feat(llama.cpp): upgrade and use libmtmd (#5379)

* WIP

* wip

* wip

* Make it compile

* Update json.hpp

* this shouldn't be private for now

* Add logs

* Reset auto detected template

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Re-enable grammars

* This seems to be broken - 360a9c98e1 (diff-a18a8e64e12a01167d8e98fc)[…]cccf0d4eed09d76d879L2998-L3207

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Placeholder

* Simplify image loading

* use completion type

* disable streaming

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* correctly return timings

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Remove some debug logging

* Adapt tests

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Keep header

* embedding: do not use oai type

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Sync from server.cpp

* Use utils and json directly from llama.cpp

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Sync with upstream

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* fix: copy json.hpp from the correct location

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* fix: add httplib

* sync llama.cpp

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Embeddiongs: set OAICOMPAT_TYPE_EMBEDDING

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* feat: sync with server.cpp by including it

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* make it darwin-compatible

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

backend/cpp/llama/CMakeLists.txt

@@ -1,17 +1,17 @@
 
 ## XXX: In some versions of CMake clip wasn't being built before llama.
 ## This is an hack for now, but it should be fixed in the future.
-set(TARGET myclip)
-add_library(${TARGET} clip.cpp clip.h clip-impl.h llava.cpp llava.h)
-install(TARGETS ${TARGET} LIBRARY)
-target_include_directories(myclip PUBLIC .)
-target_include_directories(myclip PUBLIC ../..)
-target_include_directories(myclip PUBLIC ../../common)
-target_link_libraries(${TARGET} PRIVATE common ggml llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
-if (NOT MSVC)
-    target_compile_options(${TARGET} PRIVATE -Wno-cast-qual) # stb_image.h
-endif()
+# set(TARGET myclip)
+# add_library(${TARGET} clip.cpp clip.h clip-impl.h llava.cpp llava.h)
+# install(TARGETS ${TARGET} LIBRARY)
+# target_include_directories(myclip PUBLIC .)
+# target_include_directories(myclip PUBLIC ../..)
+# target_include_directories(myclip PUBLIC ../../common)
+# target_link_libraries(${TARGET} PRIVATE common ggml llama ${CMAKE_THREAD_LIBS_INIT})
+# target_compile_features(${TARGET} PRIVATE cxx_std_11)
+# if (NOT MSVC)
+#     target_compile_options(${TARGET} PRIVATE -Wno-cast-qual) # stb_image.h
+# endif()
 # END CLIP hack
 
 
@@ -74,8 +74,12 @@ add_library(hw_grpc_proto
   ${hw_proto_srcs}
   ${hw_proto_hdrs} )
 
-add_executable(${TARGET} grpc-server.cpp utils.hpp json.hpp)
-target_link_libraries(${TARGET} PRIVATE common llama myclip ${CMAKE_THREAD_LIBS_INIT} absl::flags hw_grpc_proto
+add_executable(${TARGET} grpc-server.cpp utils.hpp json.hpp httplib.h)
+
+target_include_directories(${TARGET} PRIVATE ../llava)
+target_include_directories(${TARGET} PRIVATE ${CMAKE_SOURCE_DIR})
+
+target_link_libraries(${TARGET} PRIVATE common llama mtmd ${CMAKE_THREAD_LIBS_INIT} absl::flags hw_grpc_proto
   absl::flags_parse
   gRPC::${_REFLECTION}
   gRPC::${_GRPC_GRPCPP}