fix includes with help from include-what-you-use

examples/baby-llama/baby-llama.cpp

@@ -1,11 +1,13 @@
 #include "ggml.h"
 #include "train.h"
 
-#include <vector>
+#include <algorithm>
 #include <cassert>
+#include <cmath>
+#include <cstdint>
+#include <cstdio>
 #include <cstdlib>
 #include <cstring>
-#include <random>
 #include <vector>
 
 #if defined(_MSC_VER)

examples/batched-bench/batched-bench.cpp

@@ -1,10 +1,11 @@
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
 
 #include <algorithm>
-#include <cmath>
+#include <cstdint>
 #include <cstdio>
-#include <string>
+#include <cstdlib>
 #include <vector>
 
 // mutates the input string

examples/batched/batched.cpp

@@ -1,9 +1,11 @@
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
 
 #include <algorithm>
-#include <cmath>
+#include <cstdint>
 #include <cstdio>
+#include <cstdlib>
 #include <string>
 #include <vector>
 

examples/beam-search/beam-search.cpp

@@ -1,29 +1,14 @@
 #include "common.h"
 #include "llama.h"
 
+#include <algorithm>
 #include <cassert>
-#include <cinttypes>
-#include <cmath>
 #include <cstdio>
-#include <cstring>
-#include <ctime>
-#include <fstream>
 #include <iostream>
 #include <string>
+#include <tuple>
 #include <vector>
 
-#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
-#include <signal.h>
-#include <unistd.h>
-#elif defined (_WIN32)
-#define WIN32_LEAN_AND_MEAN
-#ifndef NOMINMAX
-#   define NOMINMAX
-#endif
-#include <windows.h>
-#include <signal.h>
-#endif
-
 // Used for debugging to print out beam tokens.
 struct ostream_beam_view {
     llama_context * ctx;

examples/benchmark/benchmark-matmult.cpp

@@ -1,20 +1,11 @@
 #include "common.h"
 #include "ggml.h"
 
-#include <locale.h>
-#include <assert.h>
-#include <math.h>
-#include <cstring>
-#include <cstdio>
 #include <cinttypes>
-#include <unordered_map>
-#include <queue>
-#include <string.h>
-#include <cassert>
-#include <fstream>
+#include <cstdio>
+#include <cstdlib>
 #include <string>
-#include <iterator>
-#include <algorithm>
+#include <vector>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data

examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp

@@ -1,19 +1,22 @@
+#include "common.h"
 #include "ggml.h"
 #include "llama.h"
-#include "common.h"
 
-#include <unordered_map>
-#include <vector>
-#include <cassert>
-#include <climits>
-#include <cstring>
-#include <cstdarg>
-#include <ctime>
-#include <random>
-#include <stdexcept>
-#include <sstream>
 #include <algorithm>
+#include <cassert>
+#include <cerrno>
+#include <climits>
+#include <cmath>
+#include <cstdarg>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <sstream>
 #include <string>
+#include <unordered_map>
+#include <utility>
+#include <vector>
 
 // GGUF keys & tensor names.
 

examples/embedding/embedding.cpp

@@ -1,7 +1,12 @@
 #include "common.h"
 #include "llama.h"
 
+#include <algorithm>
+#include <cstdio>
 #include <ctime>
+#include <random>
+#include <tuple>
+#include <vector>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data

examples/export-lora/export-lora.cpp

@@ -1,11 +1,16 @@
-
 #include "common.h"
 #include "ggml.h"
 #include "ggml-alloc.h"
 
-#include <vector>
+#include <algorithm>
+#include <cerrno>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
 #include <string>
 #include <thread>
+#include <vector>
 
 static const size_t tensor_alignment = 32;
 

examples/finetune/finetune.cpp

@@ -1,19 +1,19 @@
-#include "ggml.h"
-#include "ggml-alloc.h"
-#include "llama.h"
 #include "common.h"
+#include "ggml-alloc.h"
+#include "ggml.h"
+#include "llama.h"
 #include "train.h"
-#include <unordered_map>
-#include <vector>
-#include <cassert>
-#include <climits>
-#include <cstring>
-#include <cstdarg>
-#include <ctime>
-#include <random>
-#include <stdexcept>
+
 #include <algorithm>
+#include <cerrno>
+#include <cmath>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <ctime>
 #include <string>
+#include <vector>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data

examples/infill/infill.cpp

@@ -1,19 +1,17 @@
 #include "common.h"
-
 #include "console.h"
 #include "llama.h"
-#include "grammar-parser.h"
+#include "sampling.h"
 
-#include <cassert>
-#include <cinttypes>
-#include <cmath>
+#include <algorithm>
 #include <cstdio>
-#include <cstring>
+#include <cstdlib>
 #include <ctime>
 #include <fstream>
-#include <iostream>
+#include <random>
 #include <sstream>
 #include <string>
+#include <tuple>
 #include <vector>
 
 #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))

examples/llama-bench/llama-bench.cpp

@@ -1,26 +1,30 @@
+#include "ggml.h"
+#include "llama.h"
+#include "common.h"
+#include "ggml-cuda.h"
+
 #include <algorithm>
 #include <array>
 #include <cassert>
+#include <cctype>
 #include <chrono>
 #include <cinttypes>
 #include <clocale>
 #include <cmath>
 #include <cstdio>
+#include <cstdlib>
 #include <cstring>
 #include <ctime>
 #include <iterator>
 #include <map>
+#include <memory>
 #include <numeric>
 #include <regex>
 #include <sstream>
 #include <string>
+#include <utility>
 #include <vector>
 
-#include "ggml.h"
-#include "llama.h"
-#include "common.h"
-#include "ggml-cuda.h"
-
 // utils
 static uint64_t get_time_ns() {
     using clock = std::chrono::high_resolution_clock;

examples/llava/clip.cpp

@@ -2,21 +2,23 @@
 // so there might be still unnecessary artifacts hanging around
 // I'll gradually clean and extend it
 
-#include <cassert>
+#include "clip.h"
+#include "ggml-alloc.h"
+#include "ggml.h"
+
+#include <algorithm>
+#include <climits>
 #include <cmath>
+#include <cstdarg>
+#include <cstdio>
 #include <cstdlib>
 #include <cstring>
 #include <fstream>
-#include <iostream>
-#include <map>
 #include <regex>
 #include <stdexcept>
+#include <string>
 #include <vector>
 
-#include "clip.h"
-#include "ggml.h"
-#include "ggml-alloc.h"
-
 #define STB_IMAGE_IMPLEMENTATION
 #include "stb_image.h"
 

examples/llava/clip.h

@@ -1,7 +1,8 @@
 #ifndef CLIP_H
 #define CLIP_H
 
-#include "ggml.h"
+#include <stddef.h>
+#include <stdint.h>
 
 struct clip_ctx;
 

examples/llava/llava.cpp

@@ -1,11 +1,14 @@
 #include "clip.h"
-#include "llava-utils.h"
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
+#include "llava-utils.h"
 
+#include <cstdint>
 #include <cstdio>
 #include <cstdlib>
-#include <vector>
+#include <cstring>
+#include <string>
 
 static void show_additional_info(int /*argc*/, char ** argv) {
     printf("\n example usage: %s -m <llava-v1.5-7b/ggml-model-q5_k.gguf> --mmproj <llava-v1.5-7b/mmproj-model-f16.gguf> --image <path/to/an/image.jpg> [--temp 0.1] [-p \"describe the image in detail.\"]\n", argv[0]);

examples/main/main.cpp

@@ -1,18 +1,17 @@
 #include "common.h"
-
 #include "console.h"
 #include "llama.h"
+#include "sampling.h"
 
-#include <cassert>
-#include <cinttypes>
-#include <cmath>
+#include <algorithm>
 #include <cstdio>
-#include <cstring>
+#include <cstdlib>
 #include <ctime>
 #include <fstream>
-#include <iostream>
+#include <random>
 #include <sstream>
 #include <string>
+#include <tuple>
 #include <vector>
 
 #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))

examples/parallel/parallel.cpp

@@ -2,13 +2,20 @@
 // The clients submite requests to the server and they are processed in parallel.
 
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
+#include "sampling.h"
 
-#include <cmath>
+#include <algorithm>
+#include <cctype>
+#include <cstdint>
 #include <cstdio>
-#include <string>
-#include <vector>
+#include <cstdlib>
 #include <ctime>
+#include <istream>
+#include <string>
+#include <tuple>
+#include <vector>
 
 // trim whitespace from the beginning and end of a string
 static std::string trim(const std::string & str) {

examples/perplexity/perplexity.cpp

@@ -1,13 +1,20 @@
 #include "common.h"
 #include "llama.h"
 
+#include <algorithm>
+#include <chrono>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
 #include <ctime>
-#include <sstream>
-#include <thread>
+#include <iterator>
 #include <mutex>
+#include <random>
+#include <sstream>
+#include <string>
+#include <thread>
+#include <tuple>
+#include <utility>
 #include <vector>
 
 #if defined(_MSC_VER)

examples/quantize-stats/quantize-stats.cpp

@@ -8,15 +8,16 @@
 #include <cinttypes>
 #include <cmath>
 #include <cstdio>
+#include <cstdlib>
 #include <cstring>
-#include <map>
+#include <iterator>
+#include <mutex>
 #include <numeric>
 #include <regex>
 #include <string>
-#include <unordered_map>
-#include <vector>
 #include <thread>
-#include <mutex>
+#include <utility>
+#include <vector>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data

examples/quantize/quantize.cpp

@@ -1,10 +1,14 @@
 #include "common.h"
 #include "llama.h"
 
+#include <cctype>
+#include <cstdint>
 #include <cstdio>
+#include <cstdlib>
 #include <cstring>
-#include <vector>
+#include <exception>
 #include <string>
+#include <vector>
 
 struct quant_option {
     std::string name;

examples/save-load-state/save-load-state.cpp

@@ -1,9 +1,11 @@
 #include "common.h"
 #include "llama.h"
 
-#include <vector>
+#include <cstdint>
 #include <cstdio>
-#include <chrono>
+#include <string>
+#include <tuple>
+#include <vector>
 
 int main(int argc, char ** argv) {
     gpt_params params;

examples/server/server.cpp

@@ -1,6 +1,7 @@
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
-#include "grammar-parser.h"
+#include "sampling.h"
 
 #include "../llava/clip.h"
 
@@ -20,10 +21,28 @@
 #include "completion.js.hpp"
 #include "json-schema-to-grammar.mjs.hpp"
 
-#include <cstddef>
-#include <thread>
-#include <mutex>
+#include <algorithm>
+#include <cctype>
 #include <chrono>
+#include <cmath>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <ctime>
+#include <exception>
+#include <functional>
+#include <istream>
+#include <iterator>
+#include <mutex>
+#include <numeric>
+#include <stdexcept>
+#include <string>
+#include <thread>
+#include <tuple>
+#include <unordered_map>
+#include <utility>
+#include <vector>
 
 #ifndef SERVER_VERBOSE
 #define SERVER_VERBOSE 1

examples/simple/simple.cpp

@@ -1,9 +1,8 @@
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
 
-#include <cmath>
 #include <cstdio>
-#include <string>
 #include <vector>
 
 int main(int argc, char ** argv) {

examples/speculative/speculative.cpp

@@ -1,9 +1,13 @@
 #include "common.h"
+#include "ggml.h"
 #include "llama.h"
+#include "sampling.h"
 
-#include <cmath>
+#include <algorithm>
 #include <cstdio>
+#include <cstring>
 #include <string>
+#include <tuple>
 #include <vector>
 
 #define SPEC_VOCAB_MAX_SIZE_DIFFERENCE  100

examples/train-text-from-scratch/train-text-from-scratch.cpp

@@ -1,19 +1,18 @@
-#include "ggml.h"
-#include "ggml-alloc.h"
 #include "common.h"
-#include "train.h"
+#include "ggml-alloc.h"
+#include "ggml.h"
 #include "llama.h"
-#include <unordered_map>
-#include <vector>
-#include <cassert>
-#include <climits>
-#include <cstring>
-#include <cstdarg>
-#include <ctime>
-#include <random>
-#include <stdexcept>
+#include "train.h"
+
 #include <algorithm>
+#include <cmath>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <ctime>
 #include <string>
+#include <vector>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data