_t   lm_gguf_get_data_offset(const struct lm_gguf_context * ctx);
+
+    LM_GGML_API int64_t      lm_gguf_get_n_kv(const struct lm_gguf_context * ctx);
+    LM_GGML_API int64_t      lm_gguf_find_key(const struct lm_gguf_context * ctx, const char * key); // returns -1 if key is not found
+    LM_GGML_API const char * lm_gguf_get_key (const struct lm_gguf_context * ctx, int64_t key_id);
+
+    LM_GGML_API enum lm_gguf_type lm_gguf_get_kv_type (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API enum lm_gguf_type lm_gguf_get_arr_type(const struct lm_gguf_context * ctx, int64_t key_id);
+
+    // will abort if the wrong type is used for the key
+    LM_GGML_API uint8_t      lm_gguf_get_val_u8  (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API int8_t       lm_gguf_get_val_i8  (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API uint16_t     lm_gguf_get_val_u16 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API int16_t      lm_gguf_get_val_i16 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API uint32_t     lm_gguf_get_val_u32 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API int32_t      lm_gguf_get_val_i32 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API float        lm_gguf_get_val_f32 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API uint64_t     lm_gguf_get_val_u64 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API int64_t      lm_gguf_get_val_i64 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API double       lm_gguf_get_val_f64 (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API bool         lm_gguf_get_val_bool(const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API const char * lm_gguf_get_val_str (const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API const void * lm_gguf_get_val_data(const struct lm_gguf_context * ctx, int64_t key_id);
+    LM_GGML_API size_t       lm_gguf_get_arr_n   (const struct lm_gguf_context * ctx, int64_t key_id);
+
+    // get raw pointer to the first element of the array with the given key_id
+    // for bool arrays, note that they are always stored as int8 on all platforms (usually this makes no difference)
+    LM_GGML_API const void * lm_gguf_get_arr_data(const struct lm_gguf_context * ctx, int64_t key_id);
+
+    // get ith C string from array with given key_id
+    LM_GGML_API const char * lm_gguf_get_arr_str (const struct lm_gguf_context * ctx, int64_t key_id, size_t i);
+
+    LM_GGML_API int64_t        lm_gguf_get_n_tensors    (const struct lm_gguf_context * ctx);
+    LM_GGML_API int64_t        lm_gguf_find_tensor      (const struct lm_gguf_context * ctx, const char * name); // returns -1 if the tensor is not found
+    LM_GGML_API size_t         lm_gguf_get_tensor_offset(const struct lm_gguf_context * ctx, int64_t tensor_id);
+    LM_GGML_API const char *   lm_gguf_get_tensor_name  (const struct lm_gguf_context * ctx, int64_t tensor_id);
+    LM_GGML_API enum lm_ggml_type lm_gguf_get_tensor_type  (const struct lm_gguf_context * ctx, int64_t tensor_id);
+    LM_GGML_API size_t         lm_gguf_get_tensor_size  (const struct lm_gguf_context * ctx, int64_t tensor_id);
+
+    // removes key if it exists, returns id that the key had prior to removal (-1 if it didn't exist)
+    LM_GGML_API int64_t lm_gguf_remove_key(struct lm_gguf_context * ctx, const char * key);
+
+    // overrides an existing KV pair or adds a new one, the new KV pair is always at the back
+    LM_GGML_API void lm_gguf_set_val_u8  (struct lm_gguf_context * ctx, const char * key, uint8_t      val);
+    LM_GGML_API void lm_gguf_set_val_i8  (struct lm_gguf_context * ctx, const char * key, int8_t       val);
+    LM_GGML_API void lm_gguf_set_val_u16 (struct lm_gguf_context * ctx, const char * key, uint16_t     val);
+    LM_GGML_API void lm_gguf_set_val_i16 (struct lm_gguf_context * ctx, const char * key, int16_t      val);
+    LM_GGML_API void lm_gguf_set_val_u32 (struct lm_gguf_context * ctx, const char * key, uint32_t     val);
+    LM_GGML_API void lm_gguf_set_val_i32 (struct lm_gguf_context * ctx, const char * key, int32_t      val);
+    LM_GGML_API void lm_gguf_set_val_f32 (struct lm_gguf_context * ctx, const char * key, float        val);
+    LM_GGML_API void lm_gguf_set_val_u64 (struct lm_gguf_context * ctx, const char * key, uint64_t     val);
+    LM_GGML_API void lm_gguf_set_val_i64 (struct lm_gguf_context * ctx, const char * key, int64_t      val);
+    LM_GGML_API void lm_gguf_set_val_f64 (struct lm_gguf_context * ctx, const char * key, double       val);
+    LM_GGML_API void lm_gguf_set_val_bool(struct lm_gguf_context * ctx, const char * key, bool         val);
+    LM_GGML_API void lm_gguf_set_val_str (struct lm_gguf_context * ctx, const char * key, const char * val);
+
+    // creates a new array with n elements of the given type and copies the corresponding number of bytes from data
+    LM_GGML_API void lm_gguf_set_arr_data(struct lm_gguf_context * ctx, const char * key, enum lm_gguf_type type, const void * data, size_t n);
+
+    // creates a new array with n strings and copies the corresponding strings from data
+    LM_GGML_API void lm_gguf_set_arr_str (struct lm_gguf_context * ctx, const char * key, const char ** data, size_t n);
+
+    // set or add KV pairs from another context
+    LM_GGML_API void lm_gguf_set_kv(struct lm_gguf_context * ctx, const struct lm_gguf_context * src);
+
+    // add tensor to GGUF context, tensor name must be unique
+    LM_GGML_API void lm_gguf_add_tensor(struct lm_gguf_context * ctx, const struct lm_ggml_tensor * tensor);
+
+    // after changing a tensor's type, the offsets of all tensors with higher indices are immediately recalculated
+    //   in such a way that the tensor data remains as one contiguous block (except for padding)
+    LM_GGML_API void lm_gguf_set_tensor_type(struct lm_gguf_context * ctx, const char * name, enum lm_ggml_type type);
+
+    // assumes that at least lm_gguf_get_tensor_size bytes can be read from data
+    LM_GGML_API void lm_gguf_set_tensor_data(struct lm_gguf_context * ctx, const char * name, const void * data);
+
+    // writing gguf files can be done in 3 ways:
+    //
+    // - write the entire lm_gguf_context to a binary file in a single pass:
+    //
+    //   lm_gguf_write_to_file(ctx, fname, /*only_meta =*/ false);
+    //
+    // - write only the meta data to a file, then re-open the file and append the tensor data:
+    //
+    //   lm_gguf_write_to_file(ctx, fname, /*only_meta =*/ true);
+    //   FILE * f = fopen(fname, "ab");
+    //   fwrite(f, ...); // write tensor data
+    //   fclose(f);
+    //
+    // - 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");
+    //   const size_t size_meta = lm_gguf_get_meta_size(ctx);
+    //   fseek(f, size_meta, SEEK_SET);
+    //   fwrite(f, ...); // write tensor data
+    //   void * data = malloc(size_meta);
+    //   lm_gguf_get_meta_data(ctx, data);
+    //   rewind(f);
+    //   fwrite(data, 1, data, f);
+    //   free(data);
+    //   fclose(f);
+    //
+
+    // write the entire context to a binary file
+    LM_GGML_API bool lm_gguf_write_to_file(const struct lm_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
+    LM_GGML_API size_t lm_gguf_get_meta_size(const struct lm_gguf_context * ctx);
+
+    // writes the meta data to pointer "data"
+    LM_GGML_API void   lm_gguf_get_meta_data(const struct lm_gguf_context * ctx, void * data);
+
+#ifdef  __cplusplus
+}
+#endif

package/cpp/json-partial.cpp

@@ -0,0 +1,256 @@
+#include "json-partial.h"
+
+#include "log.h"
+
+#include "nlohmann/json.hpp"
+
+#include <string>
+
+using json = nlohmann::ordered_json;
+
+enum common_json_stack_element_type {
+    COMMON_JSON_STACK_ELEMENT_OBJECT,
+    COMMON_JSON_STACK_ELEMENT_KEY,
+    COMMON_JSON_STACK_ELEMENT_ARRAY,
+};
+
+struct common_json_stack_element {
+    common_json_stack_element_type type;
+    std::string key;
+};
+
+bool common_json_parse(
+    const std::string & input,
+    const std::string & healing_marker,
+    common_json & out)
+{
+    std::string::const_iterator it = input.begin();
+    const auto end = input.end();
+    return common_json_parse(it, end, healing_marker, out);
+}
+
+bool common_json_parse(
+    std::string::const_iterator & it,
+    const std::string::const_iterator & end,
+    const std::string & healing_marker,
+    common_json & out)
+{
+    // // https://json.nlohmann.me/features/parsing/sax_interface/
+    struct json_error_locator : public nlohmann::json_sax<json> {
+        std::size_t position;
+        bool found_error;
+        std::string last_token;
+        std::string exception_message;
+        std::vector<common_json_stack_element> stack;
+
+        json_error_locator() : position(0), found_error(false) {}
+
+        bool parse_error(std::size_t position, const std::string & last_token, const json::exception & ex) override { // NOLINT
+            this->position = position - 1;
+            this->found_error = true;
+            this->last_token = last_token;
+            this->exception_message = ex.what();
+            return false;
+        }
+        void close_value() {
+            if (!stack.empty() && (stack.back().type == COMMON_JSON_STACK_ELEMENT_KEY)) {
+                stack.pop_back();
+            }
+        }
+        bool null() override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool boolean(bool) override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool number_integer(number_integer_t) override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool number_unsigned(number_unsigned_t) override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool number_float(number_float_t, const string_t &) override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool string(string_t &) override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool binary(binary_t &) override { // NOLINT
+            close_value();
+            return true;
+        }
+        bool start_object(std::size_t) override { // NOLINT
+            stack.push_back({COMMON_JSON_STACK_ELEMENT_OBJECT, ""});
+            return true;
+        }
+        bool end_object() override {
+            LM_GGML_ASSERT(!stack.empty() && stack.back().type == COMMON_JSON_STACK_ELEMENT_OBJECT);
+            stack.pop_back();
+            close_value();
+            return true;
+        }
+        bool key(string_t & key) override { // NOLINT
+            stack.push_back({COMMON_JSON_STACK_ELEMENT_KEY, key});
+            return true;
+        }
+        bool start_array(std::size_t) override { // NOLINT
+            stack.push_back({COMMON_JSON_STACK_ELEMENT_ARRAY, ""});
+            return true;
+        }
+        bool end_array() override {
+            LM_GGML_ASSERT(!stack.empty() && stack.back().type == COMMON_JSON_STACK_ELEMENT_ARRAY);
+            stack.pop_back();
+            close_value();
+            return true;
+        }
+    };
+    json_error_locator err_loc;
+    auto start = it;
+    json::sax_parse(it, end, &err_loc);
+
+    if (err_loc.found_error) {
+        it = start;
+        auto temptative_end = it + err_loc.position;
+        // LOG_DBG("Error at position %zu (is_end = %s): %s\n", err_loc.position, temptative_end == end ? "true" : "false", err_loc.exception_message.c_str());
+
+        auto input = std::string(it, temptative_end);
+        try {
+            out.json = json::parse(input);
+            // out.json = json::parse(it, temptative_end);
+            it = temptative_end;
+            return true;
+        } catch (const std::exception & ex) {
+            // No, needs healing.
+            LOG_DBG("Failed to parse up to error: %s: <<<%s>>>\n", ex.what(), std::string(it, temptative_end).c_str());
+        }
+        auto can_parse = [](const std::string & str) {
+            try {
+                auto _ = json::parse(str); // NOLINT
+                return true;
+            } catch (const std::exception &) {
+                return false;
+            }
+        };
+        if (!healing_marker.empty() && !err_loc.stack.empty()) {
+            std::string str(it, temptative_end);
+            auto last_non_sp_pos = str.find_last_not_of(" \n\r\t");
+            if (last_non_sp_pos == std::string::npos) {
+                throw std::runtime_error("Cannot heal a truncated JSON that stopped in an unknown location");
+            }
+            auto last_non_sp_char = str[last_non_sp_pos];
+            // Used to detect stops on a number, which may not be complete.
+            auto was_maybe_number = [&]() {
+                if (!str.empty() && std::isspace(str.back())) {
+                    return false;
+                }
+                return std::isdigit(last_non_sp_char) ||
+                    last_non_sp_char == '.' ||
+                    last_non_sp_char == 'e' ||
+                    last_non_sp_char == 'E' ||
+                    last_non_sp_char == '-';
+            };
+
+            std::string closing;
+            for (size_t i = err_loc.stack.size(); i > 0; i--) {
+                auto & el = err_loc.stack[i - 1];
+                if (el.type == COMMON_JSON_STACK_ELEMENT_OBJECT) {
+                    closing += "}";
+                } else if (el.type == COMMON_JSON_STACK_ELEMENT_ARRAY) {
+                    closing += "]";
+                } else if (el.type != COMMON_JSON_STACK_ELEMENT_KEY) {
+                    throw std::runtime_error("Unexpected stack element type");
+                }
+            }
+
+            const auto & magic_seed = out.healing_marker.marker = healing_marker;//"$llama.cpp.json$";
+
+            if (err_loc.stack.back().type == COMMON_JSON_STACK_ELEMENT_KEY) {
+                // We're inside an object value
+                if (last_non_sp_char == ':' && can_parse(str + "1" + closing)) {
+                    // Was about to create an object value
+                    str += (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\"" + closing;
+                } else if (can_parse(str + ": 1" + closing)) {
+                    str += (out.healing_marker.json_dump_marker = ":\"" + magic_seed) + "\"" + closing;
+                } else if (last_non_sp_char == '{' && can_parse(str + closing)) {
+                    // Was about to create an object
+                    str += (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\": 1" + closing;
+                } else if (can_parse(str + "\"" + closing)) {
+                    // Was inside an object value string
+                    str += (out.healing_marker.json_dump_marker = magic_seed) + "\"" + closing;
+                } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\"" + closing)) {
+                    // Was inside an object value string after an escape
+                    str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\"" + closing;
+                } else {
+                    // find last :
+                    auto last_pos = str.find_last_of(':');
+                    if (last_pos == std::string::npos) {
+                        throw std::runtime_error("Cannot heal a truncated JSON that stopped in an unknown location");
+                    }
+                    // Cutting back to opening : for object value
+                    str = str.substr(0, last_pos + 1) + (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\"" + closing;
+                }
+            } else if (err_loc.stack.back().type == COMMON_JSON_STACK_ELEMENT_ARRAY) {
+                if ((last_non_sp_char == ',' || last_non_sp_char == '[') && can_parse(str + "1" + closing)) {
+                    // Was about to create an array value
+                    str += (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\"" + closing;
+                } else if (can_parse(str + "\"" + closing)) {
+                    // Was inside an array value string
+                    str += (out.healing_marker.json_dump_marker = magic_seed) + "\"" + closing;
+                } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\"" + closing)) {
+                    // Was inside an array value string after an escape
+                    str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\"" + closing;
+                } else if (!was_maybe_number() && can_parse(str + ", 1" + closing)) {
+                    // Had just finished a value
+                    str += (out.healing_marker.json_dump_marker = ",\"" + magic_seed) + "\"" + closing;
+                } else {
+                    auto last_pos = str.find_last_of("[,");
+                    if (last_pos == std::string::npos) {
+                        throw std::runtime_error("Cannot heal a truncated JSON array stopped in an unknown location");
+                    }
+                    // Cutting back to last [ or , for array value
+                    str = str.substr(0, last_pos + 1) + (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\"" + closing;
+                }
+            } else if (err_loc.stack.back().type == COMMON_JSON_STACK_ELEMENT_OBJECT) {
+                if ((last_non_sp_char == '{' && can_parse(str + closing)) ||
+                        (last_non_sp_char == ',' && can_parse(str + "\"\": 1" + closing))) {
+                    // Was about to create an object key+value
+                    str += (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\": 1" + closing;
+                } else if (!was_maybe_number() && can_parse(str + ",\"\": 1" + closing)) {
+                    // Was about to create an object key+value
+                    str += (out.healing_marker.json_dump_marker = ",\"" + magic_seed) + "\": 1" + closing;
+                } else if (can_parse(str + "\": 1" + closing)) {
+                    // Was inside an object key string
+                    str += (out.healing_marker.json_dump_marker = magic_seed) + "\": 1" + closing;
+                } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\": 1" + closing)) {
+                    // Was inside an object key string after an escape
+                    str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\": 1" + closing;
+                } else {
+                    auto last_pos = str.find_last_of(':');
+                    if (last_pos == std::string::npos) {
+                        throw std::runtime_error("Cannot heal a truncated JSON object stopped in an unknown location");
+                    }
+                    // fprintf(stderr, "Cutting back to last : for object key+value\n");
+                    str = str.substr(0, last_pos + 1) + (out.healing_marker.json_dump_marker = "\"" + magic_seed) + "\"" + closing;
+                }
+            } else {
+                throw std::runtime_error("Cannot heal a truncated JSON object stopped in an unknown location");
+            }
+            // fprintf(stderr, "HEALED:\nSTRING <<<\n%s\n>>>\n\nmagic_cut: <<<\n%s\n>>>\n\n", str.c_str(), out.healing_marker.json_dump_marker.c_str());
+            out.json = json::parse(str);
+            it = temptative_end;
+            return true;
+        }
+        // TODO: handle unclosed top-level primitive if the stack was empty but we got an error (e.g. "tru", "\"", etc...)
+        // fprintf(stderr, "Closing: TODO\n");
+        return false;
+    }
+    out.json = json::parse(it, end);
+    it = end;
+    return true;
+}

package/cpp/json-partial.h

@@ -0,0 +1,38 @@
+#pragma once
+
+#include "nlohmann/json.hpp"
+
+// Healing marker (empty if the JSON was fully parsed / wasn't healed).
+struct common_healing_marker {
+    // Raw marker.
+    std::string marker;
+
+    // Cutting the `common_json.json.dump()` string at the (only) occurrence of this marker should yield the original partial JSON string (modulo spaces / if it had the same dump format).
+    std::string json_dump_marker;
+};
+
+// Represents a parsed JSON object, with its optional healing marker (a JSON dump fragment that can be used to find the position of healing in the JSON dump string)
+struct common_json {
+    nlohmann::ordered_json json;
+
+    common_healing_marker healing_marker;
+};
+
+// Parse the JSON string, healing (closing) any partial JSON if `healing_marker` is not empty.
+//
+// Healing completes partial JSON strings by adding a (possibly modified) healing marker, then whatever is needed to close the JSON.
+// This allows to parse the resulting healed JSON string, yet be able to cut it again if needed at the healing marker.
+// (this is used when parsing JSON outputs from the models, then crafting partial JSONs for the partial tool calls in OAI format).
+//
+// For instance, parsing `{` with a healing marker `foo` will produce a healed JSON `{"foo":1}`, w/ json_dump_marker = `"foo"` (which can be used to break the JSON again).
+bool common_json_parse(
+    const std::string & input,
+    const std::string & healing_marker,
+    common_json & out);
+
+// Parse the JSON string (see overload above), but advancing an iterator to the end of the input when the (potentially partial) parsing succeeds.
+bool common_json_parse(
+    std::string::const_iterator & it,
+    const std::string::const_iterator & end,
+    const std::string & healing_marker,
+    common_json & out);