feat(backend): rfdetr-cpp native object detection + segmentation backend (#10028)

Adds a Go native gRPC backend that dlopens librfdetrcpp.so (built from mudler/rf-detr.cpp at the pinned RFDETR_VERSION) via purego and exposes the rfdetr.cpp inference pipeline through LocalAI's existing Detect RPC. Supports all 5 RF-DETR detection variants (Nano/Small/Base/Medium/Large) and 6 segmentation variants (SegNano/SegSmall/SegMedium/SegLarge/ SegXLarge/Seg2XLarge) with F32/F16/Q8_0/Q4_K quantizations. Pre-built GGUFs ship at mudler/rfdetr-cpp-* on HuggingFace. Detection returns Bbox + class_name + confidence; segmentation also returns PNG-encoded per-detection masks via the rfdetr_capi accessor functions (rfdetr_capi_get_detection_{class_id,box,score,class_name, mask_png}). End-to-end verified through POST /v1/detection: HTTP -> gRPC -> purego dlopen -> rfdetr.cpp -> ggml -> response (9 detections on the detection model, 21 detections + valid PNG masks on the seg-nano model against the kitchen fixture). Wiring: - backend/go/rfdetr-cpp/{main.go,gorfdetrcpp.go,CMakeLists.txt, Makefile,run.sh,package.sh,test.sh,.gitignore} - Top-level Makefile: BACKEND_RFDETR_CPP, docker-build target, .NOTPARALLEL, prepare-test-extra, test-extra - backend/go/rfdetr-cpp/Makefile: `test` target invoked by test-extra - .github/backend-matrix.yml: CPU + CUDA-12/13 + L4T CUDA-12/13 (arm64) + HIP + Vulkan (amd64 + arm64) + SYCL f32/f16 - backend/index.yaml: rfdetr-cpp meta anchor + latest/development image entries for every matrix tag-suffix - .github/workflows/bump_deps.yaml: RFDETR_VERSION pin tracking (mudler/rf-detr.cpp branch main) - gallery/index.yaml: 11 rfdetr-cpp-* entries (nano + 4 detection variants + 6 seg variants), all backed by mudler/rfdetr-cpp-* on HuggingFace with sha256 pinning on the F16 default - core/gallery/importers/rfdetr.go: GGUF auto-routing for HF imports (mudler/rfdetr-cpp-* repos route to rfdetr-cpp, Transformer-format repos stay on the Python rfdetr backend; explicit preferences.backend overrides both heuristics) - core/gallery/importers/rfdetr_test.go: table-driven coverage of the auto-routing + a live mudler/rfdetr-cpp-nano cross-check scripts/changed-backends.js needs no change: the existing Dockerfile.golang -> backend/go/${item.backend}/ branch already routes the 9 rfdetr-cpp matrix entries to the correct backend path. Assisted-by: Claude:claude-opus-4-7 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>
2026-06-01 04:28:59 -04:00 · 2026-05-27 18:43:57 +02:00
parent 893e69cbf8
commit 7a4ca8f60d
18 changed files with 1697 additions and 6 deletions
--- a/backend/go/rfdetr-cpp/.gitignore
+++ b/backend/go/rfdetr-cpp/.gitignore
@@ -0,0 +1,7 @@
+sources/
+build*/
+package/
+librfdetrcpp*.so
+rfdetr-cpp
+test-models/
+test-data/
--- a/backend/go/rfdetr-cpp/CMakeLists.txt
+++ b/backend/go/rfdetr-cpp/CMakeLists.txt
@@ -0,0 +1,79 @@
+cmake_minimum_required(VERSION 3.18)
+project(librfdetrcpp LANGUAGES C CXX)
+
+set(CMAKE_POSITION_INDEPENDENT_CODE ON)
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+# Static-link ggml + rfdetr so the resulting .so has no runtime dependency on
+# extra ggml/rfdetr shared libraries — only on libc/libstdc++/libgomp, which
+# the LocalAI package step bundles into the docker image.
+set(BUILD_SHARED_LIBS OFF CACHE BOOL "Build static libraries" FORCE)
+
+# rfdetr.cpp build switches: skip CLI/tests, keep static lib.
+set(RFDETR_BUILD_CLI OFF CACHE BOOL "Disable rfdetr CLI" FORCE)
+set(RFDETR_BUILD_TESTS OFF CACHE BOOL "Disable rfdetr tests" FORCE)
+set(RFDETR_SHARED OFF CACHE BOOL "Build rfdetr as static lib" FORCE)
+
+# rt-detr.cpp's top-level CMakeLists invokes
+# `bash ${CMAKE_SOURCE_DIR}/scripts/apply_ggml_patches.sh` to apply its
+# in-tree ggml patches before descending into the submodule. When we
+# `add_subdirectory` it from a parent project, `CMAKE_SOURCE_DIR` points
+# at *our* directory, not theirs, so the script path resolves wrong.
+#
+# Run the patches script ourselves up front (it's idempotent — re-running
+# is a no-op once patches are applied) so the rt-detr.cpp configure step
+# is essentially a no-op for the patch hook.
+set(RFDETR_CPP_SRC ${CMAKE_CURRENT_SOURCE_DIR}/sources/rt-detr.cpp)
+if(EXISTS ${RFDETR_CPP_SRC}/scripts/apply_ggml_patches.sh)
+    execute_process(
+        COMMAND bash ${RFDETR_CPP_SRC}/scripts/apply_ggml_patches.sh
+        RESULT_VARIABLE _rfdetr_patch_result
+        OUTPUT_VARIABLE _rfdetr_patch_output
+        ERROR_VARIABLE  _rfdetr_patch_error
+        OUTPUT_STRIP_TRAILING_WHITESPACE
+        ERROR_STRIP_TRAILING_WHITESPACE)
+    if(NOT _rfdetr_patch_result EQUAL 0)
+        message(FATAL_ERROR
+            "Failed to apply ggml patches (exit ${_rfdetr_patch_result}):\n"
+            "stdout:\n${_rfdetr_patch_output}\n"
+            "stderr:\n${_rfdetr_patch_error}")
+    endif()
+    message(STATUS "${_rfdetr_patch_output}")
+endif()
+
+# Stage a shim 'scripts/apply_ggml_patches.sh' under our source dir so that
+# rt-detr.cpp's CMakeLists — which calls
+#   bash ${CMAKE_SOURCE_DIR}/scripts/apply_ggml_patches.sh
+# — finds an idempotent no-op there. The real patches have already been
+# applied above; this just satisfies the path lookup.
+file(MAKE_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/scripts)
+file(WRITE ${CMAKE_CURRENT_SOURCE_DIR}/scripts/apply_ggml_patches.sh
+"#!/usr/bin/env bash
+# Shim - patches were already applied by the parent CMakeLists.
+exit 0
+")
+execute_process(COMMAND chmod +x ${CMAKE_CURRENT_SOURCE_DIR}/scripts/apply_ggml_patches.sh)
+
+add_subdirectory(./sources/rt-detr.cpp)
+
+# rfdetr.cpp's C-API symbols already live inside librfdetr (src/rfdetr_capi.cpp
+# is compiled into the lib). We re-export them via a MODULE library that
+# whole-archive-links rfdetr so the symbols are visible at dlopen time.
+add_library(rfdetrcpp MODULE
+    sources/rt-detr.cpp/src/rfdetr_capi.cpp)
+
+target_include_directories(rfdetrcpp PRIVATE
+    sources/rt-detr.cpp/include
+    sources/rt-detr.cpp/src
+    sources/rt-detr.cpp/third_party/stb
+)
+
+target_link_libraries(rfdetrcpp PRIVATE rfdetr ggml)
+
+if(CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 9.0)
+    target_link_libraries(rfdetrcpp PRIVATE stdc++fs)
+endif()
+
+set_property(TARGET rfdetrcpp PROPERTY CXX_STANDARD 17)
+set_target_properties(rfdetrcpp PROPERTIES LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
--- a/backend/go/rfdetr-cpp/Makefile
+++ b/backend/go/rfdetr-cpp/Makefile
@@ -0,0 +1,135 @@
+CMAKE_ARGS?=
+BUILD_TYPE?=
+NATIVE?=false
+
+GOCMD?=go
+GO_TAGS?=
+JOBS?=$(shell nproc --ignore=1)
+
+# rt-detr.cpp (GitHub redirects the historical mudler/rt-detr.cpp to the new
+# mudler/rf-detr.cpp slug). Pin to a specific commit if you need a stable
+# build; leaving this on `master` always picks up the latest C-API surface
+# (incl. the per-detection accessor functions used by gorfdetrcpp.go).
+RFDETR_REPO?=https://github.com/mudler/rf-detr.cpp.git
+RFDETR_VERSION?=main
+
+ifeq ($(NATIVE),false)
+	CMAKE_ARGS+=-DGGML_NATIVE=OFF
+endif
+
+# Forward LocalAI's BUILD_TYPE to the matching ggml backend switch.
+ifeq ($(BUILD_TYPE),cublas)
+	CMAKE_ARGS+=-DGGML_CUDA=ON -DRFDETR_GGML_CUDA=ON
+else ifeq ($(BUILD_TYPE),openblas)
+	CMAKE_ARGS+=-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+else ifeq ($(BUILD_TYPE),clblas)
+	CMAKE_ARGS+=-DGGML_CLBLAST=ON
+else ifeq ($(BUILD_TYPE),hipblas)
+	ROCM_HOME ?= /opt/rocm
+	ROCM_PATH ?= /opt/rocm
+	export CXX=$(ROCM_HOME)/llvm/bin/clang++
+	export CC=$(ROCM_HOME)/llvm/bin/clang
+	AMDGPU_TARGETS?=gfx908,gfx90a,gfx942,gfx950,gfx1030,gfx1100,gfx1101,gfx1102,gfx1200,gfx1201
+	CMAKE_ARGS+=-DGGML_HIPBLAS=ON -DRFDETR_GGML_HIPBLAS=ON -DAMDGPU_TARGETS=$(AMDGPU_TARGETS)
+else ifeq ($(BUILD_TYPE),vulkan)
+	CMAKE_ARGS+=-DGGML_VULKAN=ON -DRFDETR_GGML_VULKAN=ON
+else ifeq ($(OS),Darwin)
+	ifneq ($(BUILD_TYPE),metal)
+		CMAKE_ARGS+=-DGGML_METAL=OFF
+	else
+		CMAKE_ARGS+=-DGGML_METAL=ON
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+		CMAKE_ARGS+=-DRFDETR_GGML_METAL=ON
+	endif
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DGGML_SYCL_F16=ON
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx
+endif
+
+sources/rt-detr.cpp:
+	mkdir -p sources && \
+	git clone --recursive $(RFDETR_REPO) sources/rt-detr.cpp && \
+	cd sources/rt-detr.cpp && \
+	git checkout $(RFDETR_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+# Detect OS
+UNAME_S := $(shell uname -s)
+
+# Only build CPU variants on Linux
+ifeq ($(UNAME_S),Linux)
+	VARIANT_TARGETS = librfdetrcpp-avx.so librfdetrcpp-avx2.so librfdetrcpp-avx512.so librfdetrcpp-fallback.so
+else
+	# On non-Linux (e.g., Darwin), build only fallback variant
+	VARIANT_TARGETS = librfdetrcpp-fallback.so
+endif
+
+rfdetr-cpp: main.go gorfdetrcpp.go $(VARIANT_TARGETS)
+	CGO_ENABLED=0 $(GOCMD) build -tags "$(GO_TAGS)" -o rfdetr-cpp ./
+
+package: rfdetr-cpp
+	bash package.sh
+
+build: package
+
+clean: purge
+	rm -rf librfdetrcpp*.so rfdetr-cpp package sources
+
+purge:
+	rm -rf build*
+
+# Build all variants (Linux only)
+ifeq ($(UNAME_S),Linux)
+librfdetrcpp-avx.so: sources/rt-detr.cpp
+	rm -rfv build-$@
+	$(info ${GREEN}I rfdetr-cpp build info:avx${RESET})
+	SO_TARGET=$@ CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) librfdetrcpp-custom
+	rm -rfv build-$@
+
+librfdetrcpp-avx2.so: sources/rt-detr.cpp
+	rm -rfv build-$@
+	$(info ${GREEN}I rfdetr-cpp build info:avx2${RESET})
+	SO_TARGET=$@ CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=off -DGGML_FMA=on -DGGML_F16C=on -DGGML_BMI2=on" $(MAKE) librfdetrcpp-custom
+	rm -rfv build-$@
+
+librfdetrcpp-avx512.so: sources/rt-detr.cpp
+	rm -rfv build-$@
+	$(info ${GREEN}I rfdetr-cpp build info:avx512${RESET})
+	SO_TARGET=$@ CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=on -DGGML_FMA=on -DGGML_F16C=on -DGGML_BMI2=on" $(MAKE) librfdetrcpp-custom
+	rm -rfv build-$@
+endif
+
+# Build fallback variant (all platforms)
+librfdetrcpp-fallback.so: sources/rt-detr.cpp
+	rm -rfv build-$@
+	$(info ${GREEN}I rfdetr-cpp build info:fallback${RESET})
+	SO_TARGET=$@ CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) librfdetrcpp-custom
+	rm -rfv build-$@
+
+librfdetrcpp-custom: CMakeLists.txt
+	mkdir -p build-$(SO_TARGET) && \
+	cd build-$(SO_TARGET) && \
+	cmake .. $(CMAKE_ARGS) && \
+	cmake --build . --config Release -j$(JOBS) && \
+	cd .. && \
+	mv build-$(SO_TARGET)/librfdetrcpp.so ./$(SO_TARGET)
+
+all: rfdetr-cpp package
+
+# `test` is invoked by the top-level Makefile's `test-extra` target. It builds
+# the backend binary + the fallback shared library (needed for dlopen at
+# runtime), then runs test.sh which downloads the test models + COCO image
+# and exercises the gRPC Load/Detect wire path via the Go smoke test in
+# main_test.go for both the detection and segmentation models.
+test: rfdetr-cpp librfdetrcpp-fallback.so
+	bash test.sh
--- a/backend/go/rfdetr-cpp/gorfdetrcpp.go
+++ b/backend/go/rfdetr-cpp/gorfdetrcpp.go
@@ -0,0 +1,195 @@
+package main
+
+// gorfdetrcpp.go - gRPC handlers (Load, Detect) for the rfdetr-cpp backend.
+//
+// Embeds base.SingleThread to default unimplemented RPCs to "not supported"
+// while we only implement object detection.
+
+import (
+	"encoding/base64"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strconv"
+	"unsafe"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+)
+
+// Default upper bound on detections returned per image. RF-DETR's decoder
+// queries are limited to a few hundred; 300 is a safe ceiling.
+const defaultTopK = 300
+
+// rfdetr_handle_t is a uintptr-typed opaque handle (see include/rfdetr_capi.h).
+var (
+	// rfdetr_capi_load(const char* model_path, int n_threads, rfdetr_handle_t* out_handle) -> int
+	CapiLoad func(modelPath string, nThreads int32, outHandle *uintptr) int32
+	// rfdetr_capi_unload(rfdetr_handle_t handle) -> int
+	CapiUnload func(handle uintptr) int32
+	// rfdetr_capi_detect_path(handle, image_path, threshold, top_k, out_json) -> int
+	CapiDetectPath func(handle uintptr, imagePath string, threshold float32, topK uint32, outJSON *uintptr) int32
+	// rfdetr_capi_detect_buffer(handle, bytes, len, threshold, top_k, out_json) -> int
+	CapiDetectBuffer func(handle uintptr, bytes uintptr, length uintptr, threshold float32, topK uint32, outJSON *uintptr) int32
+	// rfdetr_capi_free_string(char* s)
+	CapiFreeString func(s uintptr)
+	// rfdetr_capi_get_n_detections(handle) -> int
+	CapiGetNDetections func(handle uintptr) int32
+	// rfdetr_capi_get_detection_class_id(handle, i) -> int
+	CapiGetDetectionClassID func(handle uintptr, i int32) int32
+	// rfdetr_capi_get_detection_box(handle, i, out_xyxy[4]) -> int (0 on success)
+	CapiGetDetectionBox func(handle uintptr, i int32, outXYXY uintptr) int32
+	// rfdetr_capi_get_detection_score(handle, i) -> float
+	CapiGetDetectionScore func(handle uintptr, i int32) float32
+	// rfdetr_capi_get_detection_class_name(handle, i, buf, buf_size) -> int (needed/written; two-call sizing)
+	CapiGetDetectionClassName func(handle uintptr, i int32, buf uintptr, bufSize int32) int32
+	// rfdetr_capi_get_detection_mask_png(handle, i, buf, buf_size) -> int (needed/written; 0 means no mask)
+	CapiGetDetectionMaskPNG func(handle uintptr, i int32, buf uintptr, bufSize int32) int32
+)
+
+type RFDetrCpp struct {
+	base.SingleThread
+	handle uintptr
+}
+
+// Load loads the GGUF model at opts.ModelFile (joined with opts.ModelPath if relative)
+// and stores the handle for later Detect calls.
+func (r *RFDetrCpp) Load(opts *pb.ModelOptions) error {
+	modelFile := opts.ModelFile
+	if modelFile == "" {
+		modelFile = opts.Model
+	}
+	if modelFile == "" {
+		return fmt.Errorf("rfdetr-cpp: ModelFile is empty")
+	}
+
+	var modelPath string
+	if filepath.IsAbs(modelFile) {
+		modelPath = modelFile
+	} else {
+		modelPath = filepath.Join(opts.ModelPath, modelFile)
+	}
+
+	if _, err := os.Stat(modelPath); err != nil {
+		return fmt.Errorf("rfdetr-cpp: model file not found: %s: %w", modelPath, err)
+	}
+
+	threads := opts.Threads
+	if threads <= 0 {
+		threads = 4
+	}
+
+	// Release previous model if any (re-Load).
+	if r.handle != 0 {
+		CapiUnload(r.handle)
+		r.handle = 0
+	}
+
+	var h uintptr
+	rc := CapiLoad(modelPath, threads, &h)
+	if rc != 0 || h == 0 {
+		return fmt.Errorf("rfdetr-cpp: rfdetr_capi_load failed with rc=%d for %s", rc, modelPath)
+	}
+	r.handle = h
+	return nil
+}
+
+// Detect runs object detection on the base64-encoded image in opts.Src at
+// opts.Threshold, returning one pb.Detection per result. Seg models also
+// populate Detection.Mask with PNG-encoded mask bytes.
+func (r *RFDetrCpp) Detect(opts *pb.DetectOptions) (pb.DetectResponse, error) {
+	if r.handle == 0 {
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: model not loaded")
+	}
+
+	// Decode base64 image and write to temp file.
+	imgData, err := base64.StdEncoding.DecodeString(opts.Src)
+	if err != nil {
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: failed to decode base64 image: %w", err)
+	}
+
+	tmpFile, err := os.CreateTemp("", "rfdetr-*.img")
+	if err != nil {
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: failed to create temp file: %w", err)
+	}
+	defer func() { _ = os.Remove(tmpFile.Name()) }()
+
+	if _, err := tmpFile.Write(imgData); err != nil {
+		_ = tmpFile.Close()
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: failed to write temp file: %w", err)
+	}
+	if err := tmpFile.Close(); err != nil {
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: failed to close temp file: %w", err)
+	}
+
+	threshold := opts.Threshold
+	if threshold <= 0 {
+		threshold = 0.5
+	}
+
+	// JSON output from detect_path is unused: we read structured detections via
+	// the accessor functions. Still must free the returned string.
+	var jsonPtr uintptr
+	rc := CapiDetectPath(r.handle, tmpFile.Name(), threshold, uint32(defaultTopK), &jsonPtr)
+	if jsonPtr != 0 {
+		CapiFreeString(jsonPtr)
+	}
+	if rc != 0 {
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: detect failed with rc=%d", rc)
+	}
+
+	n := CapiGetNDetections(r.handle)
+	if n < 0 {
+		return pb.DetectResponse{}, fmt.Errorf("rfdetr-cpp: invalid n_detections=%d", n)
+	}
+
+	detections := make([]*pb.Detection, 0, n)
+	for i := int32(0); i < n; i++ {
+		var bbox [4]float32 // x1, y1, x2, y2
+		if rc := CapiGetDetectionBox(r.handle, i, uintptr(unsafe.Pointer(&bbox[0]))); rc != 0 {
+			continue
+		}
+		cid := CapiGetDetectionClassID(r.handle, i)
+		score := CapiGetDetectionScore(r.handle, i)
+
+		// Two-call sizing for class_name.
+		var className string
+		nameSize := CapiGetDetectionClassName(r.handle, i, 0, 0)
+		if nameSize > 1 {
+			buf := make([]byte, nameSize)
+			written := CapiGetDetectionClassName(r.handle, i, uintptr(unsafe.Pointer(&buf[0])), nameSize)
+			// `written` is the same number (needed bytes including NUL); strip NUL.
+			if written > 0 && int(written) <= len(buf) {
+				className = string(buf[:written-1])
+			} else {
+				className = string(buf[:len(buf)-1])
+			}
+		}
+		if className == "" {
+			className = strconv.Itoa(int(cid))
+		}
+
+		// Two-call sizing for mask PNG (returns 0 when no mask).
+		var mask []byte
+		maskSize := CapiGetDetectionMaskPNG(r.handle, i, 0, 0)
+		if maskSize > 0 {
+			maskBuf := make([]byte, maskSize)
+			CapiGetDetectionMaskPNG(r.handle, i, uintptr(unsafe.Pointer(&maskBuf[0])), maskSize)
+			mask = maskBuf
+		}
+
+		detections = append(detections, &pb.Detection{
+			X:          bbox[0],
+			Y:          bbox[1],
+			Width:      bbox[2] - bbox[0],
+			Height:     bbox[3] - bbox[1],
+			Confidence: score,
+			ClassName:  className,
+			Mask:       mask,
+		})
+	}
+
+	return pb.DetectResponse{
+		Detections: detections,
+	}, nil
+}
--- a/backend/go/rfdetr-cpp/main.go
+++ b/backend/go/rfdetr-cpp/main.go
@@ -0,0 +1,61 @@
+package main
+
+// main.go - entry point for the rfdetr-cpp gRPC backend.
+//
+// Dlopens librfdetrcpp-<variant>.so via purego at the path in
+// RFDETR_LIBRARY (set by run.sh based on /proc/cpuinfo), registers the
+// rfdetr_capi_* C ABI symbols, then starts the gRPC server.
+
+import (
+	"flag"
+	"os"
+
+	"github.com/ebitengine/purego"
+	grpc "github.com/mudler/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+type LibFuncs struct {
+	FuncPtr any
+	Name    string
+}
+
+func main() {
+	// Get library name from environment variable, default to fallback
+	libName := os.Getenv("RFDETR_LIBRARY")
+	if libName == "" {
+		libName = "./librfdetrcpp-fallback.so"
+	}
+
+	rfdetrLib, err := purego.Dlopen(libName, purego.RTLD_NOW|purego.RTLD_GLOBAL)
+	if err != nil {
+		panic(err)
+	}
+
+	libFuncs := []LibFuncs{
+		{&CapiLoad, "rfdetr_capi_load"},
+		{&CapiUnload, "rfdetr_capi_unload"},
+		{&CapiDetectPath, "rfdetr_capi_detect_path"},
+		{&CapiDetectBuffer, "rfdetr_capi_detect_buffer"},
+		{&CapiFreeString, "rfdetr_capi_free_string"},
+		{&CapiGetNDetections, "rfdetr_capi_get_n_detections"},
+		{&CapiGetDetectionClassID, "rfdetr_capi_get_detection_class_id"},
+		{&CapiGetDetectionBox, "rfdetr_capi_get_detection_box"},
+		{&CapiGetDetectionScore, "rfdetr_capi_get_detection_score"},
+		{&CapiGetDetectionClassName, "rfdetr_capi_get_detection_class_name"},
+		{&CapiGetDetectionMaskPNG, "rfdetr_capi_get_detection_mask_png"},
+	}
+
+	for _, lf := range libFuncs {
+		purego.RegisterLibFunc(lf.FuncPtr, rfdetrLib, lf.Name)
+	}
+
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &RFDetrCpp{}); err != nil {
+		panic(err)
+	}
+}
--- a/backend/go/rfdetr-cpp/main_test.go
+++ b/backend/go/rfdetr-cpp/main_test.go
@@ -0,0 +1,220 @@
+package main
+
+// main_test.go - end-to-end smoke test for the rfdetr-cpp gRPC backend.
+//
+// Spawns the compiled rfdetr-cpp binary on a free local port, dials it via
+// gRPC, and exercises LoadModel + Detect against the test fixtures
+// downloaded by test.sh. Two scenarios:
+//
+//   1. detection — loads rfdetr-nano-q8_0.gguf and asserts at least one
+//      detection comes back with a non-empty class name and a bounding box
+//      of non-zero size.
+//   2. segmentation — loads rfdetr-seg-nano-q8_0.gguf and additionally
+//      asserts that at least one detection carries a PNG-encoded mask blob
+//      (verified by PNG magic bytes).
+//
+// Both specs Skip cleanly if their fixtures are missing so the test target
+// stays usable on a fresh checkout where models haven't been downloaded.
+
+import (
+	"context"
+	"encoding/base64"
+	"fmt"
+	"net"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"testing"
+	"time"
+
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/credentials/insecure"
+)
+
+func TestRFDetrCpp(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "rfdetr-cpp backend smoke suite")
+}
+
+// freePort grabs an ephemeral TCP port and immediately releases it so the
+// spawned backend can bind to it. There is a tiny TOCTOU window here but in
+// practice it's adequate for a smoke test on a quiet runner.
+func freePort() int {
+	l, err := net.Listen("tcp", "127.0.0.1:0")
+	Expect(err).ToNot(HaveOccurred(), "freePort listen")
+	port := l.Addr().(*net.TCPAddr).Port
+	Expect(l.Close()).To(Succeed())
+	return port
+}
+
+// startBackend spawns the rfdetr-cpp binary on the given port and waits
+// until it accepts TCP connections (up to 10s). The returned cleanup func
+// kills the process and reaps it.
+func startBackend(port int) func() {
+	binary, err := filepath.Abs("./rfdetr-cpp")
+	Expect(err).ToNot(HaveOccurred())
+	if _, err := os.Stat(binary); err != nil {
+		Skip(fmt.Sprintf("backend binary not built: %s (run `make rfdetr-cpp` first)", binary))
+	}
+
+	libPath, err := filepath.Abs("./librfdetrcpp-fallback.so")
+	Expect(err).ToNot(HaveOccurred())
+	if _, err := os.Stat(libPath); err != nil {
+		Skip(fmt.Sprintf("fallback library not built: %s (run `make librfdetrcpp-fallback.so` first)", libPath))
+	}
+
+	addr := fmt.Sprintf("127.0.0.1:%d", port)
+	cmd := exec.Command(binary, "--addr", addr)
+	cmd.Env = append(os.Environ(), "RFDETR_LIBRARY="+libPath)
+	cmd.Stdout = os.Stderr
+	cmd.Stderr = os.Stderr
+	Expect(cmd.Start()).To(Succeed())
+
+	cleanup := func() {
+		if cmd.Process != nil {
+			_ = cmd.Process.Kill()
+			_, _ = cmd.Process.Wait()
+		}
+	}
+
+	deadline := time.Now().Add(10 * time.Second)
+	for time.Now().Before(deadline) {
+		c, err := net.DialTimeout("tcp", addr, 200*time.Millisecond)
+		if err == nil {
+			_ = c.Close()
+			return cleanup
+		}
+		time.Sleep(200 * time.Millisecond)
+	}
+
+	cleanup()
+	Fail(fmt.Sprintf("backend did not become ready on %s within 10s", addr))
+	return func() {}
+}
+
+// loadTestImage reads the COCO test image downloaded by test.sh and returns
+// its base64-encoded content (the wire format accepted by the Detect RPC).
+func loadTestImage() string {
+	imgPath, err := filepath.Abs("test-data/test.jpg")
+	Expect(err).ToNot(HaveOccurred())
+	imgBytes, err := os.ReadFile(imgPath)
+	if err != nil {
+		Skip(fmt.Sprintf("test image not present: %s (run test.sh first)", imgPath))
+	}
+	return base64.StdEncoding.EncodeToString(imgBytes)
+}
+
+// dialBackend opens a gRPC client connection to the spawned backend.
+func dialBackend(port int) (pb.BackendClient, func()) {
+	addr := fmt.Sprintf("127.0.0.1:%d", port)
+	conn, err := grpc.NewClient(addr, grpc.WithTransportCredentials(insecure.NewCredentials()))
+	Expect(err).ToNot(HaveOccurred())
+	return pb.NewBackendClient(conn), func() { _ = conn.Close() }
+}
+
+// modelPathOrSkip resolves a model file under ./test-models/ and Skip()s
+// the current spec if it's missing.
+func modelPathOrSkip(name string) string {
+	modelDir, err := filepath.Abs("test-models")
+	Expect(err).ToNot(HaveOccurred())
+	modelPath := filepath.Join(modelDir, name)
+	if _, err := os.Stat(modelPath); err != nil {
+		Skip(fmt.Sprintf("model not present: %s (run test.sh first)", modelPath))
+	}
+	return modelPath
+}
+
+var _ = Describe("rfdetr-cpp backend", func() {
+	It("runs object detection against a known-good COCO image", func() {
+		modelPath := modelPathOrSkip("rfdetr-nano-q8_0.gguf")
+		imgB64 := loadTestImage()
+
+		port := freePort()
+		cleanup := startBackend(port)
+		defer cleanup()
+
+		client, closeConn := dialBackend(port)
+		defer closeConn()
+
+		ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+		defer cancel()
+
+		loadResp, err := client.LoadModel(ctx, &pb.ModelOptions{
+			Model:     "rfdetr-nano-q8_0.gguf",
+			ModelFile: modelPath,
+			Threads:   2,
+		})
+		Expect(err).ToNot(HaveOccurred(), "LoadModel")
+		Expect(loadResp.GetSuccess()).To(BeTrue(), "LoadModel reported failure: %s", loadResp.GetMessage())
+
+		detResp, err := client.Detect(ctx, &pb.DetectOptions{
+			Src:       imgB64,
+			Threshold: 0.5,
+		})
+		Expect(err).ToNot(HaveOccurred(), "Detect")
+		Expect(detResp.GetDetections()).ToNot(BeEmpty(), "no detections returned on a known-good COCO image")
+
+		_, _ = fmt.Fprintf(GinkgoWriter, "detection OK: %d detections\n", len(detResp.GetDetections()))
+		for i, d := range detResp.GetDetections() {
+			Expect(d.GetClassName()).ToNot(BeEmpty(), "detection %d has empty class_name", i)
+			Expect(d.GetConfidence()).To(BeNumerically(">=", float32(0.5)),
+				"detection %d below threshold", i)
+			Expect(d.GetWidth()).To(BeNumerically(">", float32(0)),
+				"detection %d has non-positive width", i)
+			Expect(d.GetHeight()).To(BeNumerically(">", float32(0)),
+				"detection %d has non-positive height", i)
+		}
+	})
+
+	It("runs segmentation and returns PNG-encoded masks", func() {
+		modelPath := modelPathOrSkip("rfdetr-seg-nano-q8_0.gguf")
+		imgB64 := loadTestImage()
+
+		port := freePort()
+		cleanup := startBackend(port)
+		defer cleanup()
+
+		client, closeConn := dialBackend(port)
+		defer closeConn()
+
+		ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+		defer cancel()
+
+		loadResp, err := client.LoadModel(ctx, &pb.ModelOptions{
+			Model:     "rfdetr-seg-nano-q8_0.gguf",
+			ModelFile: modelPath,
+			Threads:   2,
+		})
+		Expect(err).ToNot(HaveOccurred(), "LoadModel")
+		Expect(loadResp.GetSuccess()).To(BeTrue(), "LoadModel reported failure: %s", loadResp.GetMessage())
+
+		detResp, err := client.Detect(ctx, &pb.DetectOptions{
+			Src:       imgB64,
+			Threshold: 0.5,
+		})
+		Expect(err).ToNot(HaveOccurred(), "Detect")
+		Expect(detResp.GetDetections()).ToNot(BeEmpty(), "no detections returned from segmentation model")
+
+		haveMask := false
+		for i, d := range detResp.GetDetections() {
+			m := d.GetMask()
+			if len(m) == 0 {
+				continue
+			}
+			haveMask = true
+			// Verify PNG magic: 89 50 4E 47 ("\x89PNG").
+			Expect(len(m)).To(BeNumerically(">=", 4), "detection %d mask too short", i)
+			Expect([]byte{m[0], m[1], m[2], m[3]}).To(Equal([]byte{0x89, 'P', 'N', 'G'}),
+				"detection %d mask is not a PNG", i)
+		}
+		Expect(haveMask).To(BeTrue(),
+			"segmentation model returned %d detections but none carried a mask",
+			len(detResp.GetDetections()))
+
+		_, _ = fmt.Fprintf(GinkgoWriter, "segmentation OK: %d detections, at least one with PNG mask\n",
+			len(detResp.GetDetections()))
+	})
+})
--- a/backend/go/rfdetr-cpp/package.sh
+++ b/backend/go/rfdetr-cpp/package.sh
@@ -0,0 +1,59 @@
+#!/bin/bash
+
+# Script to copy the appropriate libraries based on architecture
+
+set -e
+
+CURDIR=$(dirname "$(realpath $0)")
+REPO_ROOT="${CURDIR}/../../.."
+
+# Create lib directory
+mkdir -p $CURDIR/package/lib
+
+cp -avf $CURDIR/librfdetrcpp-*.so $CURDIR/package/
+cp -avf $CURDIR/rfdetr-cpp $CURDIR/package/
+cp -fv $CURDIR/run.sh $CURDIR/package/
+
+# Detect architecture and copy appropriate libraries
+if [ -f "/lib64/ld-linux-x86-64.so.2" ]; then
+    # x86_64 architecture
+    echo "Detected x86_64 architecture, copying x86_64 libraries..."
+    cp -arfLv /lib64/ld-linux-x86-64.so.2 $CURDIR/package/lib/ld.so
+    cp -arfLv /lib/x86_64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
+    cp -arfLv /lib/x86_64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
+    cp -arfLv /lib/x86_64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
+    cp -arfLv /lib/x86_64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
+elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
+    # ARM64 architecture
+    echo "Detected ARM64 architecture, copying ARM64 libraries..."
+    cp -arfLv /lib/ld-linux-aarch64.so.1 $CURDIR/package/lib/ld.so
+    cp -arfLv /lib/aarch64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
+    cp -arfLv /lib/aarch64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
+    cp -arfLv /lib/aarch64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
+    cp -arfLv /lib/aarch64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
+elif [ $(uname -s) = "Darwin" ]; then
+    echo "Detected Darwin"
+else
+    echo "Error: Could not detect architecture"
+    exit 1
+fi
+
+# Package GPU libraries based on BUILD_TYPE
+GPU_LIB_SCRIPT="${REPO_ROOT}/scripts/build/package-gpu-libs.sh"
+if [ -f "$GPU_LIB_SCRIPT" ]; then
+    echo "Packaging GPU libraries for BUILD_TYPE=${BUILD_TYPE:-cpu}..."
+    source "$GPU_LIB_SCRIPT" "$CURDIR/package/lib"
+    package_gpu_libs
+fi
+
+echo "Packaging completed successfully"
+ls -liah $CURDIR/package/
+ls -liah $CURDIR/package/lib/
--- a/backend/go/rfdetr-cpp/run.sh
+++ b/backend/go/rfdetr-cpp/run.sh
@@ -0,0 +1,52 @@
+#!/bin/bash
+set -ex
+
+# Get the absolute current dir where the script is located
+CURDIR=$(dirname "$(realpath $0)")
+
+cd /
+
+echo "CPU info:"
+if [ "$(uname)" != "Darwin" ]; then
+	grep -e "model\sname" /proc/cpuinfo | head -1
+	grep -e "flags" /proc/cpuinfo | head -1
+fi
+
+LIBRARY="$CURDIR/librfdetrcpp-fallback.so"
+
+if [ "$(uname)" != "Darwin" ]; then
+	if grep -q -e "\savx\s" /proc/cpuinfo ; then
+		echo "CPU:    AVX    found OK"
+		if [ -e $CURDIR/librfdetrcpp-avx.so ]; then
+			LIBRARY="$CURDIR/librfdetrcpp-avx.so"
+		fi
+	fi
+
+	if grep -q -e "\savx2\s" /proc/cpuinfo ; then
+		echo "CPU:    AVX2   found OK"
+		if [ -e $CURDIR/librfdetrcpp-avx2.so ]; then
+			LIBRARY="$CURDIR/librfdetrcpp-avx2.so"
+		fi
+	fi
+
+	# Check avx 512
+	if grep -q -e "\savx512f\s" /proc/cpuinfo ; then
+		echo "CPU:    AVX512F found OK"
+		if [ -e $CURDIR/librfdetrcpp-avx512.so ]; then
+			LIBRARY="$CURDIR/librfdetrcpp-avx512.so"
+		fi
+	fi
+fi
+
+export LD_LIBRARY_PATH=$CURDIR/lib:$LD_LIBRARY_PATH
+export RFDETR_LIBRARY=$LIBRARY
+
+# If there is a lib/ld.so, use it
+if [ -f $CURDIR/lib/ld.so ]; then
+	echo "Using lib/ld.so"
+	echo "Using library: $LIBRARY"
+	exec $CURDIR/lib/ld.so $CURDIR/rfdetr-cpp "$@"
+fi
+
+echo "Using library: $LIBRARY"
+exec $CURDIR/rfdetr-cpp "$@"
--- a/backend/go/rfdetr-cpp/test.sh
+++ b/backend/go/rfdetr-cpp/test.sh
@@ -0,0 +1,55 @@
+#!/bin/bash
+set -e
+
+CURDIR=$(dirname "$(realpath $0)")
+
+echo "Running rfdetr-cpp backend tests..."
+
+# Test models from the mudler/rfdetr-cpp-* HuggingFace repos. Both the
+# detection (nano-q8_0, ~36 MB) and segmentation (seg-nano-q8_0, ~40 MB)
+# variants are downloaded so the Go smoke test exercises both code paths.
+RFDETR_MODEL_DIR="${RFDETR_MODEL_DIR:-$CURDIR/test-models}"
+
+RFDETR_DET_FILE="${RFDETR_DET_FILE:-rfdetr-nano-q8_0.gguf}"
+RFDETR_DET_URL="${RFDETR_DET_URL:-https://huggingface.co/mudler/rfdetr-cpp-nano/resolve/main/rfdetr-nano-q8_0.gguf}"
+
+RFDETR_SEG_FILE="${RFDETR_SEG_FILE:-rfdetr-seg-nano-q8_0.gguf}"
+RFDETR_SEG_URL="${RFDETR_SEG_URL:-https://huggingface.co/mudler/rfdetr-cpp-seg-nano/resolve/main/rfdetr-seg-nano-q8_0.gguf}"
+
+mkdir -p "$RFDETR_MODEL_DIR"
+
+if [ ! -f "$RFDETR_MODEL_DIR/$RFDETR_DET_FILE" ]; then
+    echo "Downloading rfdetr nano-q8_0 detection model..."
+    curl -L -o "$RFDETR_MODEL_DIR/$RFDETR_DET_FILE" "$RFDETR_DET_URL" --progress-bar
+fi
+
+if [ ! -f "$RFDETR_MODEL_DIR/$RFDETR_SEG_FILE" ]; then
+    echo "Downloading rfdetr seg-nano-q8_0 segmentation model..."
+    curl -L -o "$RFDETR_MODEL_DIR/$RFDETR_SEG_FILE" "$RFDETR_SEG_URL" --progress-bar
+fi
+
+# Use a real COCO test image from the upstream rf-detr.cpp repo (~46 KB).
+# A synthetic 64x64 red PNG was too synthetic to elicit detections from a
+# real model — the smoke test would always trivially pass with zero
+# detections.
+TEST_IMAGE_DIR="$CURDIR/test-data"
+TEST_IMAGE_FILE="$TEST_IMAGE_DIR/test.jpg"
+TEST_IMAGE_URL="${TEST_IMAGE_URL:-https://raw.githubusercontent.com/mudler/rf-detr.cpp/main/tests/fixtures/ci/test_image.jpg}"
+
+mkdir -p "$TEST_IMAGE_DIR"
+if [ ! -f "$TEST_IMAGE_FILE" ]; then
+    echo "Downloading COCO test image..."
+    curl -L -o "$TEST_IMAGE_FILE" "$TEST_IMAGE_URL" --progress-bar
+fi
+
+echo "rfdetr-cpp test setup complete."
+echo "  detection model:    $RFDETR_MODEL_DIR/$RFDETR_DET_FILE"
+echo "  segmentation model: $RFDETR_MODEL_DIR/$RFDETR_SEG_FILE"
+echo "  test image:         $TEST_IMAGE_FILE"
+
+# Run the Go smoke test: spawns the backend binary on a free port, calls
+# LoadModel + Detect via gRPC for both detection and segmentation models.
+echo ""
+echo "Running Go smoke test..."
+cd "$CURDIR"
+go test -v -timeout 5m ./...
--- a/backend/index.yaml
+++ b/backend/index.yaml
@@ -253,6 +253,34 @@
    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-sam3-cpp"
    intel: "intel-sycl-f32-sam3-cpp"
    vulkan: "vulkan-sam3-cpp"
+- &rfdetrcpp
+  name: "rfdetr-cpp"
+  alias: "rfdetr-cpp"
+  license: apache-2.0
+  description: |
+    Native RF-DETR object detection and instance segmentation in C/C++
+    using GGML. Loads pre-built GGUF weights from the mudler/rfdetr-cpp-*
+    family (Nano/Small/Base/Medium/Large + SegNano/SegSmall/SegMedium)
+    and returns bounding boxes, class labels, confidence scores, and
+    (for segmentation variants) PNG-encoded per-detection masks.
+  urls:
+    - https://github.com/mudler/rf-detr.cpp
+  tags:
+    - object-detection
+    - image-segmentation
+    - rfdetr
+    - gpu
+    - cpu
+  capabilities:
+    default: "cpu-rfdetr-cpp"
+    nvidia: "cuda12-rfdetr-cpp"
+    nvidia-cuda-12: "cuda12-rfdetr-cpp"
+    nvidia-cuda-13: "cuda13-rfdetr-cpp"
+    nvidia-l4t: "nvidia-l4t-arm64-rfdetr-cpp"
+    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-rfdetr-cpp"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-rfdetr-cpp"
+    intel: "intel-sycl-f32-rfdetr-cpp"
+    vulkan: "vulkan-rfdetr-cpp"
 - &vllm
  name: "vllm"
  license: apache-2.0
@@ -2349,6 +2377,99 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-vulkan-sam3-cpp"
  mirrors:
    - localai/localai-backends:master-gpu-vulkan-sam3-cpp
+## rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "rfdetr-cpp-development"
+  capabilities:
+    default: "cpu-rfdetr-cpp-development"
+    nvidia: "cuda12-rfdetr-cpp-development"
+    nvidia-cuda-12: "cuda12-rfdetr-cpp-development"
+    nvidia-cuda-13: "cuda13-rfdetr-cpp-development"
+    nvidia-l4t: "nvidia-l4t-arm64-rfdetr-cpp-development"
+    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-rfdetr-cpp-development"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-rfdetr-cpp-development"
+    intel: "intel-sycl-f32-rfdetr-cpp-development"
+    vulkan: "vulkan-rfdetr-cpp-development"
+- !!merge <<: *rfdetrcpp
+  name: "cpu-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-cpu-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cpu-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-cpu-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cuda12-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cuda12-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cuda13-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cuda13-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "nvidia-l4t-arm64-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-arm64-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-arm64-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "nvidia-l4t-arm64-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-arm64-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-arm64-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cuda13-nvidia-l4t-arm64-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-cuda-13-arm64-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-cuda-13-arm64-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "cuda13-nvidia-l4t-arm64-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "intel-sycl-f32-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f32-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-sycl-f32-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "intel-sycl-f32-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f32-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-sycl-f32-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "intel-sycl-f16-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f16-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-sycl-f16-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "intel-sycl-f16-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-sycl-f16-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "vulkan-rfdetr-cpp"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-vulkan-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:latest-gpu-vulkan-rfdetr-cpp
+- !!merge <<: *rfdetrcpp
+  name: "vulkan-rfdetr-cpp-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-vulkan-rfdetr-cpp"
+  mirrors:
+    - localai/localai-backends:master-gpu-vulkan-rfdetr-cpp
 ## Rerankers
 - !!merge <<: *rerankers
  name: "rerankers-development"