ggml-org · ggerganov · May 30, 2025 · May 29, 2025 · May 29, 2025 · May 29, 2025
diff --git a/.editorconfig b/.editorconfig
@@ -49,6 +49,6 @@ charset = unset
 trim_trailing_whitespace = unset
 insert_final_newline = unset
 
-[tools/mtmd/vendor/miniaudio.h]
+[vendor/miniaudio/miniaudio.h]
 trim_trailing_whitespace = unset
 insert_final_newline = unset
diff --git a/common/CMakeLists.txt b/common/CMakeLists.txt
@@ -58,23 +58,20 @@ add_library(${TARGET} STATIC
     arg.cpp
     arg.h
     base64.hpp
-    chat.cpp
-    chat.h
     chat-parser.cpp
     chat-parser.h
+    chat.cpp
+    chat.h
     common.cpp
     common.h
     console.cpp
     console.h
-    json-schema-to-grammar.cpp
-    json.hpp
-    json-partial.h
     json-partial.cpp
+    json-partial.h
+    json-schema-to-grammar.cpp
     llguidance.cpp
     log.cpp
     log.h
-    minja/chat-template.hpp
-    minja/minja.hpp
     ngram-cache.cpp
     ngram-cache.h
     regex-partial.cpp
@@ -147,7 +144,7 @@ if (LLAMA_LLGUIDANCE)
     set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance ${LLGUIDANCE_PLATFORM_LIBS})
 endif ()
 
-target_include_directories(${TARGET} PUBLIC .)
+target_include_directories(${TARGET} PUBLIC . ../vendor)
 target_compile_features   (${TARGET} PUBLIC cxx_std_17)
 target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
 

diff --git a/common/arg.cpp b/common/arg.cpp
@@ -1,10 +1,11 @@
-#include "gguf.h" // for reading GGUF splits
 #include "arg.h"
 
+#include "chat.h"
 #include "common.h"
+#include "gguf.h" // for reading GGUF splits
+#include "json-schema-to-grammar.h"
 #include "log.h"
 #include "sampling.h"
-#include "chat.h"
 
 // fix problem with std::min and std::max
 #if defined(_WIN32)
@@ -15,6 +16,9 @@
 #include <windows.h>
 #endif
 
+#define JSON_ASSERT GGML_ASSERT
+#include <nlohmann/json.hpp>
+
 #include <algorithm>
 #include <climits>
 #include <cstdarg>
@@ -34,8 +38,6 @@
 #include <future>
 #endif
 
-#include "json-schema-to-grammar.h"
-
 using json = nlohmann::ordered_json;
 
 std::initializer_list<enum llama_example> mmproj_examples = {

diff --git a/common/chat-parser.h b/common/chat-parser.h
@@ -2,9 +2,10 @@
 
 #include "chat.h"
 #include "json-partial.h"
-#include "json.hpp"
 #include "regex-partial.h"
 
+#include <nlohmann/json.hpp>
+
 #include <optional>
 #include <string>
 #include <vector>

diff --git a/common/chat.cpp b/common/chat.cpp
@@ -1,13 +1,14 @@
 #include "chat.h"
 #include "chat-parser.h"
 #include "common.h"
+#include "json-partial.h"
 #include "json-schema-to-grammar.h"
 #include "log.h"
-#include "json-partial.h"
-#include "minja/chat-template.hpp"
-#include "minja/minja.hpp"
 #include "regex-partial.h"
 
+#include <minja/chat-template.hpp>
+#include <minja/minja.hpp>
+
 #include <cstdio>
 #include <exception>
 #include <iostream>
@@ -16,7 +17,6 @@
 #include <string>
 #include <vector>
 
-
 static std::string format_time(const std::chrono::system_clock::time_point & now, const std::string & format) {
     auto time = std::chrono::system_clock::to_time_t(now);
     auto local_time = *std::localtime(&time);

diff --git a/common/json-partial.cpp b/common/json-partial.cpp
@@ -1,9 +1,10 @@
-#include <json-partial.h>
-#include "ggml.h"
+#include "json-partial.h"
+
 #include "log.h"
-#include <string>
 
-#include <json.hpp>
+#include <nlohmann/json.hpp>
+
+#include <string>
 
 using json = nlohmann::ordered_json;
 

diff --git a/common/json-partial.h b/common/json-partial.h
@@ -1,5 +1,6 @@
 #pragma once
-#include <json.hpp>
+
+#include <nlohmann/json.hpp>
 
 // Healing marker (empty if the JSON was fully parsed / wasn't healed).
 struct common_healing_marker {

diff --git a/common/json-schema-to-grammar.cpp b/common/json-schema-to-grammar.cpp
@@ -1,8 +1,9 @@
 #include "json-schema-to-grammar.h"
 #include "common.h"
 
+#include <nlohmann/json.hpp>
+
 #include <algorithm>
-#include <fstream>
 #include <map>
 #include <regex>
 #include <sstream>

diff --git a/common/json-schema-to-grammar.h b/common/json-schema-to-grammar.h
@@ -1,9 +1,9 @@
 #pragma once
 
-#include "ggml.h"
-// Change JSON_ASSERT from assert() to GGML_ASSERT:
-#define JSON_ASSERT GGML_ASSERT
-#include "json.hpp"
+#include <nlohmann/json_fwd.hpp>
+
+#include <functional>
+#include <string>
 
 std::string json_schema_to_grammar(const nlohmann::ordered_json & schema,
                                    bool force_gbnf = false);

diff --git a/scripts/sync_vendor.py b/scripts/sync_vendor.py
@@ -0,0 +1,22 @@
+#!/usr/bin/env python3
+
+import urllib.request
+
+vendor = {
+    "https://github.com/nlohmann/json/releases/latest/download/json.hpp":     "vendor/nlohmann/json.hpp",
+    "https://github.com/nlohmann/json/releases/latest/download/json_fwd.hpp": "vendor/nlohmann/json_fwd.hpp",
+
+    # sync manually
+    # "https://raw.githubusercontent.com/ochafik/minja/refs/heads/main/include/minja/minja.hpp":         "vendor/minja/minja.hpp",
+    # "https://raw.githubusercontent.com/ochafik/minja/refs/heads/main/include/minja/chat-template.hpp": "vendor/minja/chat-template.hpp",
+
+    "https://raw.githubusercontent.com/nothings/stb/refs/heads/master/stb_image.h": "vendor/stb/stb_image.h",
+
+    "https://github.com/mackron/miniaudio/raw/refs/tags/0.11.22/miniaudio.h": "vendor/miniaudio/miniaudio.h",
+
+    "https://raw.githubusercontent.com/yhirose/cpp-httplib/refs/tags/v0.20.1/httplib.h": "vendor/cpp-httplib/httplib.h",
+}
+
+for url, filename in vendor.items():
+    print(f"downloading {url} to {filename}") # noqa: NP100
+    urllib.request.urlretrieve(url, filename)
diff --git a/tests/test-chat.cpp b/tests/test-chat.cpp
@@ -5,16 +5,17 @@
 //
 //    cmake -B build && cmake --build build --parallel && ./build/bin/test-chat ../minja/build/tests/*.jinja 2>/dev/null
 //
-#include <fstream>
-#include <iostream>
-#include <json.hpp>
-#include <string>
-
 #include "chat.h"
 
 #include "../src/unicode.h"
 #include "../src/llama-grammar.h"
 
+#include <nlohmann/json.hpp>
+
+#include <fstream>
+#include <iostream>
+#include <string>
+
 using json = nlohmann::ordered_json;
 
 static std::ostream & operator<<(std::ostream & os, const common_chat_msg_diff & diff) {

diff --git a/tests/test-grammar-integration.cpp b/tests/test-grammar-integration.cpp
@@ -7,6 +7,8 @@
 #include "../src/unicode.h"
 #include "../src/llama-grammar.h"
 
+#include <nlohmann/json.hpp>
+
 #include <cassert>
 #include <string>
 #include <vector>

diff --git a/tests/test-json-schema-to-grammar.cpp b/tests/test-json-schema-to-grammar.cpp
@@ -6,6 +6,8 @@
 
 #include "../src/llama-grammar.h"
 
+#include <nlohmann/json.hpp>
+
 #include <cassert>
 #include <fstream>
 #include <sstream>

diff --git a/tools/mtmd/CMakeLists.txt b/tools/mtmd/CMakeLists.txt
@@ -23,8 +23,7 @@ add_library(mtmd_helper OBJECT
 
 target_link_libraries(mtmd_helper PRIVATE ggml llama mtmd ${CMAKE_THREAD_LIBS_INIT})
 target_include_directories(mtmd_helper PUBLIC .)
-target_include_directories(mtmd_helper PRIVATE ./vendor)
-target_include_directories(mtmd_helper PRIVATE ../..)
+target_include_directories(mtmd_helper PRIVATE ../../vendor)
 target_compile_features(mtmd_helper PRIVATE cxx_std_17)
 
 if (BUILD_SHARED_LIBS)

diff --git a/tools/mtmd/mtmd-helper.cpp b/tools/mtmd/mtmd-helper.cpp
@@ -27,10 +27,10 @@
 #define MA_NO_ENGINE
 #define MA_NO_GENERATION
 #define MA_API static
-#include "vendor/miniaudio.h"
+#include "miniaudio/miniaudio.h"
 
 #define STB_IMAGE_IMPLEMENTATION
-#include "vendor/stb_image.h"
+#include "stb/stb_image.h"
 
 #define LOG_INF(...) fprintf(stdout, __VA_ARGS__)
 #define LOG_ERR(...) fprintf(stderr, __VA_ARGS__)

diff --git a/tools/run/run.cpp b/tools/run/run.cpp
@@ -1,3 +1,13 @@
+#include "chat.h"
+#include "common.h"
+#include "llama-cpp.h"
+#include "log.h"
+
+#include "linenoise.cpp/linenoise.h"
+
+#define JSON_ASSERT GGML_ASSERT
+#include <nlohmann/json.hpp>
+
 #if defined(_WIN32)
 #    include <windows.h>
 #    include <io.h>
@@ -24,13 +34,6 @@
 #include <string>
 #include <vector>
 
-#include "chat.h"
-#include "common.h"
-#include "json.hpp"
-#include "linenoise.cpp/linenoise.h"
-#include "llama-cpp.h"
-#include "log.h"
-
 #if defined(__unix__) || (defined(__APPLE__) && defined(__MACH__)) || defined(_WIN32)
 [[noreturn]] static void sigint_handler(int) {
     printf("\n" LOG_COL_DEFAULT);

@@ -12,7 +12,6 @@ endif()
 set(TARGET_SRCS
     server.cpp
     utils.hpp
-    httplib.h
 )
 set(PUBLIC_ASSETS
     index.html.gz

@@ -11,9 +11,6 @@
 #include "mtmd.h"
 #include "mtmd-helper.h"
 
-// Change JSON_ASSERT from assert() to GGML_ASSERT:
-#define JSON_ASSERT GGML_ASSERT
-#include "json.hpp"
 // mime type for sending response
 #define MIMETYPE_JSON "application/json; charset=utf-8"
 

@@ -7,17 +7,16 @@
 #include "base64.hpp"
 #include "mtmd.h"
 #include "mtmd-helper.h"
+#include "chat.h"
 
 // increase max payload length to allow use of larger context size
 #define CPPHTTPLIB_FORM_URL_ENCODED_PAYLOAD_MAX_LENGTH 1048576
 // disable Nagle's algorithm
 #define CPPHTTPLIB_TCP_NODELAY true
-#include "httplib.h"
+#include <cpp-httplib/httplib.h>
 
-// Change JSON_ASSERT from assert() to GGML_ASSERT:
 #define JSON_ASSERT GGML_ASSERT
-#include "json.hpp"
-#include "chat.h"
+#include <nlohmann/json.hpp>
 
 #include <random>
 #include <sstream>

diff --git a/tools/tts/tts.cpp b/tools/tts/tts.cpp
@@ -5,7 +5,9 @@
 #include "sampling.h"
 #include "log.h"
 #include "llama.h"
-#include "json.hpp"
+
+#define JSON_ASSERT GGML_ASSERT
+#include <nlohmann/json.hpp>
 
 #include <algorithm>
 #include <cmath>

diff --git a/tools/server/httplib.h → vendor/cpp-httplib/httplib.h b/tools/server/httplib.h → vendor/cpp-httplib/httplib.h
@@ -8,7 +8,7 @@
 #ifndef CPPHTTPLIB_HTTPLIB_H
 #define CPPHTTPLIB_HTTPLIB_H
 
-#define CPPHTTPLIB_VERSION "0.20.0"
+#define CPPHTTPLIB_VERSION "0.20.1"
 
 /*
  * Configuration
@@ -145,6 +145,10 @@
 #define CPPHTTPLIB_LISTEN_BACKLOG 5
 #endif
 
+#ifndef CPPHTTPLIB_MAX_LINE_LENGTH
+#define CPPHTTPLIB_MAX_LINE_LENGTH 32768
+#endif
+
 /*
  * Headers
  */
@@ -3067,6 +3071,11 @@ inline bool stream_line_reader::getline() {
 #endif
 
   for (size_t i = 0;; i++) {
+    if (size() >= CPPHTTPLIB_MAX_LINE_LENGTH) {
+      // Treat exceptionally long lines as an error to
+      // prevent infinite loops/memory exhaustion
+      return false;
+    }
     char byte;
     auto n = strm_.read(&byte, 1);
 
@@ -6055,6 +6064,8 @@ inline void calc_actual_timeout(time_t max_timeout_msec, time_t duration_msec,
   auto actual_timeout_msec =
       (std::min)(max_timeout_msec - duration_msec, timeout_msec);
 
+  if (actual_timeout_msec < 0) { actual_timeout_msec = 0; }
+
   actual_timeout_sec = actual_timeout_msec / 1000;
   actual_timeout_usec = (actual_timeout_msec % 1000) * 1000;
 }
@@ -7327,8 +7338,9 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
   }
 
   // Setup `is_connection_closed` method
-  req.is_connection_closed = [&]() {
-    return !detail::is_socket_alive(strm.socket());
+  auto sock = strm.socket();
+  req.is_connection_closed = [sock]() {
+    return !detail::is_socket_alive(sock);
   };
 
   // Routing

diff --git a/tools/mtmd/vendor/miniaudio.h → vendor/miniaudio/miniaudio.h b/tools/mtmd/vendor/miniaudio.h → vendor/miniaudio/miniaudio.h
diff --git a/common/minja/chat-template.hpp → vendor/minja/chat-template.hpp b/common/minja/chat-template.hpp → vendor/minja/chat-template.hpp
@@ -22,7 +22,7 @@
 #include <string>
 #include <vector>
 
-#include <json.hpp>
+#include <nlohmann/json.hpp>
 
 using json = nlohmann::ordered_json;
 

diff --git a/common/minja/minja.hpp → vendor/minja/minja.hpp b/common/minja/minja.hpp → vendor/minja/minja.hpp
@@ -29,7 +29,7 @@
 #include <utility>
 #include <vector>
 
-#include <json.hpp>
+#include <nlohmann/json.hpp>
 
 using json = nlohmann::ordered_json;