chore(model): add silero-vad model config

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-02-03 03:02:38 -05:00 · 2024-11-26 14:28:41 +01:00
55 changed files with 294 additions and 1159 deletions
--- a/.github/labeler.yml
+++ b/.github/labeler.yml
@@ -1,11 +1,6 @@
 enhancements:
 - head-branch: ['^feature', 'feature']

-dependencies:
- any:
-  - changed-files:
-    - any-glob-to-any-file: 'Makefile'
-
 kind/documentation:
 - any:
  - changed-files:
--- a/.github/workflows/bump_deps.yaml
+++ b/.github/workflows/bump_deps.yaml
@@ -12,14 +12,23 @@ jobs:
          - repository: "ggerganov/llama.cpp"
            variable: "CPPLLAMA_VERSION"
            branch: "master"
+          - repository: "go-skynet/go-ggml-transformers.cpp"
+            variable: "GOGGMLTRANSFORMERS_VERSION"
+            branch: "master"
+          - repository: "donomii/go-rwkv.cpp"
+            variable: "RWKV_VERSION"
+            branch: "main"
          - repository: "ggerganov/whisper.cpp"
            variable: "WHISPER_CPP_VERSION"
            branch: "master"
-          - repository: "PABannier/bark.cpp"
-            variable: "BARKCPP_VERSION"
+          - repository: "go-skynet/go-bert.cpp"
+            variable: "BERT_VERSION"
+            branch: "master"
+          - repository: "go-skynet/bloomz.cpp"
+            variable: "BLOOMZ_VERSION"
            branch: "main"
-          - repository: "leejet/stable-diffusion.cpp"
-            variable: "STABLEDIFFUSION_GGML_VERSION"
+          - repository: "mudler/go-ggllm.cpp"
+            variable: "GOGGLLM_VERSION"
            branch: "master"
          - repository: "mudler/go-stable-diffusion"
            variable: "STABLEDIFFUSION_VERSION"
--- a/.gitignore
+++ b/.gitignore
@@ -2,7 +2,6 @@
 /sources/
 __pycache__/
 *.a
-*.o
 get-sources
 prepare-sources
 /backend/cpp/llama/grpc-server
--- a/104
+++ b/104
@@ -8,12 +8,16 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=cc98896db858df7aa40d0e16a505883ef196a482
+CPPLLAMA_VERSION?=47f931c8f9a26c072d71224bc8013cc66ea9e445

 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
 WHISPER_CPP_VERSION?=6266a9f9e56a5b925e9892acf650f3eb1245814d

+# bert.cpp version
+BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
+BERT_VERSION?=710044b124545415f555e4260d16b146c725a6e4
+
 # go-piper version
 PIPER_REPO?=https://github.com/mudler/go-piper
 PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0
@@ -26,14 +30,6 @@ STABLEDIFFUSION_VERSION?=4a3cd6aeae6f66ee57eae9a0075f8c58c3a6a38f
 TINYDREAM_REPO?=https://github.com/M0Rf30/go-tiny-dream
 TINYDREAM_VERSION?=c04fa463ace9d9a6464313aa5f9cd0f953b6c057

-# bark.cpp
-BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
-BARKCPP_VERSION?=v1.0.0
-
-# stablediffusion.cpp (ggml)
-STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=4570715727f35e5a07a76796d823824c8f42206c
-
 ONNX_VERSION?=1.20.0
 ONNX_ARCH?=x64
 ONNX_OS?=linux
@@ -202,6 +198,7 @@ ifeq ($(findstring tts,$(GO_TAGS)),tts)
 endif

 ALL_GRPC_BACKENDS=backend-assets/grpc/huggingface
+ALL_GRPC_BACKENDS+=backend-assets/grpc/bert-embeddings
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx2
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-fallback
@@ -209,14 +206,6 @@ ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-grpc
 ALL_GRPC_BACKENDS+=backend-assets/util/llama-cpp-rpc-server
 ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
-
-ifeq ($(ONNX_OS),linux)
-ifeq ($(ONNX_ARCH),x64)
-	ALL_GRPC_BACKENDS+=backend-assets/grpc/bark-cpp
-	ALL_GRPC_BACKENDS+=backend-assets/grpc/stablediffusion-ggml
-endif
-endif
-
 ALL_GRPC_BACKENDS+=backend-assets/grpc/local-store
 ALL_GRPC_BACKENDS+=backend-assets/grpc/silero-vad
 ALL_GRPC_BACKENDS+=$(OPTIONAL_GRPC)
@@ -239,6 +228,19 @@ endif

 all: help

+## BERT embeddings
+sources/go-bert.cpp:
+	mkdir -p sources/go-bert.cpp
+	cd sources/go-bert.cpp && \
+	git init && \
+	git remote add origin $(BERT_REPO) && \
+	git fetch origin && \
+	git checkout $(BERT_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/go-bert.cpp/libgobert.a: sources/go-bert.cpp
+	$(MAKE) -C sources/go-bert.cpp libgobert.a
+
 ## go-llama.cpp
 sources/go-llama.cpp:
 	mkdir -p sources/go-llama.cpp
@@ -252,23 +254,6 @@ sources/go-llama.cpp:
 sources/go-llama.cpp/libbinding.a: sources/go-llama.cpp
 	$(MAKE) -C sources/go-llama.cpp BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a

-## bark.cpp
-sources/bark.cpp:
-	git clone --recursive $(BARKCPP_REPO) sources/bark.cpp && \
-	cd sources/bark.cpp && \
-	git checkout $(BARKCPP_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/bark.cpp/build/libbark.a: sources/bark.cpp
-	cd sources/bark.cpp && \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) .. && \
-	cmake --build . --config Release
-
-backend/go/bark/libbark.a: sources/bark.cpp/build/libbark.a
-	$(MAKE) -C backend/go/bark libbark.a
-
 ## go-piper
 sources/go-piper:
 	mkdir -p sources/go-piper
@@ -282,7 +267,7 @@ sources/go-piper:
 sources/go-piper/libpiper_binding.a: sources/go-piper
 	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o

-## stable diffusion (onnx)
+## stable diffusion
 sources/go-stable-diffusion:
 	mkdir -p sources/go-stable-diffusion
 	cd sources/go-stable-diffusion && \
@@ -295,30 +280,6 @@ sources/go-stable-diffusion:
 sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
 	CPATH="$(CPATH):/usr/include/opencv4" $(MAKE) -C sources/go-stable-diffusion libstablediffusion.a

-## stablediffusion (ggml)
-sources/stablediffusion-ggml.cpp:
-	git clone --recursive $(STABLEDIFFUSION_GGML_REPO) sources/stablediffusion-ggml.cpp && \
-	cd sources/stablediffusion-ggml.cpp && \
-	git checkout $(STABLEDIFFUSION_GGML_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/stablediffusion-ggml.cpp/build/libstable-diffusion.a: sources/stablediffusion-ggml.cpp
-	cd sources/stablediffusion-ggml.cpp && \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) .. && \
-	cmake --build . --config Release
-
-backend/go/image/stablediffusion-ggml/libsd.a: sources/stablediffusion-ggml.cpp/build/libstable-diffusion.a
-	$(MAKE) -C backend/go/image/stablediffusion-ggml libsd.a
-
-backend-assets/grpc/stablediffusion-ggml: backend/go/image/stablediffusion-ggml/libsd.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ LIBRARY_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion-ggml ./backend/go/image/stablediffusion-ggml/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/stablediffusion-ggml
-endif
-
 sources/onnxruntime:
 	mkdir -p sources/onnxruntime
 	curl -L https://github.com/microsoft/onnxruntime/releases/download/v$(ONNX_VERSION)/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz -o sources/onnxruntime/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
@@ -359,11 +320,12 @@ sources/whisper.cpp:
 sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
 	cd sources/whisper.cpp && $(MAKE) libwhisper.a libggml.a

-get-sources: sources/go-llama.cpp sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp
+get-sources: sources/go-llama.cpp sources/go-piper sources/whisper.cpp sources/go-bert.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp

 replace:
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-bert.cpp=$(CURDIR)/sources/go-bert.cpp
 	$(GOCMD) mod edit -replace github.com/M0Rf30/go-tiny-dream=$(CURDIR)/sources/go-tiny-dream
 	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
 	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(CURDIR)/sources/go-stable-diffusion
@@ -372,6 +334,7 @@ replace:
 dropreplace:
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp/bindings/go
+	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-bert.cpp
 	$(GOCMD) mod edit -dropreplace github.com/M0Rf30/go-tiny-dream
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-piper
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-stable-diffusion
@@ -386,6 +349,7 @@ rebuild: ## Rebuilds the project
 	$(MAKE) -C sources/go-llama.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-stable-diffusion clean
+	$(MAKE) -C sources/go-bert.cpp clean
 	$(MAKE) -C sources/go-piper clean
 	$(MAKE) -C sources/go-tiny-dream clean
 	$(MAKE) build
@@ -400,9 +364,7 @@ clean: ## Remove build related file
 	rm -rf release/
 	rm -rf backend-assets/*
 	$(MAKE) -C backend/cpp/grpc clean
-	$(MAKE) -C backend/go/bark clean
 	$(MAKE) -C backend/cpp/llama clean
-	$(MAKE) -C backend/go/image/stablediffusion-ggml clean
 	rm -rf backend/cpp/llama-* || true
 	$(MAKE) dropreplace
 	$(MAKE) protogen-clean
@@ -745,6 +707,13 @@ backend-assets/espeak-ng-data: sources/go-piper sources/go-piper/libpiper_bindin
 backend-assets/grpc: protogen-go replace
 	mkdir -p backend-assets/grpc

+backend-assets/grpc/bert-embeddings: sources/go-bert.cpp sources/go-bert.cpp/libgobert.a backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-bert.cpp LIBRARY_PATH=$(CURDIR)/sources/go-bert.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/bert-embeddings
+endif
+
 backend-assets/grpc/huggingface: backend-assets/grpc
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/huggingface ./backend/go/llm/langchain/
 ifneq ($(UPX),)
@@ -804,6 +773,10 @@ backend-assets/grpc/llama-cpp-fallback: backend-assets/grpc backend/cpp/llama/ll
 	$(info ${GREEN}I llama-cpp build info:fallback${RESET})
 	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-fallback" build-llama-cpp-grpc-server
 	cp -rfv backend/cpp/llama-fallback/grpc-server backend-assets/grpc/llama-cpp-fallback
+# TODO: every binary should have its own folder instead, so can have different metal implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp backend/cpp/llama-fallback/llama.cpp/build/bin/ggml-metal.metal backend-assets/grpc/
+endif

 backend-assets/grpc/llama-cpp-cuda: backend-assets/grpc backend/cpp/llama/llama.cpp
 	cp -rf backend/cpp/llama backend/cpp/llama-cuda
@@ -851,13 +824,6 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/llama-ggml
 endif

-backend-assets/grpc/bark-cpp: backend/go/bark/libbark.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/bark/ LIBRARY_PATH=$(CURDIR)/backend/go/bark/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/bark-cpp
-endif
-
 backend-assets/grpc/piper: sources/go-piper sources/go-piper/libpiper_binding.a backend-assets/grpc backend-assets/espeak-ng-data
 	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/sources/go-piper \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
--- a/README.md
+++ b/README.md
@@ -92,8 +92,6 @@ local-ai run oci://localai/phi-2:latest

 ## 📰 Latest project news

- Dec 2024: stablediffusion.cpp backend (ggml) added ( https://github.com/mudler/LocalAI/pull/4289 )
- Nov 2024: Bark.cpp backend added ( https://github.com/mudler/LocalAI/pull/4287 )
 - Nov 2024: Voice activity detection models (**VAD**) added to the API: https://github.com/mudler/LocalAI/pull/4204
 - Oct 2024: examples moved to [LocalAI-examples](https://github.com/mudler/LocalAI-examples)
 - Aug 2024:  🆕 FLUX-1, [P2P Explorer](https://explorer.localai.io)
--- a/aio/cpu/embeddings.yaml
+++ b/aio/cpu/embeddings.yaml
@@ -1,7 +1,7 @@
 name: text-embedding-ada-002
-embeddings: true
+backend: bert-embeddings
 parameters:
-  model: huggingface://hugging-quants/Llama-3.2-1B-Instruct-Q4_K_M-GGUF/llama-3.2-1b-instruct-q4_k_m.gguf
+  model: huggingface://mudler/all-MiniLM-L6-v2/ggml-model-q4_0.bin

 usage: |
    You can test this model with curl like this:
--- a/backend/backend.proto
+++ b/backend/backend.proto
@@ -240,8 +240,6 @@ message ModelOptions {

  repeated string LoraAdapters = 60;
  repeated float LoraScales = 61;
-
-  repeated string Options = 62;
 }

 message Result {
--- a/backend/cpp/llama/Makefile
+++ b/backend/cpp/llama/Makefile
@@ -30,7 +30,9 @@ else ifeq ($(OS),Darwin)
 		CMAKE_ARGS+=-DGGML_METAL=OFF
 	else
 		CMAKE_ARGS+=-DGGML_METAL=ON
-		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+# Until this is tested properly, we disable embedded metal file
+# as we already embed it as part of the LocalAI assets
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=OFF
 		TARGET+=--target ggml-metal
 	endif
 endif
--- a/backend/go/bark/Makefile
+++ b/backend/go/bark/Makefile
@@ -1,25 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-
-BUILD_TYPE?=
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../sources/bark.cpp/examples -I$(INCLUDE_PATH)/../../../sources/bark.cpp/spm-headers -I$(INCLUDE_PATH)/../../../sources/bark.cpp -O3 -DNDEBUG -std=c++17 -fPIC
-LDFLAGS  = -L$(LIBRARY_PATH) -L$(LIBRARY_PATH)/../../../sources/bark.cpp/build/examples -lbark -lstdc++ -lm
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-gobark.o:
-	$(CXX) $(CXXFLAGS) gobark.cpp -o gobark.o -c $(LDFLAGS)
-
-libbark.a: gobark.o
-	cp $(INCLUDE_PATH)/../../../sources/bark.cpp/build/libbark.a ./
-	$(AR) rcs libbark.a gobark.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml.c.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-alloc.c.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-backend.c.o
-
-clean:
-	rm -f gobark.o libbark.a
--- a/backend/go/bark/gobark.cpp
+++ b/backend/go/bark/gobark.cpp
@@ -1,85 +0,0 @@
-#include <iostream>
-#include <tuple>
-
-#include "bark.h"
-#include "gobark.h"
-#include "common.h"
-#include "ggml.h"
-
-struct bark_context *c;
-
-void bark_print_progress_callback(struct bark_context *bctx, enum bark_encoding_step step, int progress, void *user_data) {
-    if (step == bark_encoding_step::SEMANTIC) {
-        printf("\rGenerating semantic tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::COARSE) {
-        printf("\rGenerating coarse tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::FINE) {
-        printf("\rGenerating fine tokens... %d%%", progress);
-    }
-    fflush(stdout);
-}
-
-int load_model(char *model) {
-    // initialize bark context
-    struct bark_context_params ctx_params = bark_context_default_params();
-    bark_params params;
-
-    params.model_path = model;
-
-   // ctx_params.verbosity = verbosity;
-    ctx_params.progress_callback = bark_print_progress_callback;
-    ctx_params.progress_callback_user_data = nullptr;
-
-    struct bark_context *bctx = bark_load_model(params.model_path.c_str(), ctx_params, params.seed);
-    if (!bctx) {
-        fprintf(stderr, "%s: Could not load model\n", __func__);
-        return 1;
-    }
-
-    c = bctx;
-
-    return 0;
-}
-
-int tts(char *text,int  threads, char *dst ) {
-
-    ggml_time_init();
-    const int64_t t_main_start_us = ggml_time_us();
-
-    // generate audio
-    if (!bark_generate_audio(c, text, threads)) {
-        fprintf(stderr, "%s: An error occured. If the problem persists, feel free to open an issue to report it.\n", __func__);
-        return 1;
-    }
-
-    const float *audio_data = bark_get_audio_data(c);
-    if (audio_data == NULL) {
-        fprintf(stderr, "%s: Could not get audio data\n", __func__);
-        return 1;
-    }
-
-    const int audio_arr_size = bark_get_audio_data_size(c);
-
-    std::vector<float> audio_arr(audio_data, audio_data + audio_arr_size);
-
-    write_wav_on_disk(audio_arr, dst);
-
-    // report timing
-    {
-        const int64_t t_main_end_us = ggml_time_us();
-        const int64_t t_load_us = bark_get_load_time(c);
-        const int64_t t_eval_us = bark_get_eval_time(c);
-
-        printf("\n\n");
-        printf("%s:     load time = %8.2f ms\n", __func__, t_load_us / 1000.0f);
-        printf("%s:     eval time = %8.2f ms\n", __func__, t_eval_us / 1000.0f);
-        printf("%s:    total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us) / 1000.0f);
-    }
-    
-    return 0;
-}
-
-int unload() {
-    bark_free(c);
-}
-
--- a/backend/go/bark/gobark.go
+++ b/backend/go/bark/gobark.go
@@ -1,52 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/../../../sources/bark.cpp/ -I${SRCDIR}/../../../sources/bark.cpp/encodec.cpp -I${SRCDIR}/../../../sources/bark.cpp/examples -I${SRCDIR}/../../../sources/bark.cpp/spm-headers
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/../../../sources/bark.cpp/build/examples -L${SRCDIR}/../../../sources/bark.cpp/build/encodec.cpp/ -lbark -lencodec -lcommon
-// #include <gobark.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type Bark struct {
-	base.SingleThread
-	threads int
-}
-
-func (sd *Bark) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	ret := C.load_model(modelFile)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
-
-func (sd *Bark) TTS(opts *pb.TTSRequest) error {
-	t := C.CString(opts.Text)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	threads := C.int(sd.threads)
-
-	ret := C.tts(t, threads, dst)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
--- a/backend/go/bark/gobark.h
+++ b/backend/go/bark/gobark.h
@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model);
-int tts(char *text,int  threads, char *dst );
-#ifdef __cplusplus
-}
-#endif
--- a/backend/go/image/stablediffusion-ggml/Makefile
+++ b/backend/go/image/stablediffusion-ggml/Makefile
@@ -1,21 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-
-BUILD_TYPE?=
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/ggml/include -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp -O3 -DNDEBUG -std=c++17 -fPIC
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-gosd.o:
-	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c
-
-libsd.a: gosd.o
-	cp $(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/build/libstable-diffusion.a ./libsd.a
-	$(AR) rcs libsd.a gosd.o
-
-clean:
-	rm -f gosd.o libsd.a
--- a/backend/go/image/stablediffusion-ggml/gosd.cpp
+++ b/backend/go/image/stablediffusion-ggml/gosd.cpp
@@ -1,228 +0,0 @@
-#include <stdio.h>
-#include <string.h>
-#include <time.h>
-#include <iostream>
-#include <random>
-#include <string>
-#include <vector>
-#include "gosd.h"
-
-// #include "preprocessing.hpp"
-#include "flux.hpp"
-#include "stable-diffusion.h"
-
-#define STB_IMAGE_IMPLEMENTATION
-#define STB_IMAGE_STATIC
-#include "stb_image.h"
-
-#define STB_IMAGE_WRITE_IMPLEMENTATION
-#define STB_IMAGE_WRITE_STATIC
-#include "stb_image_write.h"
-
-#define STB_IMAGE_RESIZE_IMPLEMENTATION
-#define STB_IMAGE_RESIZE_STATIC
-#include "stb_image_resize.h"
-
-// Names of the sampler method, same order as enum sample_method in stable-diffusion.h
-const char* sample_method_str[] = {
-    "euler_a",
-    "euler",
-    "heun",
-    "dpm2",
-    "dpm++2s_a",
-    "dpm++2m",
-    "dpm++2mv2",
-    "ipndm",
-    "ipndm_v",
-    "lcm",
-};
-
-// Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
-const char* schedule_str[] = {
-    "default",
-    "discrete",
-    "karras",
-    "exponential",
-    "ays",
-    "gits",
-};
-
-sd_ctx_t* sd_c;
-
-sample_method_t sample_method;
-
-int load_model(char *model, char* options[], int threads, int diff) {
-    fprintf (stderr, "Loading model!\n");
-
-    char *stableDiffusionModel = "";
-    if (diff == 1 ) {
-        stableDiffusionModel = model;
-        model = "";
-    }
-
-    // decode options. Options are in form optname:optvale, or if booleans only optname.
-    char *clip_l_path  = "";
-    char *clip_g_path  = "";
-    char *t5xxl_path  = "";
-    char *vae_path  = "";
-    char *scheduler = "";
-    char *sampler = "";
-
-    // If options is not NULL, parse options
-    for (int i = 0; options[i] != NULL; i++) {
-        char *optname = strtok(options[i], ":");
-        char *optval = strtok(NULL, ":");
-        if (optval == NULL) {
-            optval = "true";
-        }
-
-        if (!strcmp(optname, "clip_l_path")) {
-            clip_l_path = optval;
-        }
-        if (!strcmp(optname, "clip_g_path")) {
-            clip_g_path = optval;
-        }
-        if (!strcmp(optname, "t5xxl_path")) {
-            t5xxl_path = optval;
-        }
-        if (!strcmp(optname, "vae_path")) {
-            vae_path = optval;
-        }
-        if (!strcmp(optname, "scheduler")) {
-            scheduler = optval;
-        }
-        if (!strcmp(optname, "sampler")) {
-            sampler = optval;
-        }
-    }
-
-    int sample_method_found = -1;
-    for (int m = 0; m < N_SAMPLE_METHODS; m++) {
-        if (!strcmp(sampler, sample_method_str[m])) {
-            sample_method_found = m;
-        }
-    }
-    if (sample_method_found == -1) {
-        fprintf(stderr, "Invalid sample method, default to EULER_A!\n");
-        sample_method_found = EULER_A;
-    }
-    sample_method = (sample_method_t)sample_method_found;
-
-    int schedule_found            = -1;
-    for (int d = 0; d < N_SCHEDULES; d++) {
-        if (!strcmp(scheduler, schedule_str[d])) {
-            schedule_found = d;
-                fprintf (stderr, "Found scheduler: %s\n", scheduler);
-
-        }
-    }
-
-    if (schedule_found == -1) {
-        fprintf (stderr, "Invalid scheduler! using DEFAULT\n");
-        schedule_found = DEFAULT;
-    }
-
-    schedule_t schedule = (schedule_t)schedule_found;
-    
-    fprintf (stderr, "Creating context\n");
-    sd_ctx_t* sd_ctx = new_sd_ctx(model,
-                                  clip_l_path,
-                                  clip_g_path,
-                                  t5xxl_path,
-                                  stableDiffusionModel,
-                                  vae_path,
-                                  "",
-                                  "",
-                                  "",
-                                  "",
-                                  "",
-                                  false,
-                                  false,
-                                  false,
-                                  threads,
-                                  SD_TYPE_COUNT,
-                                  STD_DEFAULT_RNG,
-                                  schedule,
-                                  false,
-                                  false,
-                                  false,
-                                  false);
-
-    if (sd_ctx == NULL) {
-        fprintf (stderr, "failed loading model (generic error)\n");
-        return 1;
-    }
-    fprintf (stderr, "Created context: OK\n");
-
-    sd_c = sd_ctx;
-
-    return 0;
-}
-
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed , char *dst, float cfg_scale) {
-
-    sd_image_t* results;
-
-    std::vector<int> skip_layers = {7, 8, 9};
-
-    fprintf (stderr, "Generating image\n");
-
-    results = txt2img(sd_c,
-                            text,
-                            negativeText,
-                            -1, //clip_skip
-                            cfg_scale, // sfg_scale
-                            3.5f,
-                            width,
-                            height,
-                            sample_method, 
-                            steps,
-                            seed,
-                            1,
-                            NULL,
-                            0.9f,
-                            20.f,
-                            false,
-                            "",
-                            skip_layers.data(),
-                            skip_layers.size(),
-                            0,
-                            0.01,
-                            0.2);
-
-    if (results == NULL) {
-        fprintf (stderr, "NO results\n");
-        return 1;
-    }
-
-    if (results[0].data == NULL) {
-        fprintf (stderr, "Results with no data\n");
-        return 1;
-    }
-
-    fprintf (stderr, "Writing PNG\n");
-
-    fprintf (stderr, "DST: %s\n", dst);
-    fprintf (stderr, "Width: %d\n", results[0].width);
-    fprintf (stderr, "Height: %d\n", results[0].height);
-    fprintf (stderr, "Channel: %d\n", results[0].channel);
-    fprintf (stderr, "Data: %p\n", results[0].data);
-
-    stbi_write_png(dst, results[0].width, results[0].height, results[0].channel,
-                       results[0].data, 0, NULL);
-    fprintf (stderr, "Saved resulting image to '%s'\n", dst);
-
-    // TODO: free results. Why does it crash?
-
-    free(results[0].data);
-    results[0].data = NULL;
-    free(results);
-    fprintf (stderr, "gen_image is done", dst);
-
-    return 0;
-}
-
-int unload() {
-    free_sd_ctx(sd_c);
-}
-
--- a/backend/go/image/stablediffusion-ggml/gosd.go
+++ b/backend/go/image/stablediffusion-ggml/gosd.go
@@ -1,96 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/ggml/include
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/build/ggml/src/ggml-cpu -L${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/build/ggml/src -lsd -lstdc++ -lm -lggml -lggml-base -lggml-cpu -lgomp
-// #include <gosd.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"strings"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-	"github.com/mudler/LocalAI/pkg/utils"
-)
-
-type SDGGML struct {
-	base.SingleThread
-	threads      int
-	sampleMethod string
-	cfgScale     float32
-}
-
-func (sd *SDGGML) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	var options **C.char
-	// prepare the options array to pass to C
-
-	size := C.size_t(unsafe.Sizeof((*C.char)(nil)))
-	length := C.size_t(len(opts.Options))
-	options = (**C.char)(C.malloc(length * size))
-	view := (*[1 << 30]*C.char)(unsafe.Pointer(options))[0:len(opts.Options):len(opts.Options)]
-
-	var diffusionModel int
-
-	var oo []string
-	for _, op := range opts.Options {
-		if op == "diffusion_model" {
-			diffusionModel = 1
-			continue
-		}
-
-		// If it's an option path, we resolve absolute path from the model path
-		if strings.Contains(op, ":") && strings.Contains(op, "path") {
-			data := strings.Split(op, ":")
-			data[1] = filepath.Join(opts.ModelPath, data[1])
-			if err := utils.VerifyPath(data[1], opts.ModelPath); err == nil {
-				oo = append(oo, strings.Join(data, ":"))
-			}
-		} else {
-			oo = append(oo, op)
-		}
-	}
-
-	fmt.Fprintf(os.Stderr, "Options: %+v\n", oo)
-
-	for i, x := range oo {
-		view[i] = C.CString(x)
-	}
-
-	sd.cfgScale = opts.CFGScale
-
-	ret := C.load_model(modelFile, options, C.int(opts.Threads), C.int(diffusionModel))
-	if ret != 0 {
-		return fmt.Errorf("could not load model")
-	}
-
-	return nil
-}
-
-func (sd *SDGGML) GenerateImage(opts *pb.GenerateImageRequest) error {
-	t := C.CString(opts.PositivePrompt)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	negative := C.CString(opts.NegativePrompt)
-	defer C.free(unsafe.Pointer(negative))
-
-	ret := C.gen_image(t, negative, C.int(opts.Width), C.int(opts.Height), C.int(opts.Step), C.int(opts.Seed), dst, C.float(sd.cfgScale))
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
--- a/backend/go/image/stablediffusion-ggml/gosd.h
+++ b/backend/go/image/stablediffusion-ggml/gosd.h
@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model, char* options[], int threads, int diffusionModel);
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed, char *dst, float cfg_scale);
-#ifdef __cplusplus
-}
-#endif
--- a/backend/go/image/stablediffusion-ggml/main.go
+++ b/backend/go/image/stablediffusion-ggml/main.go
@@ -1,20 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-import (
-	"flag"
-
-	grpc "github.com/mudler/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &SDGGML{}); err != nil {
-		panic(err)
-	}
-}
--- a/backend/go/llm/bert/bert.go
+++ b/backend/go/llm/bert/bert.go
@@ -0,0 +1,34 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	bert "github.com/go-skynet/go-bert.cpp"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+)
+
+type Embeddings struct {
+	base.SingleThread
+	bert *bert.Bert
+}
+
+func (llm *Embeddings) Load(opts *pb.ModelOptions) error {
+	model, err := bert.New(opts.ModelFile)
+	llm.bert = model
+	return err
+}
+
+func (llm *Embeddings) Embeddings(opts *pb.PredictOptions) ([]float32, error) {
+
+	if len(opts.EmbeddingTokens) > 0 {
+		tokens := []int{}
+		for _, t := range opts.EmbeddingTokens {
+			tokens = append(tokens, int(t))
+		}
+		return llm.bert.TokenEmbeddings(tokens, bert.SetThreads(int(opts.Threads)))
+	}
+
+	return llm.bert.Embeddings(opts.Embeddings, bert.SetThreads(int(opts.Threads)))
+}
--- a/backend/go/llm/bert/main.go
+++ b/backend/go/llm/bert/main.go
@@ -1,6 +1,7 @@
 package main

 // Note: this is started internally by LocalAI and a server is allocated for each model
+
 import (
 	"flag"

@@ -14,7 +15,7 @@ var (
 func main() {
 	flag.Parse()

-	if err := grpc.StartServer(*addr, &Bark{}); err != nil {
+	if err := grpc.StartServer(*addr, &Embeddings{}); err != nil {
 		panic(err)
 	}
 }
--- a/backend/python/autogptq/requirements.txt
+++ b/backend/python/autogptq/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
 transformers
--- a/backend/python/bark/requirements.txt
+++ b/backend/python/bark/requirements.txt
@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
--- a/backend/python/common/template/requirements.txt
+++ b/backend/python/common/template/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 grpcio-tools
--- a/backend/python/coqui/requirements.txt
+++ b/backend/python/coqui/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
 packaging==24.1
--- a/backend/python/diffusers/requirements.txt
+++ b/backend/python/diffusers/requirements.txt
@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.68.1
+grpcio==1.68.0
 pillow
 protobuf
 certifi
--- a/backend/python/exllama2/requirements.txt
+++ b/backend/python/exllama2/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
 wheel
--- a/backend/python/mamba/requirements.txt
+++ b/backend/python/mamba/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
--- a/backend/python/openvoice/requirements-intel.txt
+++ b/backend/python/openvoice/requirements-intel.txt
@@ -2,7 +2,7 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 librosa==0.9.1
 faster-whisper==0.9.0
--- a/backend/python/openvoice/requirements.txt
+++ b/backend/python/openvoice/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 librosa
 faster-whisper
--- a/backend/python/parler-tts/requirements.txt
+++ b/backend/python/parler-tts/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.1
+grpcio==1.68.0
 certifi
 llvmlite==0.43.0
--- a/backend/python/rerankers/requirements.txt
+++ b/backend/python/rerankers/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
--- a/backend/python/sentencetransformers/requirements.txt
+++ b/backend/python/sentencetransformers/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
 datasets
--- a/backend/python/transformers-musicgen/requirements.txt
+++ b/backend/python/transformers-musicgen/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 scipy==1.14.0
 certifi
--- a/backend/python/transformers/requirements.txt
+++ b/backend/python/transformers/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/vall-e-x/requirements.txt
+++ b/backend/python/vall-e-x/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
--- a/backend/python/vllm/install.sh
+++ b/backend/python/vllm/install.sh
@@ -22,7 +22,7 @@ if [ "x${BUILD_TYPE}" == "x" ] && [ "x${FROM_SOURCE}" == "xtrue" ]; then
            git clone https://github.com/vllm-project/vllm
        fi
        pushd vllm
-            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.68.1 protobuf bitsandbytes
+            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.68.0 protobuf bitsandbytes
            uv pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
            VLLM_TARGET_DEVICE=cpu python setup.py install
        popd
--- a/backend/python/vllm/requirements.txt
+++ b/backend/python/vllm/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.1
+grpcio==1.68.0
 protobuf
 certifi
 setuptools
--- a/core/backend/options.go
+++ b/core/backend/options.go
@@ -122,7 +122,7 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		CUDA:                 c.CUDA || c.Diffusers.CUDA,
 		SchedulerType:        c.Diffusers.SchedulerType,
 		PipelineType:         c.Diffusers.PipelineType,
-		CFGScale:             c.CFGScale,
+		CFGScale:             c.Diffusers.CFGScale,
 		LoraAdapter:          c.LoraAdapter,
 		LoraScale:            c.LoraScale,
 		LoraAdapters:         c.LoraAdapters,
@@ -132,7 +132,6 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		IMG2IMG:              c.Diffusers.IMG2IMG,
 		CLIPModel:            c.Diffusers.ClipModel,
 		CLIPSubfolder:        c.Diffusers.ClipSubFolder,
-		Options:              c.Options,
 		CLIPSkip:             int32(c.Diffusers.ClipSkip),
 		ControlNet:           c.Diffusers.ControlNet,
 		ContextSize:          int32(ctxSize),
--- a/core/config/backend_config.go
+++ b/core/config/backend_config.go
@@ -72,8 +72,6 @@ type BackendConfig struct {

 	Description string `yaml:"description"`
 	Usage       string `yaml:"usage"`
-
-	Options []string `yaml:"options"`
 }

 type File struct {
@@ -99,15 +97,16 @@ type GRPC struct {
 }

 type Diffusers struct {
-	CUDA             bool   `yaml:"cuda"`
-	PipelineType     string `yaml:"pipeline_type"`
-	SchedulerType    string `yaml:"scheduler_type"`
-	EnableParameters string `yaml:"enable_parameters"` // A list of comma separated parameters to specify
-	IMG2IMG          bool   `yaml:"img2img"`           // Image to Image Diffuser
-	ClipSkip         int    `yaml:"clip_skip"`         // Skip every N frames
-	ClipModel        string `yaml:"clip_model"`        // Clip model to use
-	ClipSubFolder    string `yaml:"clip_subfolder"`    // Subfolder to use for clip model
-	ControlNet       string `yaml:"control_net"`
+	CUDA             bool    `yaml:"cuda"`
+	PipelineType     string  `yaml:"pipeline_type"`
+	SchedulerType    string  `yaml:"scheduler_type"`
+	EnableParameters string  `yaml:"enable_parameters"` // A list of comma separated parameters to specify
+	CFGScale         float32 `yaml:"cfg_scale"`         // Classifier-Free Guidance Scale
+	IMG2IMG          bool    `yaml:"img2img"`           // Image to Image Diffuser
+	ClipSkip         int     `yaml:"clip_skip"`         // Skip every N frames
+	ClipModel        string  `yaml:"clip_model"`        // Clip model to use
+	ClipSubFolder    string  `yaml:"clip_subfolder"`    // Subfolder to use for clip model
+	ControlNet       string  `yaml:"control_net"`
 }

 // LLMConfig is a struct that holds the configuration that are
@@ -165,8 +164,6 @@ type LLMConfig struct {
 	YarnAttnFactor float32 `yaml:"yarn_attn_factor"`
 	YarnBetaFast   float32 `yaml:"yarn_beta_fast"`
 	YarnBetaSlow   float32 `yaml:"yarn_beta_slow"`
-
-	CFGScale float32 `yaml:"cfg_scale"` // Classifier-Free Guidance Scale
 }

 // AutoGPTQ is a struct that holds the configuration specific to the AutoGPTQ backend
--- a/core/gallery/models_test.go
+++ b/core/gallery/models_test.go
@@ -12,8 +12,6 @@ import (
 	"gopkg.in/yaml.v3"
 )

-const bertEmbeddingsURL = `https://gist.githubusercontent.com/mudler/0a080b166b87640e8644b09c2aee6e3b/raw/f0e8c26bb72edc16d9fbafbfd6638072126ff225/bert-embeddings-gallery.yaml`
-
 var _ = Describe("Model test", func() {

 	Context("Downloading", func() {
@@ -49,7 +47,7 @@ var _ = Describe("Model test", func() {

 			gallery := []GalleryModel{{
 				Name: "bert",
-				URL:  bertEmbeddingsURL,
+				URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
 			}}
 			out, err := yaml.Marshal(gallery)
 			Expect(err).ToNot(HaveOccurred())
@@ -68,7 +66,7 @@ var _ = Describe("Model test", func() {
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(models)).To(Equal(1))
 			Expect(models[0].Name).To(Equal("bert"))
-			Expect(models[0].URL).To(Equal(bertEmbeddingsURL))
+			Expect(models[0].URL).To(Equal("https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml"))
 			Expect(models[0].Installed).To(BeFalse())

 			err = InstallModelFromGallery(galleries, "test@bert", tempdir, GalleryModel{}, func(s1, s2, s3 string, f float64) {}, true)
@@ -80,7 +78,7 @@ var _ = Describe("Model test", func() {
 			content := map[string]interface{}{}
 			err = yaml.Unmarshal(dat, &content)
 			Expect(err).ToNot(HaveOccurred())
-			Expect(content["usage"]).To(ContainSubstring("You can test this model with curl like this"))
+			Expect(content["backend"]).To(Equal("bert-embeddings"))

 			models, err = AvailableGalleryModels(galleries, tempdir)
 			Expect(err).ToNot(HaveOccurred())
--- a/core/http/app_test.go
+++ b/core/http/app_test.go
@@ -5,12 +5,14 @@ import (
 	"context"
 	"embed"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"net/http"
 	"os"
 	"path/filepath"
 	"runtime"
+	"strings"

 	"github.com/mudler/LocalAI/core/config"
 	. "github.com/mudler/LocalAI/core/http"
@@ -238,8 +240,6 @@ func postInvalidRequest(url string) (error, int) {
 	return nil, resp.StatusCode
 }

-const bertEmbeddingsURL = `https://gist.githubusercontent.com/mudler/0a080b166b87640e8644b09c2aee6e3b/raw/f0e8c26bb72edc16d9fbafbfd6638072126ff225/bert-embeddings-gallery.yaml`
-
 //go:embed backend-assets/*
 var backendAssets embed.FS

@@ -279,13 +279,13 @@ var _ = Describe("API test", func() {
 			g := []gallery.GalleryModel{
 				{
 					Name: "bert",
-					URL:  bertEmbeddingsURL,
+					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
 				},
 				{
 					Name:            "bert2",
-					URL:             bertEmbeddingsURL,
+					URL:             "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
 					Overrides:       map[string]interface{}{"foo": "bar"},
-					AdditionalFiles: []gallery.File{{Filename: "foo.yaml", URI: bertEmbeddingsURL}},
+					AdditionalFiles: []gallery.File{{Filename: "foo.yaml", URI: "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml"}},
 				},
 			}
 			out, err := yaml.Marshal(g)
@@ -383,7 +383,7 @@ var _ = Describe("API test", func() {
 				content := map[string]interface{}{}
 				err = yaml.Unmarshal(dat, &content)
 				Expect(err).ToNot(HaveOccurred())
-				Expect(content["usage"]).To(ContainSubstring("You can test this model with curl like this"))
+				Expect(content["backend"]).To(Equal("bert-embeddings"))
 				Expect(content["foo"]).To(Equal("bar"))

 				models, err = getModels("http://127.0.0.1:9090/models/available")
@@ -402,7 +402,7 @@ var _ = Describe("API test", func() {
 			It("overrides models", func() {

 				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
-					URL:  bertEmbeddingsURL,
+					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
 					Name: "bert",
 					Overrides: map[string]interface{}{
 						"backend": "llama",
@@ -451,7 +451,7 @@ var _ = Describe("API test", func() {
 			})
 			It("apply models without overrides", func() {
 				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
-					URL:       bertEmbeddingsURL,
+					URL:       "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
 					Name:      "bert",
 					Overrides: map[string]interface{}{},
 				})
@@ -471,7 +471,7 @@ var _ = Describe("API test", func() {
 				content := map[string]interface{}{}
 				err = yaml.Unmarshal(dat, &content)
 				Expect(err).ToNot(HaveOccurred())
-				Expect(content["usage"]).To(ContainSubstring("You can test this model with curl like this"))
+				Expect(content["backend"]).To(Equal("bert-embeddings"))
 			})

 			It("runs openllama(llama-ggml backend)", Label("llama"), func() {
@@ -806,7 +806,7 @@ var _ = Describe("API test", func() {
 		It("returns the models list", func() {
 			models, err := client.ListModels(context.TODO())
 			Expect(err).ToNot(HaveOccurred())
-			Expect(len(models.Models)).To(Equal(7)) // If "config.yaml" should be included, this should be 8?
+			Expect(len(models.Models)).To(Equal(6)) // If "config.yaml" should be included, this should be 8?
 		})
 		It("can generate completions via ggml", func() {
 			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel.ggml", Prompt: testPrompt})
@@ -866,8 +866,8 @@ var _ = Describe("API test", func() {
 				},
 			)
 			Expect(err).ToNot(HaveOccurred(), err)
-			Expect(len(resp.Data[0].Embedding)).To(BeNumerically("==", 2048))
-			Expect(len(resp.Data[1].Embedding)).To(BeNumerically("==", 2048))
+			Expect(len(resp.Data[0].Embedding)).To(BeNumerically("==", 384))
+			Expect(len(resp.Data[1].Embedding)).To(BeNumerically("==", 384))

 			sunEmbedding := resp.Data[0].Embedding
 			resp2, err := client.CreateEmbeddings(
@@ -911,6 +911,71 @@ var _ = Describe("API test", func() {
 			})
 		})

+		Context("backends", func() {
+			It("runs rwkv completion", func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "rwkv_test", Prompt: "Count up to five: one, two, three, four,"})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices) > 0).To(BeTrue())
+				Expect(resp.Choices[0].Text).To(ContainSubstring("five"))
+
+				stream, err := client.CreateCompletionStream(context.TODO(), openai.CompletionRequest{
+					Model: "rwkv_test", Prompt: "Count up to five: one, two, three, four,", Stream: true,
+				})
+				Expect(err).ToNot(HaveOccurred())
+				defer stream.Close()
+
+				tokens := 0
+				text := ""
+				for {
+					response, err := stream.Recv()
+					if errors.Is(err, io.EOF) {
+						break
+					}
+
+					Expect(err).ToNot(HaveOccurred())
+					text += response.Choices[0].Text
+					tokens++
+				}
+				Expect(text).ToNot(BeEmpty())
+				Expect(text).To(ContainSubstring("five"))
+				Expect(tokens).ToNot(Or(Equal(1), Equal(0)))
+			})
+			It("runs rwkv chat completion", func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				resp, err := client.CreateChatCompletion(context.TODO(),
+					openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices) > 0).To(BeTrue())
+				Expect(strings.ToLower(resp.Choices[0].Message.Content)).To(Or(ContainSubstring("sure"), ContainSubstring("five")))
+
+				stream, err := client.CreateChatCompletionStream(context.TODO(), openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
+				Expect(err).ToNot(HaveOccurred())
+				defer stream.Close()
+
+				tokens := 0
+				text := ""
+				for {
+					response, err := stream.Recv()
+					if errors.Is(err, io.EOF) {
+						break
+					}
+
+					Expect(err).ToNot(HaveOccurred())
+					text += response.Choices[0].Delta.Content
+					tokens++
+				}
+				Expect(text).ToNot(BeEmpty())
+				Expect(strings.ToLower(text)).To(Or(ContainSubstring("sure"), ContainSubstring("five")))
+
+				Expect(tokens).ToNot(Or(Equal(1), Equal(0)))
+			})
+		})
+
 		// See tests/integration/stores_test
 		Context("Stores", Label("stores"), func() {

--- a/docs/content/docs/features/embeddings.md
+++ b/docs/content/docs/features/embeddings.md
@@ -27,6 +27,39 @@ embeddings: true
 # .. other parameters
 ```

+## Bert embeddings
+
+To use `bert.cpp` models you can use the `bert` embedding backend.
+
+An example model config file:
+
+```yaml
+name: text-embedding-ada-002
+parameters:
+  model: bert
+backend: bert-embeddings
+embeddings: true
+# .. other parameters
+```
+
+The `bert` backend uses [bert.cpp](https://github.com/skeskinen/bert.cpp) and uses `ggml` models.
+
+For instance you can download the `ggml` quantized version of `all-MiniLM-L6-v2` from https://huggingface.co/skeskinen/ggml:
+
+```bash
+wget https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-q4_0.bin -O models/bert
+```
+
+To test locally (LocalAI server running on `localhost`),
+you can use `curl` (and `jq` at the end to prettify):
+
+```bash
+curl http://localhost:8080/embeddings -X POST -H "Content-Type: application/json" -d '{
+  "input": "Your text string goes here",
+  "model": "text-embedding-ada-002"
+}' | jq "."
+```
+
 ## Huggingface embeddings

 To use `sentence-transformers` and models in `huggingface` you can use the `sentencetransformers` embedding backend.
@@ -54,26 +87,17 @@ The `sentencetransformers` backend uses Python [sentence-transformers](https://g

 ## Llama.cpp embeddings

-Embeddings with `llama.cpp` are supported with the `llama-cpp` backend, it needs to be enabled with `embeddings` set to `true`.
+Embeddings with `llama.cpp` are supported with the `llama` backend.

 ```yaml
 name: my-awesome-model
-backend: llama-cpp
+backend: llama
 embeddings: true
 parameters:
  model: ggml-file.bin
 # ...
 ```

-Then you can use the API to generate embeddings:
-
-```bash
-curl http://localhost:8080/embeddings -X POST -H "Content-Type: application/json" -d '{
-  "input": "My text",
-  "model": "my-awesome-model"
-}' | jq "."
-```
-
 ## 💡 Examples

 - Example that uses LLamaIndex and LocalAI as embedding: [here](https://github.com/go-skynet/LocalAI/tree/master/examples/query_data/).
--- a/docs/content/docs/features/image-generation.md
+++ b/docs/content/docs/features/image-generation.md
@@ -194,9 +194,8 @@ diffusers:
  pipeline_type: StableDiffusionPipeline
  enable_parameters: "negative_prompt,num_inference_steps,clip_skip"
  scheduler_type: "k_dpmpp_sde"
+  cfg_scale: 8
  clip_skip: 11
-
-cfg_scale: 8
 ```

 #### Configuration parameters
@@ -303,8 +302,7 @@ cuda: true
 diffusers:
  pipeline_type: StableDiffusionDepth2ImgPipeline
  enable_parameters: "negative_prompt,num_inference_steps,image"
-
-cfg_scale: 6
+  cfg_scale: 6
 ```

 ```bash
--- a/docs/content/docs/features/model-gallery.md
+++ b/docs/content/docs/features/model-gallery.md
@@ -300,7 +300,7 @@ curl $LOCALAI/models/apply -H "Content-Type: application/json" -d '{

 ```bash
 curl $LOCALAI/models/apply -H "Content-Type: application/json" -d '{
-     "id": "bert-embeddings",
+     "url": "github:mudler/LocalAI/gallery/bert-embeddings.yaml",
     "name": "text-embedding-ada-002"
   }'  
 ```
--- a/docs/themes/hugo-theme-relearn
+++ b/docs/themes/hugo-theme-relearn
--- a/embedded/models/bert-cpp.yaml
+++ b/embedded/models/bert-cpp.yaml
@@ -0,0 +1,23 @@
+backend: bert-embeddings
+embeddings: true
+f16: true
+
+gpu_layers: 90
+mmap: true
+name: bert-cpp-minilm-v6
+
+parameters:
+  model: bert-MiniLM-L6-v2q4_0.bin
+
+download_files:
+- filename: "bert-MiniLM-L6-v2q4_0.bin"
+  sha256: "a5a174d8772c8a569faf9f3136c441f2c3855b5bf35ed32274294219533feaad"
+  uri: "https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin"
+
+usage: |
+    You can test this model with curl like this:
+
+    curl http://localhost:8080/embeddings -X POST -H "Content-Type: application/json" -d '{
+      "input": "Your text string goes here",
+      "model": "bert-cpp-minilm-v6"
+    }'
--- a/gallery/bert-embeddings.yaml
+++ b/gallery/bert-embeddings.yaml
@@ -0,0 +1,12 @@
+---
+name: "bert-embeddings"
+
+config_file: |
+    parameters:
+      model: bert-MiniLM-L6-v2q4_0.bin
+    backend: bert-embeddings
+    embeddings: true
+files:
+  - filename: "bert-MiniLM-L6-v2q4_0.bin"
+    sha256: "a5a174d8772c8a569faf9f3136c441f2c3855b5bf35ed32274294219533feaad"
+    uri: "https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin"
--- a/gallery/flux-ggml.yaml
+++ b/gallery/flux-ggml.yaml
@@ -1,12 +0,0 @@
---
-name: "flux-ggml"
-
-config_file: |
-    backend: stablediffusion-ggml
-    step: 25
-    options:
-    - "diffusion_model"
-    - "clip_l_path:clip_l.safetensors"
-    - "t5xxl_path:t5xxl_fp16.safetensors"
-    - "vae_path:ae.safetensors"
-    - "sampler:euler"
--- a/gallery/flux.yaml
+++ b/gallery/flux.yaml
@@ -11,5 +11,4 @@ config_file: |
    cuda: true
    enable_parameters: num_inference_steps
    pipeline_type: FluxPipeline
-
-  cfg_scale: 0
+    cfg_scale: 0
--- a/gallery/index.yaml
+++ b/gallery/index.yaml
@@ -1,27 +1,4 @@
 ---
- &rwkv
-  url: "github:mudler/LocalAI/gallery/rwkv.yaml@master"
-  name: "rwkv-6-world-7b"
-  license: apache-2.0
-  urls:
-    - https://huggingface.co/RWKV/rwkv-6-world-7b
-    - https://huggingface.co/bartowski/rwkv-6-world-7b-GGUF
-  tags:
-    - llm
-    - rwkv
-    - cpu
-    - gpu
-    - rnn
-  description: |
-    RWKV (pronounced RwaKuv) is an RNN with GPT-level LLM performance, and can also be directly trained like a GPT transformer (parallelizable). We are at RWKV-7.
-    So it's combining the best of RNN and transformer - great performance, fast inference, fast training, saves VRAM, "infinite" ctxlen, and free text embedding. Moreover it's 100% attention-free, and a Linux Foundation AI project.
-  overrides:
-    parameters:
-      model: rwkv-6-world-7b-Q4_K_M.gguf
-  files:
-    - filename: rwkv-6-world-7b-Q4_K_M.gguf
-      sha256: f74574186fa4584f405e92198605680db6ad00fd77974ffa14bf02073bb90273
-      uri: huggingface://bartowski/rwkv-6-world-7b-GGUF/rwkv-6-world-7b-Q4_K_M.gguf
 - &qwen25coder
  name: "qwen2.5-coder-14b"
  url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
@@ -380,7 +357,6 @@
  urls:
    - https://huggingface.co/hugging-quants/Llama-3.2-1B-Instruct-Q4_K_M-GGUF
  overrides:
-    embeddings: true
    parameters:
      model: llama-3.2-1b-instruct-q4_k_m.gguf
  files:
@@ -704,31 +680,6 @@
    - filename: Llama-Sentient-3.2-3B-Instruct.Q4_K_M.gguf
      uri: huggingface://QuantFactory/Llama-Sentient-3.2-3B-Instruct-GGUF/Llama-Sentient-3.2-3B-Instruct.Q4_K_M.gguf
      sha256: 3f855ce0522bfdc39fc826162ba6d89f15cc3740c5207da10e70baa3348b7812
- !!merge <<: *llama32
-  name: "llama-smoltalk-3.2-1b-instruct"
-  urls:
-    - https://huggingface.co/prithivMLmods/Llama-SmolTalk-3.2-1B-Instruct
-    - https://huggingface.co/mradermacher/Llama-SmolTalk-3.2-1B-Instruct-GGUF
-  description: |
-    The Llama-SmolTalk-3.2-1B-Instruct model is a lightweight, instruction-tuned model designed for efficient text generation and conversational AI tasks. With a 1B parameter architecture, this model strikes a balance between performance and resource efficiency, making it ideal for applications requiring concise, contextually relevant outputs. The model has been fine-tuned to deliver robust instruction-following capabilities, catering to both structured and open-ended queries.
-    Key Features:
-
-        Instruction-Tuned Performance: Optimized to understand and execute user-provided instructions across diverse domains.
-        Lightweight Architecture: With just 1 billion parameters, the model provides efficient computation and storage without compromising output quality.
-        Versatile Use Cases: Suitable for tasks like content generation, conversational interfaces, and basic problem-solving.
-
-    Intended Applications:
-
-        Conversational AI: Engage users with dynamic and contextually aware dialogue.
-        Content Generation: Produce summaries, explanations, or other creative text outputs efficiently.
-        Instruction Execution: Follow user commands to generate precise and relevant responses.
-  overrides:
-    parameters:
-      model: Llama-SmolTalk-3.2-1B-Instruct.Q4_K_M.gguf
-  files:
-    - filename: Llama-SmolTalk-3.2-1B-Instruct.Q4_K_M.gguf
-      sha256: 03d8d05e3821f4caa65defa82baaff658484d4405b66546431528153ceef4d9e
-      uri: huggingface://mradermacher/Llama-SmolTalk-3.2-1B-Instruct-GGUF/Llama-SmolTalk-3.2-1B-Instruct.Q4_K_M.gguf
 - &qwen25
  ## Qwen2.5
  name: "qwen2.5-14b-instruct"
@@ -1620,177 +1571,6 @@
    - filename: SteyrCannon-0.2-Qwen2.5-72b.Q4_K_M.gguf
      sha256: b34c08b77ffd25ccb0ca50b167f2215e784689205c93a0903fa9435b6cc187f0
      uri: huggingface://mradermacher/SteyrCannon-0.2-Qwen2.5-72b-GGUF/SteyrCannon-0.2-Qwen2.5-72b.Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "q2.5-ms-mistoria-72b-v2"
-  icon: https://cdn-uploads.huggingface.co/production/uploads/64545af5ec40bbbd01242ca6/5LOvUFYiMMw6pcEsOhmo2.webp
-  urls:
-    - https://huggingface.co/Steelskull/Q2.5-MS-Mistoria-72b-v2
-    - https://huggingface.co/bartowski/Q2.5-MS-Mistoria-72b-v2-GGUF
-  description: |
-    This model is my second attempt at a 72b model, as usual, my goal is to merge the robust storytelling of mutiple models while attempting to maintain intelligence.
-    models:
-      - model: EVA-UNIT-01/EVA-Qwen2.5-72B-v0.2
-      - model: ZeusLabs/Chronos-Platinum-72B
-      - model: shuttleai/shuttle-3
-  overrides:
-    parameters:
-      model: Q2.5-MS-Mistoria-72b-v2-Q4_K_M.gguf
-  files:
-    - filename: Q2.5-MS-Mistoria-72b-v2-Q4_K_M.gguf
-      sha256: 33df8aac5a790d1c286fe0fc4f9d340311f282eca19b78db6f7abb845923425c
-      uri: huggingface://bartowski/Q2.5-MS-Mistoria-72b-v2-GGUF/Q2.5-MS-Mistoria-72b-v2-Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "eva-qwen2.5-72b-v0.2"
-  urls:
-    - https://huggingface.co/EVA-UNIT-01/EVA-Qwen2.5-72B-v0.2
-    - https://huggingface.co/bartowski/EVA-Qwen2.5-72B-v0.2-GGUF
-  description: |
-    A RP/storywriting specialist model, full-parameter finetune of Qwen2.5-72B on mixture of synthetic and natural data.
-    It uses Celeste 70B 0.1 data mixture, greatly expanding it to improve versatility, creativity and "flavor" of the resulting model.
-
-    Version notes for 0.2: Optimized training hyperparameters and increased sequence length. Better instruction following deeper into context and less repetition.
-  overrides:
-    parameters:
-      model: EVA-Qwen2.5-72B-v0.2-Q4_K_M.gguf
-  files:
-    - filename: EVA-Qwen2.5-72B-v0.2-Q4_K_M.gguf
-      sha256: 03ea0ecac3ee24a332ca43cf925b669c58714b9754be0f4bc232bd996681ef4b
-      uri: huggingface://bartowski/EVA-Qwen2.5-72B-v0.2-GGUF/EVA-Qwen2.5-72B-v0.2-Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "qwq-32b-preview"
-  urls:
-    - https://huggingface.co/Qwen/QwQ-32B-Preview
-    - https://huggingface.co/bartowski/QwQ-32B-Preview-GGUF
-  description: |
-    QwQ-32B-Preview is an experimental research model developed by the Qwen Team, focused on advancing AI reasoning capabilities. As a preview release, it demonstrates promising analytical abilities while having several important limitations:
-
-    Language Mixing and Code-Switching: The model may mix languages or switch between them unexpectedly, affecting response clarity.
-    Recursive Reasoning Loops: The model may enter circular reasoning patterns, leading to lengthy responses without a conclusive answer.
-    Safety and Ethical Considerations: The model requires enhanced safety measures to ensure reliable and secure performance, and users should exercise caution when deploying it.
-    Performance and Benchmark Limitations: The model excels in math and coding but has room for improvement in other areas, such as common sense reasoning and nuanced language understanding.
-  overrides:
-    parameters:
-      model: QwQ-32B-Preview-Q4_K_M.gguf
-  files:
-    - filename: QwQ-32B-Preview-Q4_K_M.gguf
-      sha256: c499801e682e2379528090c50e106837ca1d69dc3bf3ff3a9af830a0eb49cdf6
-      uri: huggingface://bartowski/QwQ-32B-Preview-GGUF/QwQ-32B-Preview-Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "q2.5-32b-slush-i1"
-  urls:
-    - https://huggingface.co/crestf411/Q2.5-32B-Slush
-    - https://huggingface.co/mradermacher/Q2.5-32B-Slush-i1-GGUF
-  description: |
-    Slush is a two-stage model trained with high LoRA dropout, where stage 1 is a pretraining continuation on the base model, aimed at boosting the model's creativity and writing capabilities. This is then merged into the instruction tune model, and stage 2 is a fine tuning step on top of this to further enhance its roleplaying capabilities and/or to repair any damage caused in the stage 1 merge.
-    This is still early stage. As always, feedback is welcome, and begone if you demand perfection.
-    The second stage, like the Sunfall series, follows the Silly Tavern preset (ChatML), so ymmv in particular if you use some other tool and/or preset.
-  overrides:
-    parameters:
-      model: Q2.5-32B-Slush.i1-Q4_K_M.gguf
-  files:
-    - filename: Q2.5-32B-Slush.i1-Q4_K_M.gguf
-      sha256: 95aecaf43077dabc72d3b556923ede2563325e1c89863800229cfa8b7f1c9659
-      uri: huggingface://mradermacher/Q2.5-32B-Slush-i1-GGUF/Q2.5-32B-Slush.i1-Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "qwestion-24b"
-  urls:
-    - https://huggingface.co/CultriX/Qwestion-14B
-    - https://huggingface.co/mradermacher/Qwestion-24B-GGUF
-  description: |
-    This model was merged using the DARE TIES merge method using Qwen/Qwen2.5-14B as a base.
-    The following models were included in the merge:
-    allknowingroger/Qwenslerp2-14B
-    rombodawg/Rombos-LLM-V2.6-Qwen-14b
-    VAGOsolutions/SauerkrautLM-v2-14b-DPO
-    CultriX/Qwen2.5-14B-Wernicke
-  overrides:
-    parameters:
-      model: Qwestion-24B.Q4_K_M.gguf
-  files:
-    - filename: Qwestion-24B.Q4_K_M.gguf
-      sha256: 5d493bd81cfeef66d80101260145ab1d1d0428ef2191edce62b58391bd0fff0e
-      uri: huggingface://mradermacher/Qwestion-24B-GGUF/Qwestion-24B.Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "teleut-7b"
-  icon: https://cdn-uploads.huggingface.co/production/uploads/634262af8d8089ebaefd410e/UqIi8eztdptvt52Mak_1K.png
-  urls:
-    - https://huggingface.co/allura-org/Teleut-7b
-    - https://huggingface.co/QuantFactory/Teleut-7b-GGUF
-  description: |
-    A replication attempt of Tulu 3 on the Qwen 2.5 base models.
-  overrides:
-    parameters:
-      model: Teleut-7b.Q4_K_M.gguf
-  files:
-    - filename: Teleut-7b.Q4_K_M.gguf
-      sha256: 844a633ea01d793c638e99f2e07413606b3812b759e9264fbaf69c8d94eaa093
-      uri: huggingface://QuantFactory/Teleut-7b-GGUF/Teleut-7b.Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "qwen2.5-7b-homercreative-mix"
-  urls:
-    - https://huggingface.co/ZeroXClem/Qwen2.5-7B-HomerCreative-Mix
-    - https://huggingface.co/QuantFactory/Qwen2.5-7B-HomerCreative-Mix-GGUF
-  description: |
-    ZeroXClem/Qwen2.5-7B-HomerCreative-Mix is an advanced language model meticulously crafted by merging four pre-trained models using the powerful mergekit framework. This fusion leverages the Model Stock merge method to combine the creative prowess of Qandora, the instructive capabilities of Qwen-Instruct-Fusion, the sophisticated blending of HomerSlerp1, and the foundational conversational strengths of Homer-v0.5-Qwen2.5-7B. The resulting model excels in creative text generation, contextual understanding, and dynamic conversational interactions.
-    🚀 Merged Models
-
-    This model merge incorporates the following:
-
-        bunnycore/Qandora-2.5-7B-Creative: Specializes in creative text generation, enhancing the model's ability to produce imaginative and diverse content.
-
-        bunnycore/Qwen2.5-7B-Instruct-Fusion: Focuses on instruction-following capabilities, improving the model's performance in understanding and executing user commands.
-
-        allknowingroger/HomerSlerp1-7B: Utilizes spherical linear interpolation (SLERP) to blend model weights smoothly, ensuring a harmonious integration of different model attributes.
-
-        newsbang/Homer-v0.5-Qwen2.5-7B: Acts as the foundational conversational model, providing robust language comprehension and generation capabilities.
-  overrides:
-    parameters:
-      model: Qwen2.5-7B-HomerCreative-Mix.Q4_K_M.gguf
-  files:
-    - filename: Qwen2.5-7B-HomerCreative-Mix.Q4_K_M.gguf
-      sha256: fc3fdb41e068646592f89a8ae62d7b330f2bd4e97bf615aef2977930977c8ba5
-      uri: huggingface://QuantFactory/Qwen2.5-7B-HomerCreative-Mix-GGUF/Qwen2.5-7B-HomerCreative-Mix.Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "cybercore-qwen-2.1-7b"
-  urls:
-    - https://huggingface.co/bunnycore/CyberCore-Qwen-2.1-7B
-    - https://huggingface.co/QuantFactory/CyberCore-Qwen-2.1-7B-GGUF
-  description: |
-    This model was merged using the TIES merge method using rombodawg/Rombos-LLM-V2.5-Qwen-7b as a base.
-    Models Merged
-    fblgit/cybertron-v4-qw7B-UNAMGS + bunnycore/Qwen-2.1-7b-Persona-lora_model
-    fblgit/cybertron-v4-qw7B-MGS + bunnycore/Qwen-2.1-7b-Persona-lora_model
-  overrides:
-    parameters:
-      model: CyberCore-Qwen-2.1-7B.Q4_K_M.gguf
-  files:
-    - filename: CyberCore-Qwen-2.1-7B.Q4_K_M.gguf
-      sha256: 726042707a4cec29ca0355b4dc7c53a807b307d08aa8a3d4a9e76aefbbbcaadf
-      uri: huggingface://QuantFactory/CyberCore-Qwen-2.1-7B-GGUF/CyberCore-Qwen-2.1-7B.Q4_K_M.gguf
- !!merge <<: *qwen25
-  name: "homercreativeanvita-mix-qw7b"
-  icon: https://huggingface.co/suayptalha/HomerCreativeAnvita-Mix-Qw7B/resolve/main/HomerCreativeAnvita.jpeg
-  urls:
-    - https://huggingface.co/suayptalha/HomerCreativeAnvita-Mix-Qw7B
-    - https://huggingface.co/QuantFactory/HomerCreativeAnvita-Mix-Qw7B-GGUF
-  description: |
-    This model is currently ranked #1 on the Open LLM Leaderboard among models up to 13B parameters!
-    Merge Method
-
-    This model was merged using the SLERP merge method.
-    Models Merged
-
-    The following models were included in the merge:
-
-        ZeroXClem/Qwen2.5-7B-HomerAnvita-NerdMix
-        ZeroXClem/Qwen2.5-7B-HomerCreative-Mix
-  overrides:
-    parameters:
-      model: HomerCreativeAnvita-Mix-Qw7B.Q4_K_M.gguf
-  files:
-    - filename: HomerCreativeAnvita-Mix-Qw7B.Q4_K_M.gguf
-      sha256: a356f279a104bff0bbc2ef7ec136c1e774153de8893bf988083e96fb7f4bc053
-      uri: huggingface://QuantFactory/HomerCreativeAnvita-Mix-Qw7B-GGUF/HomerCreativeAnvita-Mix-Qw7B.Q4_K_M.gguf
 - &archfunct
  license: apache-2.0
  tags:
@@ -3353,73 +3133,6 @@
    - filename: Tulu-3.1-8B-SuperNova.i1-Q4_K_M.gguf
      sha256: c6cc2e1a4c3d2338973ca0050af1cf4462b3f62838f62b4c8a204f2a74eeb01f
      uri: huggingface://mradermacher/Tulu-3.1-8B-SuperNova-i1-GGUF/Tulu-3.1-8B-SuperNova.i1-Q4_K_M.gguf
- !!merge <<: *llama31
-  name: "llama-3.1-tulu-3-70b-dpo"
-  icon: "https://huggingface.co/datasets/allenai/blog-images/resolve/main/tulu3/Tulu3-logo.png"
-  urls:
-    - https://huggingface.co/allenai/Llama-3.1-Tulu-3-70B-DPO
-    - https://huggingface.co/bartowski/Llama-3.1-Tulu-3-70B-DPO-GGUF
-  description: |
-    Tülu3 is a leading instruction following model family, offering fully open-source data, code, and recipes designed to serve as a comprehensive guide for modern post-training techniques. Tülu3 is designed for state-of-the-art performance on a diversity of tasks in addition to chat, such as MATH, GSM8K, and IFEval.
-  overrides:
-    parameters:
-      model: Llama-3.1-Tulu-3-70B-DPO-Q4_K_M.gguf
-  files:
-    - filename: Llama-3.1-Tulu-3-70B-DPO-Q4_K_M.gguf
-      sha256: e2d9c59736274f9dd94f30ef3edcee68fec1d6649eb01d6bad7e3e8a6024f77d
-      uri: huggingface://bartowski/Llama-3.1-Tulu-3-70B-DPO-GGUF/Llama-3.1-Tulu-3-70B-DPO-Q4_K_M.gguf
- !!merge <<: *llama31
-  name: "llama-3.1-tulu-3-8b-sft"
-  icon: "https://huggingface.co/datasets/allenai/blog-images/resolve/main/tulu3/Tulu3-logo.png"
-  urls:
-    - https://huggingface.co/allenai/Llama-3.1-Tulu-3-8B-SFT
-    - https://huggingface.co/bartowski/Llama-3.1-Tulu-3-8B-SFT-GGUF
-  description: |
-    Tülu3 is a leading instruction following model family, offering fully open-source data, code, and recipes designed to serve as a comprehensive guide for modern post-training techniques. Tülu3 is designed for state-of-the-art performance on a diversity of tasks in addition to chat, such as MATH, GSM8K, and IFEval.
-  overrides:
-    parameters:
-      model: Llama-3.1-Tulu-3-8B-SFT-Q4_K_M.gguf
-  files:
-    - filename: Llama-3.1-Tulu-3-8B-SFT-Q4_K_M.gguf
-      sha256: 3fad2c96aa9b9de19c2cda0f88a381c47ac768ca03a95059d9f6c439791f8592
-      uri: huggingface://bartowski/Llama-3.1-Tulu-3-8B-SFT-GGUF/Llama-3.1-Tulu-3-8B-SFT-Q4_K_M.gguf
- !!merge <<: *llama31
-  icon: https://huggingface.co/Skywork/Skywork-o1-Open-Llama-3.1-8B/resolve/main/misc/misc_fig.jpg
-  name: "skywork-o1-open-llama-3.1-8b"
-  urls:
-    - https://huggingface.co/Skywork/Skywork-o1-Open-Llama-3.1-8B
-    - https://huggingface.co/QuantFactory/Skywork-o1-Open-Llama-3.1-8B-GGUF
-  description: |
-    We are excited to announce the release of the Skywork o1 Open model series, developed by the Skywork team at Kunlun Inc. This groundbreaking release introduces a series of models that incorporate o1-like slow thinking and reasoning capabilities. The Skywork o1 Open model series includes three advanced models:
-
-    Skywork o1 Open-Llama-3.1-8B: A robust chat model trained on Llama-3.1-8B, enhanced significantly with "o1-style" data to improve reasoning skills.
-
-    Skywork o1 Open-PRM-Qwen-2.5-1.5B: A specialized model designed to enhance reasoning capability through incremental process rewards, ideal for complex problem solving at a smaller scale.
-
-    Skywork o1 Open-PRM-Qwen-2.5-7B: Extends the capabilities of the 1.5B model by scaling up to handle more demanding reasoning tasks, pushing the boundaries of AI reasoning.
-
-    Different from mere reproductions of the OpenAI o1 model, the Skywork o1 Open model series not only exhibits innate thinking, planning, and reflecting capabilities in its outputs, but also shows significant improvements in reasoning skills on standard benchmarks. This series represents a strategic advancement in AI capabilities, moving a previously weaker base model towards the state-of-the-art (SOTA) in reasoning tasks.
-  overrides:
-    parameters:
-      model: Skywork-o1-Open-Llama-3.1-8B.Q4_K_M.gguf
-  files:
-    - filename: Skywork-o1-Open-Llama-3.1-8B.Q4_K_M.gguf
-      sha256: ef6a203ba585aab14f5d2ec463917a45b3ac571abd89c39e9a96a5e395ea8eea
-      uri: huggingface://QuantFactory/Skywork-o1-Open-Llama-3.1-8B-GGUF/Skywork-o1-Open-Llama-3.1-8B.Q4_K_M.gguf
- !!merge <<: *llama31
-  name: "sparse-llama-3.1-8b-2of4"
-  urls:
-    - https://huggingface.co/QuantFactory/Sparse-Llama-3.1-8B-2of4-GGUF
-    - https://huggingface.co/QuantFactory/Sparse-Llama-3.1-8B-2of4-GGUF
-  description: |
-    This is the 2:4 sparse version of Llama-3.1-8B. On the OpenLLM benchmark (version 1), it achieves an average score of 62.16, compared to 63.19 for the dense model—demonstrating a 98.37% accuracy recovery. On the Mosaic Eval Gauntlet benchmark (version v0.3), it achieves an average score of 53.85, versus 55.34 for the dense model—representing a 97.3% accuracy recovery.
-  overrides:
-    parameters:
-      model: Sparse-Llama-3.1-8B-2of4.Q4_K_M.gguf
-  files:
-    - filename: Sparse-Llama-3.1-8B-2of4.Q4_K_M.gguf
-      sha256: c481e7089ffaedd5ae8c74dccc7fb45f6509640b661fa086ae979f6fefc3fdba
-      uri: huggingface://QuantFactory/Sparse-Llama-3.1-8B-2of4-GGUF/Sparse-Llama-3.1-8B-2of4.Q4_K_M.gguf
 - &deepseek
  ## Deepseek
  url: "github:mudler/LocalAI/gallery/deepseek.yaml@master"
@@ -4213,23 +3926,6 @@
    - filename: magnum-12b-v2.5-kto.i1-Q4_K_M.gguf
      sha256: 07e91d2c6d4e42312e65a69c54f16be467575f7a596fe052993b388e38b90d76
      uri: huggingface://mradermacher/magnum-12b-v2.5-kto-i1-GGUF/magnum-12b-v2.5-kto.i1-Q4_K_M.gguf
- !!merge <<: *mistral03
-  url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
-  name: "chatty-harry_v3.0"
-  icon: https://cdn-uploads.huggingface.co/production/uploads/66c1cc08453a7ef6c5fe657a/0KzNTEtn2kJJQsw4lQeY0.png
-  urls:
-    - https://huggingface.co/Triangle104/Chatty-Harry_V3.0
-    - https://huggingface.co/QuantFactory/Chatty-Harry_V3.0-GGUF
-  description: |
-    This model was merged using the TIES merge method using Triangle104/ChatWaifu_Magnum_V0.2 as a base.
-    The following models were included in the merge: elinas/Chronos-Gold-12B-1.0
-  overrides:
-    parameters:
-      model: Chatty-Harry_V3.0.Q4_K_M.gguf
-  files:
-    - filename: Chatty-Harry_V3.0.Q4_K_M.gguf
-      sha256: 54b63bb74498576ca77b801ed096657a93cc2f6b71d707c3605fdb394bd3e622
-      uri: huggingface://QuantFactory/Chatty-Harry_V3.0-GGUF/Chatty-Harry_V3.0.Q4_K_M.gguf
 - &mudler
  ### START mudler's LocalAI specific-models
  url: "github:mudler/LocalAI/gallery/mudler.yaml@master"
@@ -4985,39 +4681,6 @@
    - filename: G2-9B-Sugarquill-v0.Q4_K_M.gguf
      sha256: 790a2f1541011b2773e22aa863ef78c8662baaa7eca5875e9573007985120187
      uri: huggingface://QuantFactory/G2-9B-Sugarquill-v0-GGUF/G2-9B-Sugarquill-v0.Q4_K_M.gguf
- !!merge <<: *gemma
-  name: "volare-i1"
-  urls:
-    - https://huggingface.co/MoxoffSpA/Volare
-    - https://huggingface.co/mradermacher/Volare-i1-GGUF
-  description: |
-    Volare is an updated version of Gemma7B, specifically fine-tuned with SFT and LoRA adjustments.
-        It's trained on publicly available datasets, like SQUAD-it, and datasets we've created in-house.
-        it's designed to understand and maintain context, making it ideal for Retrieval Augmented Generation (RAG) tasks and applications requiring contextual awareness.
-    Italian dataset.
-  overrides:
-    parameters:
-      model: Volare.i1-Q4_K_M.gguf
-  files:
-    - filename: Volare.i1-Q4_K_M.gguf
-      sha256: fa8fb9d4cb19fcb44be8d53561c9e2840f45aed738de545983ebb158ebba461b
-      uri: huggingface://mradermacher/Volare-i1-GGUF/Volare.i1-Q4_K_M.gguf
- !!merge <<: *gemma
-  name: "bggpt-gemma-2-2.6b-it-v1.0"
-  icon: https://cdn-uploads.huggingface.co/production/uploads/637e1f8cf7e01589cc17bf7e/p6d0YFHjWCQ3S12jWqO1m.png
-  urls:
-    - https://huggingface.co/QuantFactory/BgGPT-Gemma-2-2.6B-IT-v1.0-GGUF
-    - https://huggingface.co/QuantFactory/BgGPT-Gemma-2-2.6B-IT-v1.0-GGUF
-  description: |
-    INSAIT introduces BgGPT-Gemma-2-2.6B-IT-v1.0, a state-of-the-art Bulgarian language model based on google/gemma-2-2b and google/gemma-2-2b-it. BgGPT-Gemma-2-2.6B-IT-v1.0 is free to use and distributed under the Gemma Terms of Use. This model was created by INSAIT, part of Sofia University St. Kliment Ohridski, in Sofia, Bulgaria.
-    The model was built on top of Google’s Gemma 2 2B open models. It was continuously pre-trained on around 100 billion tokens (85 billion in Bulgarian) using the Branch-and-Merge strategy INSAIT presented at EMNLP’24, allowing the model to gain outstanding Bulgarian cultural and linguistic capabilities while retaining its English performance. During the pre-training stage, we use various datasets, including Bulgarian web crawl data, freely available datasets such as Wikipedia, a range of specialized Bulgarian datasets sourced by the INSAIT Institute, and machine translations of popular English datasets. The model was then instruction-fine-tuned on a newly constructed Bulgarian instruction dataset created using real-world conversations. For more information check our blogpost.
-  overrides:
-    parameters:
-      model: BgGPT-Gemma-2-2.6B-IT-v1.0.Q4_K_M.gguf
-  files:
-    - filename: BgGPT-Gemma-2-2.6B-IT-v1.0.Q4_K_M.gguf
-      sha256: 1e92fe80ccad80e97076ee26b002c2280f075dfe2507d534b46a4391a077f319
-      uri: huggingface://QuantFactory/BgGPT-Gemma-2-2.6B-IT-v1.0-GGUF/BgGPT-Gemma-2-2.6B-IT-v1.0.Q4_K_M.gguf
 - &llama3
  url: "github:mudler/LocalAI/gallery/llama3-instruct.yaml@master"
  icon: https://cdn-uploads.huggingface.co/production/uploads/642cc1c253e76b4c2286c58e/aJJxKus1wP5N-euvHEUq7.png
@@ -7643,27 +7306,6 @@
    - filename: LLAMA-3_8B_Unaligned_BETA-Q4_K_M.gguf
      sha256: 5b88fb4537339996c04e4a1b6ef6a2d555c4103b6378e273ae9c6c5e77af67eb
      uri: huggingface://bartowski/LLAMA-3_8B_Unaligned_BETA-GGUF/LLAMA-3_8B_Unaligned_BETA-Q4_K_M.gguf
- !!merge <<: *llama3
-  name: "freyja-v4.95-maldv-7b-non-fiction-i1"
-  urls:
-    - https://huggingface.co/MrRobotoAI/Freyja-v4.95-maldv-7b-NON-FICTION
-    - https://huggingface.co/mradermacher/Freyja-v4.95-maldv-7b-NON-FICTION-i1-GGUF
-  description: |
-    This model was merged using the Model Stock merge method using aifeifei798/llama3-8B-DarkIdol-2.2-Uncensored-1048K as a base.
-    The following models were included in the merge:
-        maldv/llama-3-fantasy-writer-8b
-        maldv/badger-iota-llama-3-8b
-        maldv/badger-lambda-llama-3-8b
-        maldv/badger-mu-llama-3-8b
-        maldv/badger-kappa-llama-3-8b
-        maldv/badger-writer-llama-3-8b
-  overrides:
-    parameters:
-      model: Freyja-v4.95-maldv-7b-NON-FICTION.i1-Q4_K_M.gguf
-  files:
-    - filename: Freyja-v4.95-maldv-7b-NON-FICTION.i1-Q4_K_M.gguf
-      sha256: cdc0f4de6df2ba120835fbd25c2a0ae2af8548f46d2c40c7a018c51c3d19e0c0
-      uri: huggingface://mradermacher/Freyja-v4.95-maldv-7b-NON-FICTION-i1-GGUF/Freyja-v4.95-maldv-7b-NON-FICTION.i1-Q4_K_M.gguf
 - &chatml
  ### ChatML
  url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
@@ -8906,43 +8548,6 @@
  overrides:
    parameters:
      model: black-forest-labs/FLUX.1-schnell
- name: flux.1-dev-ggml
-  license: flux-1-dev-non-commercial-license
-  url: "github:mudler/LocalAI/gallery/flux-ggml.yaml@master"
-  description: |
-    FLUX.1 [dev] is a 12 billion parameter rectified flow transformer capable of generating images from text descriptions. For more information, please read our blog post.
-    Key Features
-        Cutting-edge output quality, second only to our state-of-the-art model FLUX.1 [pro].
-        Competitive prompt following, matching the performance of closed source alternatives .
-        Trained using guidance distillation, making FLUX.1 [dev] more efficient.
-        Open weights to drive new scientific research, and empower artists to develop innovative workflows.
-        Generated outputs can be used for personal, scientific, and commercial purposes as described in the flux-1-dev-non-commercial-license.
-    This model is quantized with GGUF
-  urls:
-    - https://huggingface.co/black-forest-labs/FLUX.1-dev
-    - https://huggingface.co/city96/FLUX.1-dev-gguf
-  tags:
-    - text-to-image
-    - flux
-    - gpu
-    - cpu
-  icon: https://huggingface.co/black-forest-labs/FLUX.1-schnell/resolve/main/schnell_grid.jpeg
-  overrides:
-    parameters:
-      model: flux1-dev-Q2_K.gguf
-  files:
-    - filename: "flux1-dev-Q2_K.gguf"
-      sha256: "b8c464bc0f10076ef8f00ba040d220d90c7993f7c4245ae80227d857f65df105"
-      uri: "huggingface://city96/FLUX.1-dev-gguf/flux1-dev-Q2_K.gguf"
-    - filename: ae.safetensors
-      sha256: afc8e28272cd15db3919bacdb6918ce9c1ed22e96cb12c4d5ed0fba823529e38
-      uri: https://huggingface.co/ChuckMcSneed/FLUX.1-dev/resolve/main/ae.safetensors
-    - filename: clip_l.safetensors
-      sha256: 660c6f5b1abae9dc498ac2d21e1347d2abdb0cf6c0c0c8576cd796491d9a6cdd
-      uri: https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors
-    - filename: t5xxl_fp16.safetensors
-      sha256: 6e480b09fae049a72d2a8c5fbccb8d3e92febeb233bbe9dfe7256958a9167635
-      uri: https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp16.safetensors
 - &whisper
  ## Whisper
  url: "github:mudler/LocalAI/gallery/whisper-base.yaml@master"
@@ -9104,13 +8709,16 @@
    - filename: "ggml-model-whisper-tiny.en-q8_0.bin"
      uri: "https://ggml.ggerganov.com/ggml-model-whisper-tiny.en-q8_0.bin"
      sha256: 5bc2b3860aa151a4c6e7bb095e1fcce7cf12c7b020ca08dcec0c6d018bb7dd94
-## Bert embeddings (llama3.2 drop-in)
- !!merge <<: *llama32
+## Bert embeddings
+- url: "github:mudler/LocalAI/gallery/bert-embeddings.yaml@master"
  name: "bert-embeddings"
-  description: |
-      llama3.2 embeddings model. Using as drop-in replacement for bert-embeddings
+  license: "Apache 2.0"
+  urls:
+    - https://huggingface.co/skeskinen/ggml
  tags:
    - embeddings
+  description: |
+    Bert model that can be used for embeddings
 ## Stable Diffusion
 - url: github:mudler/LocalAI/gallery/stablediffusion.yaml@master
  license: "BSD-3"
@@ -9728,22 +9336,3 @@
    - filename: silero-vad.onnx
      uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
      sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808
- name: "bark-cpp-small"
-  url: github:mudler/LocalAI/gallery/virtual.yaml@master
-  license: mit
-  urls:
-    - https://huggingface.co/suno/bark
-    - https://huggingface.co/Green-Sky/bark-ggml
-  description: |
-    Bark is a transformer-based text-to-audio model created by Suno. Bark can generate highly realistic, multilingual speech as well as other audio - including music, background noise and simple sound effects. The model can also produce nonverbal communications like laughing, sighing and crying. To support the research community, we are providing access to pretrained model checkpoints ready for inference.
-  tags:
-    - tts
-    - cpu
-  overrides:
-    backend: bark-cpp
-    parameters:
-      model: bark-small_weights-f16.bin
-  files:
-    - filename: bark-small_weights-f16.bin
-      uri: https://huggingface.co/Green-Sky/bark-ggml/resolve/main/bark-small_weights-f16.bin
-      sha256: de1ece17e8319537b3a7909baebbd28affab23c942d5d57e648d622af4e2feaa
--- a/gallery/rwkv.yaml
+++ b/gallery/rwkv.yaml
@@ -1,24 +0,0 @@
---
-name: "rwkv"
-
-config_file: |
-    parameters:
-      top_k: 80
-      temperature: 0.9
-      max_tokens: 4098
-      top_p: 0.8
-    context_size: 4098
-
-    roles:
-      user: "User: "
-      system: "System: "
-      assistant: "Assistant: "
-
-    stopwords:
-    - 'Assistant:'
-    - '<s>'
-
-    template:
-      chat: "{{.Input}}\nAssistant: "
-      completion: |
-        {{.Input}}
--- a/gallery/silero-vad.yaml
+++ b/gallery/silero-vad.yaml
@@ -0,0 +1,8 @@
+---
+name: "silero-vad"
+
+config_file: |
+  name: vad
+  backend: silero-vad
+  parameters:
+    model: silero_vad.onnx
--- a/pkg/model/initializers.go
+++ b/pkg/model/initializers.go
@@ -45,6 +45,7 @@ const (

 	LLamaCPPGRPC = "llama-cpp-grpc"

+	BertEmbeddingsBackend  = "bert-embeddings"
 	WhisperBackend         = "whisper"
 	StableDiffusionBackend = "stablediffusion"
 	TinyDreamBackend       = "tinydream"
@@ -153,6 +154,8 @@ func orderBackends(backends map[string][]string) ([]string, error) {
 	toTheEnd := []string{
 		// last has to be huggingface
 		LCHuggingFaceBackend,
+		// then bert embeddings
+		BertEmbeddingsBackend,
 	}

 	// create an ordered map
--- a/scripts/model_gallery_info.py
+++ b/scripts/model_gallery_info.py
@@ -103,10 +103,7 @@ if __name__ == "__main__":
    if readmeFile:
        # If there is a README file, read it
        readme = fs.read_text(readmeFile)
-        try:
-            summarized_readme = summarize(readme)
-        except Exception as e:
-            print(f"Error summarizing the README: {str(e)}", file=sys.stderr)            
+        summarized_readme = summarize(readme)
        summarized_readme = format_description(summarized_readme)

    print("Model correctly processed")
--- a/tests/models_fixtures/embeddings.yaml
+++ b/tests/models_fixtures/embeddings.yaml
@@ -1,4 +1,5 @@
 name: text-embedding-ada-002
-embeddings: true
 parameters:
-  model: huggingface://hugging-quants/Llama-3.2-1B-Instruct-Q4_K_M-GGUF/llama-3.2-1b-instruct-q4_k_m.gguf
+  model: bert
+backend: bert-embeddings
+embeddings: true
--- a/tests/models_fixtures/rwkv.yaml
+++ b/tests/models_fixtures/rwkv.yaml
@@ -14,7 +14,6 @@ roles:

 stopwords:
 - 'Assistant:'
- '<s>'

 template:
  chat: |