Distribute and run LLMs with a single file.
C++ cross-platformggufllama-cpplocal-ailocal-inferencelocal-llmopen-source-aisingle-file-executablespeech-to-text
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
--- a/whisper.cpp/examples/server/server.cpp
+++ b/whisper.cpp/examples/server/server.cpp
@@ -23,6 +23,11 @@
#include <windows.h>
#endif
+#ifdef WHISPERFILE
+#include "whisperfile/slurp.h"
+#include "llamafile/llamafile.h"
+#endif
+
using namespace httplib;
using json = nlohmann::ordered_json;
@@ -261,6 +266,22 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
else if (arg == "-vmsd" || arg == "--vad-max-speech-duration-s") { params.vad_max_speech_duration_s = std::stof(argv[++i]); }
else if (arg == "-vp" || arg == "--vad-speech-pad-ms") { params.vad_speech_pad_ms = std::stoi(argv[++i]); }
else if (arg == "-vo" || arg == "--vad-samples-overlap") { params.vad_samples_overlap = std::stof(argv[++i]); }
+#ifdef WHISPERFILE
+ // llamafile flags
+ else if (arg == "--log-disable") { FLAG_log_disable = true; }
+ else if (arg == "-ng" || arg == "--no-gpu") { FLAG_gpu = LLAMAFILE_GPU_DISABLE; }
+ else if (arg == "--gpu") {
+ if (++i >= argc) {
+ fprintf(stderr, "error: missing --gpu flag value\n");
+ exit(1);
+ }
+ FLAG_gpu = llamafile_gpu_parse(argv[i]);
+ if (FLAG_gpu == LLAMAFILE_GPU_ERROR) {
+ fprintf(stderr, "error: invalid --gpu flag value: %s\n", argv[i]);
+ exit(1);
+ }
+ }
+#endif
else {
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
whisper_print_usage(argc, argv, params, sparams);
@@ -611,7 +632,8 @@ void get_req_parameters(const Request & req, whisper_params & params)
} // namespace
-int main(int argc, char ** argv) {
+// Core server logic - can be called from whisper-server main or standalone
+int whisper_server_main(int argc, char ** argv) {
ggml_backend_load_all();
whisper_params params;
@@ -818,6 +840,25 @@ int main(int argc, char ** argv) {
std::vector<float> pcmf32; // mono-channel F32 PCM
std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
+#ifdef WHISPERFILE
+ // whisperfile: use slurp_audio_file for WAV/MP3/FLAC/OGG support
+ {
+ // write incoming audio to temporary file
+ const std::string temp_filename = generate_temp_filename(sparams.tmp_dir, "whisper-server", "");
+ std::ofstream temp_file{temp_filename, std::ios::binary};
+ temp_file << audio_file.content;
+ temp_file.close();
+
+ bool ok = slurp_audio_file(temp_filename.c_str(), pcmf32, pcmf32s, params.diarize);
+ std::remove(temp_filename.c_str());
+ if (!ok) {
+ fprintf(stderr, "error: failed to read audio file\n");
+ const std::string error_resp = "{\"error\":\"failed to read audio file (supported: wav, mp3, flac, ogg)\"}";
+ res.set_content(error_resp, "application/json");
+ return;
+ }
+ }
+#else
if (sparams.ffmpeg_converter) {
// if file is not wav, convert to wav
// write to temporary file
@@ -853,6 +894,7 @@ int main(int argc, char ** argv) {
return;
}
}
+#endif
printf("Successfully loaded %s\n", filename.c_str());
@@ -1246,3 +1288,10 @@ int main(int argc, char ** argv) {
return 0;
}
+
+// Standalone entry point - only when NOT building as part of whisperfile
+#ifndef WHISPERFILE
+int main(int argc, char ** argv) {
+ return whisper_server_main(argc, argv);
+}
+#endif