diff --git a/llama/Dockerfile b/llama/Dockerfile
index c636ee5d..0e2386d4 100644
--- a/llama/Dockerfile
+++ b/llama/Dockerfile
@@ -9,7 +9,7 @@ ARG ROCM_VERSION=6.1.2
 
 ### To create a local image for building linux binaries on mac or windows with efficient incremental builds
 #
-# docker build --platform linux/amd64 -t builder-amd64 -f Dockerfile.new --target unified-builder-amd64 .
+# docker build --platform linux/amd64 -t builder-amd64 -f llama/Dockerfile --target unified-builder-amd64 .
 # docker run --platform linux/amd64 --rm -it -v $(pwd):/go/src/github.com/ollama/ollama/ builder-amd64
 #
 ### Then incremental builds will be much faster in this container
@@ -41,7 +41,7 @@ ENTRYPOINT [ "zsh" ]
 ### To create a local image for building linux binaries on mac or linux/arm64 with efficient incremental builds
 # Note: this does not contain jetson variants
 #
-# docker build --platform linux/arm64 -t builder-arm64 -f Dockerfile.new --target unified-builder-arm64 .
+# docker build --platform linux/arm64 -t builder-arm64 -f llama/Dockerfile --target unified-builder-arm64 .
 # docker run --platform linux/arm64 --rm -it -v $(pwd):/go/src/github.com/ollama/ollama/ builder-arm64
 #
 FROM --platform=linux/arm64 rockylinux:8 AS unified-builder-arm64
diff --git a/llama/make/Makefile.rocm b/llama/make/Makefile.rocm
index 865714b8..ee3fe2e4 100644
--- a/llama/make/Makefile.rocm
+++ b/llama/make/Makefile.rocm
@@ -21,7 +21,8 @@ else ifeq ($(OS),linux)
 	GPU_LIB_DIR_LINUX := $(HIP_PATH)/lib
 	GPU_COMPILER_LINUX := $(shell X=$$(which hipcc 2>/dev/null) && echo $$X)
 	GPU_COMPILER:=$(GPU_COMPILER_LINUX)
-	ROCM_TRANSITIVE_LIBS = $(shell ldd $(ROCM_LIBS) | grep "=>" | cut -f2 -d= | cut -f2 -d' '  | grep -e rocm -e amdgpu -e libtinfo -e libnuma -e libelf | sort -u )
+	ROCM_TRANSITIVE_LIBS_INITIAL = $(sort $(shell ldd $(ROCM_LIBS) | grep "=>" | cut -f2 -d= | cut -f2 -d' '  | grep -e rocm -e amdgpu -e libtinfo -e libnuma -e libelf))
+	ROCM_TRANSITIVE_LIBS = $(sort $(shell readlink -f $(ROCM_TRANSITIVE_LIBS_INITIAL)) $(ROCM_TRANSITIVE_LIBS_INITIAL))
 endif
 
 # TODO future multi-variant support for ROCm
@@ -36,14 +37,18 @@ GPU_RUNNER_DRIVER_LIB_LINK := -lamdhip64
 GPU_RUNNER_LIBS_SHORT := hipblas rocblas
 GPU_PATH_ROOT_WIN=$(dir $(GPU_LIB_DIR_WIN))
 GPU_PATH_ROOT_LINUX=$(dir $(GPU_LIB_DIR_LINUX))
-GPU_COMPILER_CFLAGS_WIN = $(CFLAGS)
+GPU_COMPILER_CFLAGS_WIN = $(CFLAGS) -D_WIN32_WINNT=0x602
 GPU_COMPILER_CFLAGS_LINUX = $(CFLAGS) -fPIC -D_GNU_SOURCE
-GPU_COMPILER_CXXFLAGS_WIN = $(CXXFLAGS)
+GPU_COMPILER_CXXFLAGS_WIN = $(CXXFLAGS) -D_WIN32_WINNT=0x602
 GPU_COMPILER_CXXFLAGS_LINUX = $(CXXFLAGS) -fPIC -D_GNU_SOURCE
 
 ROCM_LIBS = $(wildcard $(addsuffix .$(SHARED_EXT),$(addprefix $(GPU_LIB_DIR)/$(SHARED_PREFIX),$(GPU_RUNNER_LIBS_SHORT))))
-ROCM_DIST_DEPS_DIR = $(abspath $(SRC_DIR)/../dist/$(OS)-$(ARCH)-rocm)/lib/ollama
-ROCM_DIST_DEPS_LIBS = $(addprefix $(ROCM_DIST_DEPS_DIR)/,$(notdir $(ROCM_LIBS)) $(notdir $(ROCM_TRANSITIVE_LIBS)))
+ifeq ($(OS),windows)
+	ROCM_DIST_DEPS_DIR = $(abspath $(SRC_DIR)/../dist/$(OS)-$(ARCH))/lib/ollama
+else ifeq ($(OS),linux)
+	ROCM_DIST_DEPS_DIR = $(abspath $(SRC_DIR)/../dist/$(OS)-$(ARCH)-rocm)/lib/ollama
+endif
+ROCM_DIST_DEPS_LIBS= $(sort $(addprefix $(ROCM_DIST_DEPS_DIR)/,$(notdir $(ROCM_LIBS)) $(notdir $(ROCM_TRANSITIVE_LIBS))))
 ROCBLAS_DIST_DEP_MANIFEST = $(ROCM_DIST_DEPS_DIR)/rocblas/library/TensileManifest.txt
 
 ifeq ($(OS),linux)
@@ -84,8 +89,7 @@ GPU_COMPILER_CUFLAGS = \
 	-Wno-pass-failed \
 	-Wno-deprecated-declarations \
 	-Wno-unused-result \
-	-I. \
-	$(foreach arch, $(HIP_ARCHS_COMMON), --offload-arch=$(arch))
+	-I.
 
 include make/gpu.make
 
@@ -98,4 +102,4 @@ $(ROCBLAS_DIST_DEP_MANIFEST):
 	@echo "rocblas library copy complete"
 $(ROCM_DIST_DEPS_LIBS): 
 	@-mkdir -p $(dir $@)
-	$(CP) $(dir $(filter %$(notdir $@),$(ROCM_LIBS) $(ROCM_TRANSITIVE_LIBS)))/$(notdir $@)* $(dir $@)
+	$(CP) $(dir $(filter %$(notdir $@),$(ROCM_LIBS) $(ROCM_TRANSITIVE_LIBS)))/$(notdir $@) $(dir $@)
diff --git a/llama/make/gpu.make b/llama/make/gpu.make
index 7143bed6..2d63af56 100644
--- a/llama/make/gpu.make
+++ b/llama/make/gpu.make
@@ -79,7 +79,7 @@ $(GPU_RUNNER_NAME): $(BUILD_RUNNERS) $(DIST_RUNNERS) $(PAYLOAD_RUNNERS)
 # Build targets
 $(BUILD_DIR)/%.$(GPU_RUNNER_NAME).$(OBJ_EXT): %.cu
 	@-mkdir -p $(dir $@)
-	$(CCACHE) $(GPU_COMPILER) -c $(GPU_COMPILER_CUFLAGS) $(GPU_RUNNER_ARCH_FLAGS) -o $@ $<
+	$(CCACHE) $(GPU_COMPILER) -c $(GPU_COMPILER_CFLAGS) $(GPU_COMPILER_CUFLAGS) $(GPU_RUNNER_ARCH_FLAGS) -o $@ $<
 $(BUILD_DIR)/%.$(GPU_RUNNER_NAME).$(OBJ_EXT): %.c
 	@-mkdir -p $(dir $@)
 	$(CCACHE) $(GPU_COMPILER) -c $(GPU_COMPILER_CFLAGS) -o $@ $<
@@ -97,14 +97,14 @@ $(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).
 # Distribution targets
 $(RUNNERS_DIST_DIR)/%: $(RUNNERS_BUILD_DIR)/%
 	@-mkdir -p $(dir $@)
-	cp $< $@
+	$(CP) $< $@
 $(RUNNERS_DIST_DIR)/$(GPU_RUNNER_NAME)/ollama_llama_server$(EXE_EXT): $(DIST_LIB_DIR)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT)
 $(DIST_LIB_DIR)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT): $(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT)
 	@-mkdir -p $(dir $@)
-	cp $< $@
+	$(CP) $< $@
 $(DIST_GPU_RUNNER_LIB_DEPS): 
 	@-mkdir -p $(dir $@)
-	$(CP) $(GPU_LIB_DIR)/$(notdir $@)* $(dir $@)
+	$(CP) $(GPU_LIB_DIR)/$(notdir $@) $(dir $@)
 
 # Payload targets
 $(RUNNERS_PAYLOAD_DIR)/%/ollama_llama_server.gz: $(RUNNERS_BUILD_DIR)/%/ollama_llama_server