Add whisper.cpp (server) support to llamafile

Makefile

@@ -14,6 +14,7 @@ include build/rules.mk
 include llamafile/BUILD.mk
 include llama.cpp/BUILD.mk
 include stable-diffusion.cpp/BUILD.mk
+include whisper.cpp/BUILD.mk
 include double-conversion/BUILD.mk
 include stb/BUILD.mk
 
@@ -22,6 +23,8 @@ include stb/BUILD.mk
 .PHONY: o/$(MODE)/
 o/$(MODE)/:	o/$(MODE)/llamafile					\
 		o/$(MODE)/llama.cpp					\
+		o/$(MODE)/stable-diffusion.cpp					\
+		o/$(MODE)/whisper.cpp					\
 		o/$(MODE)/stb						\
 		o/$(MODE)/depend.test
 

llama.cpp/ggml-backend.c

@@ -1867,6 +1867,15 @@ int ggml_backend_sched_get_n_copies(ggml_backend_sched_t sched) {
     return sched->n_copies;
 }
 
+int ggml_backend_sched_get_n_backends(ggml_backend_sched_t sched) {
+    return sched->n_backends;
+}
+
+ggml_backend_t ggml_backend_sched_get_backend(ggml_backend_sched_t sched, int i) {
+    GGML_ASSERT(i >= 0 && i < sched->n_backends);
+    return sched->backends[i];
+}
+
 size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend) {
     int backend_index = ggml_backend_sched_backend_id(sched, backend);
     GGML_ASSERT(backend_index >= 0 && backend_index < sched->n_backends);

llama.cpp/ggml-backend.h

@@ -182,6 +182,9 @@ extern "C" {
     // Initialize backend buffers from a measure graph
     GGML_API bool                 ggml_backend_sched_reserve(ggml_backend_sched_t sched, struct ggml_cgraph * measure_graph);
 
+    GGML_API int                  ggml_backend_sched_get_n_backends(ggml_backend_sched_t sched);
+    GGML_API ggml_backend_t       ggml_backend_sched_get_backend(ggml_backend_sched_t sched, int i);
+
     // Get the number of splits of the last graph
     GGML_API int                  ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);
     GGML_API int                  ggml_backend_sched_get_n_copies(ggml_backend_sched_t sched);

llamafile/llamafile.c

@@ -294,7 +294,7 @@ struct llamafile *llamafile_open_gguf(const char *fname, const char *mode) {
         errno = EIO;
         return 0;
     }
-    if (ZIP_READ32(buf) == ZIP_READ32("GGUF")) {
+    if (ZIP_READ32(buf) == ZIP_READ32("GGUF") || ZIP_READ32(buf) == ZIP_READ32("ggml")) {
         errno = EINVAL;
         return file;
     }
@@ -312,6 +312,16 @@ size_t llamafile_size(struct llamafile *file) {
     return file->size;
 }
 
+size_t llamafile_position(struct llamafile *file) {
+    return file->position;
+}
+
+bool llamafile_eof(struct llamafile *file) {
+    if (file->fp)
+        return feof(file->fp);
+    return file->position >= file->size;
+}
+
 void *llamafile_content(struct llamafile *file) {
     return file->content;
 }

llamafile/llamafile.h

@@ -47,6 +47,8 @@ bool llamafile_seek(struct llamafile *, size_t, int);
 void *llamafile_content(struct llamafile *);
 size_t llamafile_tell(struct llamafile *);
 size_t llamafile_size(struct llamafile *);
+size_t llamafile_position(struct llamafile *);
+bool llamafile_eof(struct llamafile *file);
 FILE *llamafile_fp(struct llamafile *);
 void llamafile_ref(struct llamafile *);
 void llamafile_unref(struct llamafile *);

whisper.cpp/BUILD.mk

@@ -0,0 +1,40 @@
+#-*-mode:makefile-gmake;indent-tabs-mode:t;tab-width:8;coding:utf-8-*-┐
+#── vi: set noet ft=make ts=8 sw=8 fenc=utf-8 :vi ────────────────────┘
+
+PKGS += WHISPER_CPP
+
+WHISPER_CPP_FILES := $(wildcard whisper.cpp/*.*)
+WHISPER_CPP_HDRS = $(filter %.h,$(WHISPER_CPP_FILES))				\
+			    $(filter %.hpp,$(WHISPER_CPP_FILES))
+WHISPER_CPP_INCS = $(filter %.inc,$(WHISPER_CPP_FILES))
+WHISPER_CPP_SRCS_C = $(filter %.c,$(WHISPER_CPP_FILES))
+WHISPER_CPP_SRCS_CPP = $(filter %.cpp,$(WHISPER_CPP_FILES))
+WHISPER_CPP_SRCS = $(WHISPER_CPP_SRCS_C) $(WHISPER_CPP_SRCS_CPP)
+
+WHISPER_CPP_OBJS =								\
+	$(LLAMAFILE_OBJS)							\
+	$(WHISPER_CPP_SRCS_C:%.c=o/$(MODE)/%.o)		\
+	$(WHISPER_CPP_SRCS_CPP:%.cpp=o/$(MODE)/%.o)
+
+o/$(MODE)/whisper.cpp/whisper.cpp.a: $(WHISPER_CPP_OBJS)
+
+$(WHISPER_CPP_OBJS): private					\
+		CCFLAGS +=								\
+			-DGGML_MULTIPLATFORM
+
+$(WHISPER_CPP_OBJS): private					\
+		CXXFLAGS +=								\
+			-frtti								\
+			-Wno-alloc-size-larger-than			\
+			-Wno-deprecated-declarations
+
+o/$(MODE)/whisper.cpp/server:					\
+		o/$(MODE)/whisper.cpp/server.o			\
+		o/$(MODE)/whisper.cpp/whisper.cpp.a		\
+		o/$(MODE)/llama.cpp/llama.cpp.a	
+
+$(WHISPER_CPP_OBJS): whisper.cpp/BUILD.mk
+
+.PHONY: o/$(MODE)/whisper.cpp
+o/$(MODE)/whisper.cpp:							\
+		o/$(MODE)/whisper.cpp/server