add vulkan for whisper

Makefile

@@ -715,7 +715,7 @@ sampling.o: common/sampling.cpp common/common.h common/sampling.h common/log.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 console.o: common/console.cpp common/console.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
-expose.o: expose.cpp expose.h
+expose.o: expose.cpp expose.h model_adapter.cpp
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
 # sd.cpp objects
@@ -730,6 +730,8 @@ sdcpp_vulkan.o: otherarch/sdcpp/sdtype_adapter.cpp otherarch/sdcpp/stable-diffus
 #whisper objects
 whispercpp_default.o: otherarch/whispercpp/whisper_adapter.cpp
 	$(CXX) $(CXXFLAGS) -c $< -o $@
+whispercpp_vulkan.o: otherarch/whispercpp/whisper_adapter.cpp
+	$(CXX) $(CXXFLAGS) $(VULKAN_FLAGS) -c $< -o $@
 whispercpp_cublas.o: otherarch/whispercpp/whisper_adapter.cpp
 	$(CXX) $(CXXFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
 
@@ -942,10 +944,10 @@ koboldcpp_hipblas:
 endif
 
 ifdef VULKAN_BUILD
-koboldcpp_vulkan: ggml_v4_vulkan.o ggml-cpu.o ggml-ops.o ggml-vec.o ggml-binops.o ggml-unops.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter_vulkan.o ggml-vulkan.o ggml-vulkan-shaders.o sdcpp_vulkan.o whispercpp_default.o tts_default.o embeddings_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o ggml-backend-reg_vulkan.o ggml-repack.o $(OBJS_FULL) $(OBJS)
+koboldcpp_vulkan: ggml_v4_vulkan.o ggml-cpu.o ggml-ops.o ggml-vec.o ggml-binops.o ggml-unops.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter_vulkan.o ggml-vulkan.o ggml-vulkan-shaders.o sdcpp_vulkan.o whispercpp_vulkan.o tts_default.o embeddings_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o ggml-backend-reg_vulkan.o ggml-repack.o $(OBJS_FULL) $(OBJS)
 	$(VULKAN_BUILD)
 ifdef NOAVX2_BUILD
-koboldcpp_vulkan_noavx2: ggml_v4_vulkan_noavx2.o ggml-cpu_v4_noavx2.o ggml-ops-noavx2.o ggml-vec-noavx2.o ggml-binops.o ggml-unops.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_vulkan_noavx2.o ggml-vulkan-noext.o ggml-vulkan-shaders-noext.o sdcpp_vulkan.o whispercpp_default.o tts_default.o embeddings_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o ggml-backend-reg_vulkan.o ggml-repack.o $(OBJS_SIMPLE) $(OBJS)
+koboldcpp_vulkan_noavx2: ggml_v4_vulkan_noavx2.o ggml-cpu_v4_noavx2.o ggml-ops-noavx2.o ggml-vec-noavx2.o ggml-binops.o ggml-unops.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_vulkan_noavx2.o ggml-vulkan-noext.o ggml-vulkan-shaders-noext.o sdcpp_vulkan.o whispercpp_vulkan.o tts_default.o embeddings_default.o llavaclip_vulkan.o llava.o ggml-backend_vulkan.o ggml-backend-reg_vulkan.o ggml-repack.o $(OBJS_SIMPLE) $(OBJS)
 	$(VULKAN_BUILD)
 else
 koboldcpp_vulkan_noavx2:

model_adapter.cpp

@@ -221,7 +221,7 @@ std::string gguf_get_model_arch(const std::string & gguf_filename)
                 }
            }
        }
-       else if(vocabsiz < 31998 || vocabsiz > 33000)
+       else if((vocabsiz < 31998 || vocabsiz > 33000) && vocabsiz<51864) //avoid whisper false positive
        {
            //anything outside the llama v1 range is assumed to be NeoX
            fileformat = FileFormat::NEOX_6;
@@ -268,7 +268,11 @@ std::string gguf_get_model_arch(const std::string & gguf_filename)
                     }
                 }
            }
-
+       } 
+       else if (vocabsiz>=51864 && vocabsiz<=51865)
+       {
+            printf("\nWhisper model detected - you should load it as a whisper model instead, not a text model!\n");
+            fileformat = FileFormat::BADFORMAT; //known whisper formats, do not proceed
        }
     }
     else if(magic == 0x67676d66) //v2 format ggmf

otherarch/whispercpp/whisper.cpp

@@ -12,6 +12,10 @@
 #include "ggml-cuda.h"
 #endif
 
+#ifdef GGML_USE_VULKAN
+#include "ggml-vulkan.h"
+#endif
+
 #ifdef GGML_USE_SYCL
 #include "ggml-sycl.h"
 #endif
@@ -1262,6 +1266,16 @@ static ggml_backend_t whisper_backend_init(const whisper_context_params & params
     }
 #endif
 
+#ifdef GGML_USE_VULKAN
+    if (params.use_gpu) {
+        WHISPER_LOG_INFO("%s: using Vulkan backend\n", __func__);
+        backend_gpu = ggml_backend_vk_init(params.gpu_device);
+        if (!backend_gpu) {
+            WHISPER_LOG_ERROR("%s: ggml_backend_vk_init() failed\n", __func__);
+        }
+    }
+#endif
+
     if (backend_gpu) {
         return backend_gpu;
     }