diff --git a/backend/cpp/bonsai/Makefile b/backend/cpp/bonsai/Makefile
index e5afba624f6c..4af58ee8cbd1 100644
--- a/backend/cpp/bonsai/Makefile
+++ b/backend/cpp/bonsai/Makefile
@@ -36,6 +36,10 @@ PATCHES_DIR := $(CURRENT_MAKEFILE_DIR)/patches
 define bonsai-build
 	rm -rf $(CURRENT_MAKEFILE_DIR)/../bonsai-$(1)-build
 	cp -rf $(LLAMA_CPP_DIR) $(CURRENT_MAKEFILE_DIR)/../bonsai-$(1)-build
+	# Drop patches vendored for upstream llama.cpp: the fork tree diverges, so
+	# they reject there. Fork-specific patches live in backend/cpp/bonsai/patches/
+	# and are applied by apply-patches.sh below.
+	rm -rf $(CURRENT_MAKEFILE_DIR)/../bonsai-$(1)-build/patches
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../bonsai-$(1)-build purge
 	$(info $(GREEN)I bonsai build info:$(1)$(RESET))
 	LLAMA_REPO=$(LLAMA_REPO) LLAMA_VERSION=$(BONSAI_VERSION) \
@@ -68,6 +72,10 @@ bonsai-fallback:
 bonsai-cpu-all:
 	rm -rf $(CURRENT_MAKEFILE_DIR)/../bonsai-cpu-all-build
 	cp -rf $(LLAMA_CPP_DIR) $(CURRENT_MAKEFILE_DIR)/../bonsai-cpu-all-build
+	# Drop patches vendored for upstream llama.cpp: the fork tree diverges, so
+	# they reject there. Fork-specific patches live in backend/cpp/bonsai/patches/
+	# and are applied by apply-patches.sh below.
+	rm -rf $(CURRENT_MAKEFILE_DIR)/../bonsai-cpu-all-build/patches
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../bonsai-cpu-all-build purge
 	$(info $(GREEN)I bonsai build info:cpu-all-variants$(RESET))
 	LLAMA_REPO=$(LLAMA_REPO) LLAMA_VERSION=$(BONSAI_VERSION) \
diff --git a/backend/cpp/llama-cpp/prepare.sh b/backend/cpp/llama-cpp/prepare.sh
index 27c664e8c7d4..aec0c1d14526 100644
--- a/backend/cpp/llama-cpp/prepare.sh
+++ b/backend/cpp/llama-cpp/prepare.sh
@@ -1,17 +1,19 @@
 #!/bin/bash
 
+set -e
+
 ## Patches
 
-## Apply patches from the `patches` directory
+## Apply patches from the `patches` directory. Runs under set -e so a
+## rejected patch aborts the build here, loudly, instead of surfacing later
+## as a confusing compile error. A missing or empty patches dir is a no-op.
 if [ -d "patches" ]; then
     for patch in $(ls patches); do
         echo "Applying patch $patch"
         patch -d llama.cpp/ -p1 < patches/$patch
-    done 
+    done
 fi
 
-set -e
-
 for file in $(ls llama.cpp/tools/server/); do
     cp -rfv llama.cpp/tools/server/$file llama.cpp/tools/grpc-server/
 done
diff --git a/backend/cpp/turboquant/Makefile b/backend/cpp/turboquant/Makefile
index a32adf0b62ae..cfc2593c0670 100644
--- a/backend/cpp/turboquant/Makefile
+++ b/backend/cpp/turboquant/Makefile
@@ -37,6 +37,10 @@ PATCHES_DIR := $(CURRENT_MAKEFILE_DIR)/patches
 define turboquant-build
 	rm -rf $(CURRENT_MAKEFILE_DIR)/../turboquant-$(1)-build
 	cp -rf $(LLAMA_CPP_DIR) $(CURRENT_MAKEFILE_DIR)/../turboquant-$(1)-build
+	# Drop patches vendored for upstream llama.cpp: the fork tree diverges, so
+	# they reject there. Fork-specific patches live in backend/cpp/turboquant/patches/
+	# and are applied by apply-patches.sh below.
+	rm -rf $(CURRENT_MAKEFILE_DIR)/../turboquant-$(1)-build/patches
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../turboquant-$(1)-build purge
 	# Augment the copied grpc-server.cpp's KV-cache allow-list with the
 	# fork's turbo2/turbo3/turbo4 types. We patch the *copy*, never the
@@ -74,6 +78,10 @@ turboquant-fallback:
 turboquant-cpu-all:
 	rm -rf $(CURRENT_MAKEFILE_DIR)/../turboquant-cpu-all-build
 	cp -rf $(LLAMA_CPP_DIR) $(CURRENT_MAKEFILE_DIR)/../turboquant-cpu-all-build
+	# Drop patches vendored for upstream llama.cpp: the fork tree diverges, so
+	# they reject there. Fork-specific patches live in backend/cpp/turboquant/patches/
+	# and are applied by apply-patches.sh below.
+	rm -rf $(CURRENT_MAKEFILE_DIR)/../turboquant-cpu-all-build/patches
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../turboquant-cpu-all-build purge
 	bash $(CURRENT_MAKEFILE_DIR)/patch-grpc-server.sh $(CURRENT_MAKEFILE_DIR)/../turboquant-cpu-all-build/grpc-server.cpp
 	$(info $(GREEN)I turboquant build info:cpu-all-variants$(RESET))