llama.cpp: version bump b10969 (#18446)

llama.cpp build changes resulted in duplicate symbols between libllama and libmtmd.  This moves the compat patch into libllama with exported symbols.

LLAMA_CPP_VERSION

@@ -1 +1 @@
-b10864
+b10969

llama/compat/llama-ollama-compat.cpp

@@ -13,7 +13,6 @@
 #include <cerrno>
 #include <chrono>
 #include <cstdlib>
-#include <cstdarg>
 #include <cstdint>
 #include <cstdio>
 #include <cstring>
@@ -29,20 +28,8 @@ using namespace llama_ollama_compat::detail; // pull detail:: helpers into scope
 
 namespace {
 
-#ifdef OLLAMA_COMPAT_MTMD_BUILD
-void ollama_compat_log(const char * format, ...) {
-    std::va_list args;
-    va_start(args, format);
-    std::vfprintf(stderr, format, args);
-    va_end(args);
-}
-
-#define OLLAMA_COMPAT_LOG_INFO(...)  do { ollama_compat_log(__VA_ARGS__); } while (0)
-#define OLLAMA_COMPAT_LOG_ERROR(...) ollama_compat_log(__VA_ARGS__)
-#else
 #define OLLAMA_COMPAT_LOG_INFO(...)  do { LLAMA_LOG_INFO(__VA_ARGS__); } while (0)
 #define OLLAMA_COMPAT_LOG_ERROR(...) LLAMA_LOG_ERROR(__VA_ARGS__)
-#endif
 
 double elapsed_ms(std::chrono::steady_clock::time_point start) {
     return std::chrono::duration<double, std::milli>(std::chrono::steady_clock::now() - start).count();

llama/compat/llama-ollama-compat.h

@@ -26,6 +26,7 @@
 #include <string>
 
 #include "ggml-backend.h" // for ggml_backend_buffer_type_t
+#include "llama.h"        // for LLAMA_API
 
 struct gguf_context;
 struct ggml_context;
@@ -45,7 +46,7 @@ namespace llama_ollama_compat {
 // so there's nowhere to write the transformed bytes. Disabling mmap
 // makes the loader pre-allocate real backend buffers, after which our
 // load_op overrides land in writable memory.
-bool translate_metadata(const llama_model_loader * ml,
+LLAMA_API bool translate_metadata(const llama_model_loader * ml,
                         gguf_context * meta,
                         ggml_context * ctx,
                         std::string & arch_name,
@@ -54,19 +55,19 @@ bool translate_metadata(const llama_model_loader * ml,
 // Called from llama_model_loader's weights_map population loop. Returns
 // true to drop a tensor from the loader — used to hide embedded vision
 // tensors from the text model's view without modifying the gguf_context.
-bool should_skip_tensor(const llama_model_loader * ml, const char * tensor_name);
+LLAMA_API bool should_skip_tensor(const llama_model_loader * ml, const char * tensor_name);
 
 // Called from clip_model_loader's constructor. Rewrites the clip-facing
 // view of the metadata (arch=clip, clip.vision.* KVs, renamed tensors)
 // so the rest of clip.cpp can load a monolithic GGUF unchanged.
-void translate_clip_metadata(gguf_context * meta, ggml_context * ctx);
+LLAMA_API void translate_clip_metadata(gguf_context * meta, ggml_context * ctx);
 
 // Called from clip.cpp's tensor-loading loop, before the normal file read.
 // If this tensor was marked for type promotion by translate_clip_metadata
 // (e.g. F16->F32), performs the conversion and writes the result into
 // `cur` (host memcpy or backend_tensor_set based on `buft`). Returns true
 // when the tensor was handled — caller should skip its normal read path.
-bool maybe_load_tensor(ggml_tensor * cur,
+LLAMA_API bool maybe_load_tensor(ggml_tensor * cur,
                        const char * source_file,
                        size_t file_offset,
                        ggml_backend_buffer_type_t buft);
@@ -75,7 +76,7 @@ bool maybe_load_tensor(ggml_tensor * cur,
 // the model file path from the per-loader registry populated by
 // translate_metadata, and derives the buffer type from cur->buffer
 // internally, which keeps the call site in the patch to one line.
-bool maybe_load_text_tensor(const llama_model_loader * ml,
+LLAMA_API bool maybe_load_text_tensor(const llama_model_loader * ml,
                             ggml_tensor * cur,
                             size_t file_offset);
 
@@ -89,7 +90,7 @@ bool maybe_load_text_tensor(const llama_model_loader * ml,
 // this hook replaced), the cache holds one active tensor per loader.
 // Returns the requested range (copied into buf when buf is non-null) or
 // nullptr when no load op exists for this tensor.
-const void * maybe_load_text_tensor_range(const llama_model_loader * ml,
+LLAMA_API const void * maybe_load_text_tensor_range(const llama_model_loader * ml,
                                           ggml_tensor * cur,
                                           size_t offs,
                                           size_t size,
@@ -99,6 +100,6 @@ const void * maybe_load_text_tensor_range(const llama_model_loader * ml,
 // positive embedding size only for Ollama compatibility cases whose projector
 // metadata already follows upstream naming, but whose legacy projector type
 // is missing from that helper in the pinned llama.cpp version.
-int maybe_clip_mmproj_embd(const char * projector_type, int projection_dim);
+LLAMA_API int maybe_clip_mmproj_embd(const char * projector_type, int projection_dim);
 
 } // namespace llama_ollama_compat

llama/server/CMakeLists.txt

@@ -269,19 +269,16 @@ unset(_llama_cpp_root)
 # Link the Ollama-compat source files into the fetched llama target.
 # Kept separate from the hook patch so our .cpp/.h stay
 # on-disk in llama/compat/ rather than being copied into _deps/.
-if(_ollama_link_compat_sources AND DEFINED OLLAMA_LLAMA_CPP_COMPAT_DIR)
+if(_ollama_link_compat_sources AND DEFINED OLLAMA_LLAMA_CPP_COMPAT_DIR AND TARGET llama)
     file(GLOB _compat_sources CONFIGURE_DEPENDS
         ${OLLAMA_LLAMA_CPP_COMPAT_DIR}/*.cpp)
-    foreach(_compat_target IN ITEMS llama mtmd)
-        if(TARGET ${_compat_target})
-            target_sources(${_compat_target} PRIVATE ${_compat_sources})
-            target_include_directories(${_compat_target} PRIVATE
-                ${OLLAMA_LLAMA_CPP_COMPAT_DIR}
-                ${llama_cpp_SOURCE_DIR}/src)
-        endif()
-    endforeach()
+    target_sources(llama PRIVATE ${_compat_sources})
+    set_source_files_properties(${_compat_sources}
+        TARGET_DIRECTORY llama
+        PROPERTIES SKIP_UNITY_BUILD_INCLUSION ON)
+    target_include_directories(llama PRIVATE ${OLLAMA_LLAMA_CPP_COMPAT_DIR})
     if(TARGET mtmd)
-        target_compile_definitions(mtmd PRIVATE OLLAMA_COMPAT_MTMD_BUILD)
+        target_include_directories(mtmd PRIVATE ${OLLAMA_LLAMA_CPP_COMPAT_DIR})
     endif()
 endif()