feat: Add support for new mtmd api, add Qwen2.5-VL chat handler

Author: abetlen

CMakeLists.txt

@@ -143,46 +143,45 @@ if (LLAMA_BUILD)
         )
     endif()
 
-    # if (LLAVA_BUILD)
-    #     if (LLAMA_CUBLAS OR LLAMA_CUDA)
-    #         add_compile_definitions(GGML_USE_CUBLAS)
-    #         add_compile_definitions(GGML_USE_CUDA)
-    #     endif()
-    #
-    #     if (LLAMA_METAL)
-    #         add_compile_definitions(GGML_USE_METAL)
-    #     endif()
-    #
-    #     # Building llava
-    #     add_subdirectory(vendor/llama.cpp/tools/mtmd)
-    #     set_target_properties(llava_shared PROPERTIES OUTPUT_NAME "llava")
-    #
-    #     if (WIN32)
-    #         set_target_properties(llava_shared PROPERTIES CUDA_ARCHITECTURES OFF)
-    #     endif()
-    #     llama_cpp_python_install_target(llava_shared)
-    #     if (WIN32)
-    #         install(
-    #             FILES $<TARGET_RUNTIME_DLLS:llava_shared>
-    #             DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp/lib
-    #         )
-    #         install(
-    #             FILES $<TARGET_RUNTIME_DLLS:llava_shared>
-    #             DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp/lib
-    #         )
-    #     endif()
-    #
-    #     # Fix for llava build: Add include directory for llama.h
-    #     # Move these commands after the add_subdirectory call
-    #     target_include_directories(llava PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
-    #     target_include_directories(llava PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/ggml/include)
-    #
-    #     if (BUILD_SHARED_LIBS)
-    #         target_include_directories(llava_shared PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
-    #         target_include_directories(llava_shared PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/ggml/include)
-    #     endif()
-    #
-    #     target_include_directories(llama-llava-cli PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
-    #     target_include_directories(llama-minicpmv-cli PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
-    # endif()
+    if (LLAVA_BUILD)
+        if (LLAMA_CUBLAS OR LLAMA_CUDA)
+            add_compile_definitions(GGML_USE_CUBLAS)
+            add_compile_definitions(GGML_USE_CUDA)
+        endif()
+
+        if (LLAMA_METAL)
+            add_compile_definitions(GGML_USE_METAL)
+        endif()
+
+        # Building llava
+        add_subdirectory(vendor/llama.cpp/tools/mtmd)
+
+        if (WIN32)
+            set_target_properties(mtmd PROPERTIES CUDA_ARCHITECTURES OFF)
+        endif()
+        llama_cpp_python_install_target(mtmd)
+        if (WIN32)
+            install(
+                FILES $<TARGET_RUNTIME_DLLS:mtmd>
+                DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp/lib
+            )
+            install(
+                FILES $<TARGET_RUNTIME_DLLS:mtmd>
+                DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp/lib
+            )
+        endif()
+
+        # Fix for mtmd build: Add include directory for llama.h
+        # Move these commands after the add_subdirectory call
+        target_include_directories(mtmd PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
+        target_include_directories(mtmd PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/ggml/include)
+
+        if (BUILD_SHARED_LIBS)
+            target_include_directories(mtmd PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
+            target_include_directories(mtmd PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/ggml/include)
+        endif()
+
+        # target_include_directories(llama-llava-cli PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
+        # target_include_directories(llama-minicpmv-cli PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/vendor/llama.cpp/include)
+    endif()
 endif()