chore(stablediffusion-ncn): drop in favor of ggml implementation (#4652)

* chore(stablediffusion-ncn): drop in favor of ggml implementation

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* chore(ci): drop stablediffusion build

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* chore(tests): add

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* chore(tests): try to fixup current tests

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Try to fix tests

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* Tests improvements

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* chore(tests): use quality to specify step

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* chore(tests): switch to sd-1.5

also increase prep time for downloading models

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

.devcontainer/docker-compose-devcontainer.yml

@@ -7,7 +7,7 @@ services:
       args:
       - FFMPEG=true
       - IMAGE_TYPE=extras
-      - GO_TAGS=stablediffusion p2p tts
+      - GO_TAGS=p2p tts
     env_file:
       - ../.env
     ports:

.env

@@ -38,12 +38,12 @@
 ## Uncomment and set to true to enable rebuilding from source
 # REBUILD=true
 
-## Enable go tags, available: stablediffusion, tts
+## Enable go tags, available: p2p, tts
-## stablediffusion: image generation with stablediffusion
+## p2p: enable distributed inferencing
 ## tts: enables text-to-speech with go-piper 
 ## (requires REBUILD=true)
 #
-# GO_TAGS=stablediffusion
+# GO_TAGS=p2p
 
 ## Path where to store generated images
 # LOCALAI_IMAGE_PATH=/tmp/generated/images

.github/workflows/release.yaml

@@ -237,40 +237,7 @@ jobs:
           detached: true
           connect-timeout-seconds: 180
           limit-access-to-actor: true
-  build-stablediffusion:
+
-    runs-on: ubuntu-latest
-    steps:
-      - name: Clone
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-      - uses: actions/setup-go@v5
-        with:
-          go-version: '1.21.x'
-          cache: false
-      - name: Dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends libopencv-dev protobuf-compiler ccache upx-ucl
-          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
-          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
-      - name: Build stablediffusion
-        run: |
-          export PATH=$PATH:$GOPATH/bin
-          make backend-assets/grpc/stablediffusion
-          mkdir -p release && cp backend-assets/grpc/stablediffusion release
-        env:
-          GO_TAGS: stablediffusion
-      - uses: actions/upload-artifact@v4
-        with:
-          name: stablediffusion
-          path: release/
-      - name: Release
-        uses: softprops/action-gh-release@v2
-        if: startsWith(github.ref, 'refs/tags/')
-        with:
-          files: |
-            release/*
 
   build-macOS-x86_64:
     runs-on: macos-13

.github/workflows/test.yml

@@ -105,9 +105,7 @@ jobs:
           # Pre-build piper before we start tests in order to have shared libraries in place
           make sources/go-piper && \
           GO_TAGS="tts" make -C sources/go-piper piper.o && \
-          sudo cp -rfv sources/go-piper/piper-phonemize/pi/lib/. /usr/lib/ && \
+          sudo cp -rfv sources/go-piper/piper-phonemize/pi/lib/. /usr/lib/
-          # Pre-build stable diffusion before we install a newer version of abseil (not compatible with stablediffusion-ncn)
-          PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
         env:
           CUDA_VERSION: 12-4
       - name: Cache grpc
@@ -129,7 +127,7 @@ jobs:
           cd grpc && cd cmake/build && sudo make --jobs 5 install
       - name: Test
         run: |
-          PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" make --jobs 5 --output-sync=target test
+          PATH="$PATH:/root/go/bin" GO_TAGS="tts" make --jobs 5 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
         uses: mxschmitt/action-tmate@v3.19

.vscode/launch.json

@@ -26,7 +26,7 @@
                 "LOCALAI_P2P": "true",
                 "LOCALAI_FEDERATED": "true"
             },
-            "buildFlags": ["-tags", "stablediffusion p2p tts", "-v"],
+            "buildFlags": ["-tags", "p2p tts", "-v"],
             "envFile": "${workspaceFolder}/.env",
             "cwd": "${workspaceRoot}"
         }

Dockerfile

@@ -69,14 +69,10 @@ ENV PATH=/opt/rocm/bin:${PATH}
 # OpenBLAS requirements and stable diffusion
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        libopenblas-dev \
+        libopenblas-dev && \
-        libopencv-dev && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
-# Set up OpenCV
-RUN ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
-
 WORKDIR /build
 
 ###################################
@@ -251,7 +247,7 @@ RUN git clone --recurse-submodules --jobs 4 -b ${GRPC_VERSION} --depth 1 --shall
 
 FROM requirements-drivers AS builder-base
 
-ARG GO_TAGS="stablediffusion tts p2p"
+ARG GO_TAGS="tts p2p"
 ARG GRPC_BACKENDS
 ARG MAKEFLAGS
 ARG LD_FLAGS="-s -w"
@@ -285,35 +281,12 @@ RUN <<EOT bash
     fi
 EOT
 
-
-###################################
-###################################
-
-# This first portion of builder holds the layers specifically used to build backend-assets/grpc/stablediffusion
-# In most cases, builder is the image you should be using - however, this can save build time if one just needs to copy backend-assets/grpc/stablediffusion and nothing else.
-FROM builder-base AS builder-sd
-
-# stablediffusion does not tolerate a newer version of abseil, copy only over enough elements to build it
-COPY Makefile .
-COPY go.mod .
-COPY go.sum .
-COPY backend/backend.proto ./backend/backend.proto
-COPY backend/go/image/stablediffusion ./backend/go/image/stablediffusion
-COPY pkg/grpc ./pkg/grpc
-COPY pkg/stablediffusion ./pkg/stablediffusion
-RUN git init
-RUN make sources/go-stable-diffusion
-RUN touch prepare-sources
-
-# Actually build the backend
-RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make backend-assets/grpc/stablediffusion
-
 ###################################
 ###################################
 
 # The builder target compiles LocalAI. This target is not the target that will be uploaded to the registry.
 # Adjustments to the build process should likely be made here.
-FROM builder-sd AS builder
+FROM builder-base AS builder
 
 # Install the pre-built GRPC
 COPY --from=grpc /opt/grpc /usr/local
@@ -353,8 +326,6 @@ ARG FFMPEG
 
 COPY --from=grpc /opt/grpc /usr/local
 
-COPY --from=builder-sd /build/backend-assets/grpc/stablediffusion /build/backend-assets/grpc/stablediffusion
-
 COPY .devcontainer-scripts /.devcontainer-scripts
 
 # Add FFmpeg
@@ -427,9 +398,6 @@ COPY --from=builder /build/local-ai ./
 # Copy shared libraries for piper
 COPY --from=builder /build/sources/go-piper/piper-phonemize/pi/lib/* /usr/lib/
 
-# do not let stablediffusion rebuild (requires an older version of absl)
-COPY --from=builder-sd /build/backend-assets/grpc/stablediffusion ./backend-assets/grpc/stablediffusion
-
 # Change the shell to bash so we can use [[ tests below
 SHELL ["/bin/bash", "-c"]
 # We try to strike a balance between individual layer size (as that affects total push time) and total image size

Makefile

@@ -18,10 +18,6 @@ WHISPER_CPP_VERSION?=6266a9f9e56a5b925e9892acf650f3eb1245814d
 PIPER_REPO?=https://github.com/mudler/go-piper
 PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0
 
-# stablediffusion version
-STABLEDIFFUSION_REPO?=https://github.com/mudler/go-stable-diffusion
-STABLEDIFFUSION_VERSION?=4a3cd6aeae6f66ee57eae9a0075f8c58c3a6a38f
-
 # bark.cpp
 BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
 BARKCPP_VERSION?=v1.0.0
@@ -179,11 +175,6 @@ ifeq ($(STATIC),true)
 	LD_FLAGS+=-linkmode external -extldflags -static
 endif
 
-ifeq ($(findstring stablediffusion,$(GO_TAGS)),stablediffusion)
-#	OPTIONAL_TARGETS+=go-stable-diffusion/libstablediffusion.a
-	OPTIONAL_GRPC+=backend-assets/grpc/stablediffusion
-endif
-
 ifeq ($(findstring tts,$(GO_TAGS)),tts)
 #	OPTIONAL_TARGETS+=go-piper/libpiper_binding.a
 #	OPTIONAL_TARGETS+=backend-assets/espeak-ng-data
@@ -273,19 +264,6 @@ 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)
-sources/go-stable-diffusion:
-	mkdir -p sources/go-stable-diffusion
-	cd sources/go-stable-diffusion && \
-	git init && \
-	git remote add origin $(STABLEDIFFUSION_REPO) && \
-	git fetch origin && \
-	git checkout $(STABLEDIFFUSION_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
-	CPATH="$(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 && \
@@ -331,20 +309,18 @@ 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 backend/cpp/llama/llama.cpp
+get-sources: sources/go-llama.cpp sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp 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
 	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
-	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(CURDIR)/sources/go-stable-diffusion
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama.cpp
 
 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/mudler/go-piper
-	$(GOCMD) mod edit -dropreplace github.com/mudler/go-stable-diffusion
 	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-llama.cpp
 
 prepare-sources: get-sources replace
@@ -355,7 +331,6 @@ rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
 	$(MAKE) -C sources/go-llama.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
-	$(MAKE) -C sources/go-stable-diffusion clean
 	$(MAKE) -C sources/go-piper clean
 	$(MAKE) build
 
@@ -470,7 +445,7 @@ prepare-test: grpcs
 
 test: prepare test-models/testmodel.ggml grpcs
 	@echo 'Running tests'
-	export GO_TAGS="tts stablediffusion debug"
+	export GO_TAGS="tts debug"
 	$(MAKE) prepare-test
 	HUGGINGFACE_GRPC=$(abspath ./)/backend/python/transformers/run.sh TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
 	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!llama && !llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
@@ -816,13 +791,6 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/piper
 endif
 
-backend-assets/grpc/stablediffusion: sources/go-stable-diffusion sources/go-stable-diffusion/libstablediffusion.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/go-stable-diffusion/:/usr/include/opencv4" LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/stablediffusion
-endif
-
 backend-assets/grpc/silero-vad: backend-assets/grpc backend-assets/lib/libonnxruntime.so.1
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/onnxruntime/include/" LIBRARY_PATH=$(CURDIR)/backend-assets/lib \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/silero-vad ./backend/go/vad/silero

aio/cpu/image-gen.yaml

@@ -1,56 +1,17 @@
 name: stablediffusion
-backend: stablediffusion
+backend: stablediffusion-ggml
+cfg_scale: 4.5
+
+options:
+- sampler:euler
 parameters:
-  model: stablediffusion_assets
+  model: stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf
-
+step: 25
-license: "BSD-3"
-urls:
-- https://github.com/EdVince/Stable-Diffusion-NCNN
-- https://github.com/EdVince/Stable-Diffusion-NCNN/blob/main/LICENSE
-
-description: |
-     Stable Diffusion in NCNN with c++, supported txt2img and img2img
 
 download_files:
-- filename: "stablediffusion_assets/AutoencoderKL-256-256-fp16-opt.param"
+- filename: "stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf"
-  sha256: "18ca4b66685e21406bcf64c484b3b680b4949900415536d599cc876579c85c82"
+  sha256: "b8944e9fe0b69b36ae1b5bb0185b3a7b8ef14347fe0fa9af6c64c4829022261f"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/AutoencoderKL-256-256-fp16-opt.param"
+  uri: "huggingface://second-state/stable-diffusion-v1-5-GGUF/stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf"
-- filename: "stablediffusion_assets/AutoencoderKL-512-512-fp16-opt.param"
-  sha256: "cf45f63aacf3dbbab0f59ed92a6f2c14d9a1801314631cd3abe91e3c85639a20"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/AutoencoderKL-512-512-fp16-opt.param"
-- filename: "stablediffusion_assets/AutoencoderKL-base-fp16.param"
-  sha256: "0254a056dce61b0c27dc9ec1b78b53bcf55315c540f55f051eb841aa992701ba"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/AutoencoderKL-base-fp16.param"
-- filename: "stablediffusion_assets/AutoencoderKL-encoder-512-512-fp16.bin"
-  sha256: "ddcb79a9951b9f91e05e087739ed69da2c1c4ae30ba4168cce350b49d617c9fa"
-  uri: "https://github.com/EdVince/Stable-Diffusion-NCNN/releases/download/naifu/AutoencoderKL-encoder-512-512-fp16.bin"
-- filename: "stablediffusion_assets/AutoencoderKL-fp16.bin"
-  sha256: "f02e71f80e70252734724bbfaed5c4ddd3a8ed7e61bb2175ff5f53099f0e35dd"
-  uri: "https://github.com/EdVince/Stable-Diffusion-NCNN/releases/download/naifu/AutoencoderKL-fp16.bin"
-- filename: "stablediffusion_assets/FrozenCLIPEmbedder-fp16.bin"
-  sha256: "1c9a12f4e1dd1b295a388045f7f28a2352a4d70c3dc96a542189a3dd7051fdd6"
-  uri: "https://github.com/EdVince/Stable-Diffusion-NCNN/releases/download/naifu/FrozenCLIPEmbedder-fp16.bin"
-- filename: "stablediffusion_assets/FrozenCLIPEmbedder-fp16.param"
-  sha256: "471afbe678dd1fd3fe764ef9c6eccaccb0a7d7e601f27b462aa926b20eb368c9"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/FrozenCLIPEmbedder-fp16.param"
-- filename: "stablediffusion_assets/log_sigmas.bin"
-  sha256: "a2089f8aa4c61f9c200feaec541ab3f5c94233b28deb6d5e8bcd974fa79b68ac"
-  uri: "https://github.com/EdVince/Stable-Diffusion-NCNN/raw/main/x86/linux/assets/log_sigmas.bin"
-- filename: "stablediffusion_assets/UNetModel-256-256-MHA-fp16-opt.param"
-  sha256: "a58c380229f09491776df837b7aa7adffc0a87821dc4708b34535da2e36e3da1"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/UNetModel-256-256-MHA-fp16-opt.param"
-- filename: "stablediffusion_assets/UNetModel-512-512-MHA-fp16-opt.param"
-  sha256: "f12034067062827bd7f43d1d21888d1f03905401acf6c6eea22be23c259636fa"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/UNetModel-512-512-MHA-fp16-opt.param"
-- filename: "stablediffusion_assets/UNetModel-base-MHA-fp16.param"
-  sha256: "696f6975de49f4325b53ce32aff81861a6d6c07cd9ce3f0aae2cc405350af38d"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/UNetModel-base-MHA-fp16.param"
-- filename: "stablediffusion_assets/UNetModel-MHA-fp16.bin"
-  sha256: "d618918d011bfc1f644c0f2a33bf84931bd53b28a98492b0a8ed6f3a818852c3"
-  uri: "https://github.com/EdVince/Stable-Diffusion-NCNN/releases/download/naifu/UNetModel-MHA-fp16.bin"
-- filename: "stablediffusion_assets/vocab.txt"
-  sha256: "e30e57b6f1e47616982ef898d8922be24e535b4fa3d0110477b3a6f02ebbae7d"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/vocab.txt"
 
 usage: |
         curl http://localhost:8080/v1/images/generations \

backend/go/image/stablediffusion/main.go

@@ -1,21 +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, &Image{}); err != nil {
-		panic(err)
-	}
-}

backend/go/image/stablediffusion/stablediffusion.go

@@ -1,33 +0,0 @@
-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 (
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-	"github.com/mudler/LocalAI/pkg/stablediffusion"
-)
-
-type Image struct {
-	base.SingleThread
-	stablediffusion *stablediffusion.StableDiffusion
-}
-
-func (image *Image) Load(opts *pb.ModelOptions) error {
-	var err error
-	// Note: the Model here is a path to a directory containing the model files
-	image.stablediffusion, err = stablediffusion.New(opts.ModelFile)
-	return err
-}
-
-func (image *Image) GenerateImage(opts *pb.GenerateImageRequest) error {
-	return image.stablediffusion.GenerateImage(
-		int(opts.Height),
-		int(opts.Width),
-		int(opts.Mode),
-		int(opts.Step),
-		int(opts.Seed),
-		opts.PositivePrompt,
-		opts.NegativePrompt,
-		opts.Dst)
-}

core/config/backend_config.go

@@ -515,7 +515,7 @@ func (c *BackendConfig) GuessUsecases(u BackendConfigUsecases) bool {
 		}
 	}
 	if (u & FLAG_IMAGE) == FLAG_IMAGE {
-		imageBackends := []string{"diffusers", "stablediffusion"}
+		imageBackends := []string{"diffusers", "stablediffusion", "stablediffusion-ggml"}
 		if !slices.Contains(imageBackends, c.Backend) {
 			return false
 		}

core/config/config_test.go

@@ -48,5 +48,66 @@ var _ = Describe("Test cases for config related functions", func() {
 			// config should includes whisper-1 models's api.config
 			Expect(loadedModelNames).To(ContainElements("whisper-1"))
 		})
+
+		It("Test new loadconfig", func() {
+
+			bcl := NewBackendConfigLoader(os.Getenv("MODELS_PATH"))
+			err := bcl.LoadBackendConfigsFromPath(os.Getenv("MODELS_PATH"))
+			Expect(err).To(BeNil())
+			configs := bcl.GetAllBackendConfigs()
+			loadedModelNames := []string{}
+			for _, v := range configs {
+				loadedModelNames = append(loadedModelNames, v.Name)
+			}
+			Expect(configs).ToNot(BeNil())
+			totalModels := len(loadedModelNames)
+
+			Expect(loadedModelNames).To(ContainElements("code-search-ada-code-001"))
+
+			// config should includes text-embedding-ada-002 models's api.config
+			Expect(loadedModelNames).To(ContainElements("text-embedding-ada-002"))
+
+			// config should includes rwkv_test models's api.config
+			Expect(loadedModelNames).To(ContainElements("rwkv_test"))
+
+			// config should includes whisper-1 models's api.config
+			Expect(loadedModelNames).To(ContainElements("whisper-1"))
+
+			// create a temp directory and store a temporary model
+			tmpdir, err := os.MkdirTemp("", "test")
+			Expect(err).ToNot(HaveOccurred())
+			defer os.RemoveAll(tmpdir)
+
+			// create a temporary model
+			model := `name: "test-model"
+description: "test model"
+options:
+- foo
+- bar
+- baz
+`
+			modelFile := tmpdir + "/test-model.yaml"
+			err = os.WriteFile(modelFile, []byte(model), 0644)
+			Expect(err).ToNot(HaveOccurred())
+
+			err = bcl.LoadBackendConfigsFromPath(tmpdir)
+			Expect(err).ToNot(HaveOccurred())
+
+			configs = bcl.GetAllBackendConfigs()
+			Expect(len(configs)).ToNot(Equal(totalModels))
+
+			loadedModelNames = []string{}
+			var testModel BackendConfig
+			for _, v := range configs {
+				loadedModelNames = append(loadedModelNames, v.Name)
+				if v.Name == "test-model" {
+					testModel = v
+				}
+			}
+			Expect(loadedModelNames).To(ContainElements("test-model"))
+			Expect(testModel.Description).To(Equal("test model"))
+			Expect(testModel.Options).To(ContainElements("foo", "bar", "baz"))
+
+		})
 	})
 })

core/http/app_test.go

@@ -687,6 +687,10 @@ var _ = Describe("API test", func() {
 					Name: "model-gallery",
 					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/index.yaml",
 				},
+				{
+					Name: "localai",
+					URL:  "https://raw.githubusercontent.com/mudler/LocalAI/refs/heads/master/gallery/index.yaml",
+				},
 			}
 
 			application, err := application.New(
@@ -764,10 +768,8 @@ var _ = Describe("API test", func() {
 			}
 
 			response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
-				ID: "model-gallery@stablediffusion",
+				ID:   "localai@sd-1.5-ggml",
-				Overrides: map[string]interface{}{
+				Name: "stablediffusion",
-					"parameters": map[string]interface{}{"model": "stablediffusion_assets"},
-				},
 			})
 
 			Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
@@ -778,14 +780,14 @@ var _ = Describe("API test", func() {
 				response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
 				fmt.Println(response)
 				return response["processed"].(bool)
-			}, "360s", "10s").Should(Equal(true))
+			}, "1200s", "10s").Should(Equal(true))
 
 			resp, err := http.Post(
 				"http://127.0.0.1:9090/v1/images/generations",
 				"application/json",
 				bytes.NewBuffer([]byte(`{
-					 			"prompt": "floating hair, portrait, ((loli)), ((one girl)), cute face, hidden hands, asymmetrical bangs, beautiful detailed eyes, eye shadow, hair ornament, ribbons, bowties, buttons, pleated skirt, (((masterpiece))), ((best quality)), colorful|((part of the head)), ((((mutated hands and fingers)))), deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, Octane renderer, lowres, bad anatomy, bad hands, text",
+					 			"prompt": "a lovely cat",
-								"mode": 2,  "seed":9000,
+								"step": 1,  "seed":9000,
 					 			"size": "256x256", "n":2}`)))
 			// The response should contain an URL
 			Expect(err).ToNot(HaveOccurred(), fmt.Sprint(resp))
@@ -794,6 +796,7 @@ var _ = Describe("API test", func() {
 
 			imgUrlResp := &schema.OpenAIResponse{}
 			err = json.Unmarshal(dat, imgUrlResp)
+			Expect(err).ToNot(HaveOccurred(), fmt.Sprint(dat))
 			Expect(imgUrlResp.Data).ToNot(Or(BeNil(), BeZero()))
 			imgUrl := imgUrlResp.Data[0].URL
 			Expect(imgUrl).To(ContainSubstring("http://127.0.0.1:9090/"), imgUrl)

core/http/endpoints/openai/image.go

@@ -72,7 +72,7 @@ func ImageEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appCon
 		}
 
 		if m == "" {
-			m = model.StableDiffusionBackend
+			m = "stablediffusion"
 		}
 		log.Debug().Msgf("Loading model: %+v", m)
 
@@ -129,9 +129,9 @@ func ImageEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appCon
 
 		switch config.Backend {
 		case "stablediffusion":
-			config.Backend = model.StableDiffusionBackend
+			config.Backend = model.StableDiffusionGGMLBackend
 		case "":
-			config.Backend = model.StableDiffusionBackend
+			config.Backend = model.StableDiffusionGGMLBackend
 		}
 
 		if !strings.Contains(input.Size, "x") {

core/http/endpoints/openai/request.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"strconv"
 
 	"github.com/gofiber/fiber/v2"
 	"github.com/google/uuid"
@@ -296,6 +297,14 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 			}
 		}
 	}
+
+	// If a quality was defined as number, convert it to step
+	if input.Quality != "" {
+		q, err := strconv.Atoi(input.Quality)
+		if err == nil {
+			config.Step = q
+		}
+	}
 }
 
 func mergeRequestWithConfig(modelFile string, input *schema.OpenAIRequest, cm *config.BackendConfigLoader, loader *model.ModelLoader, debug bool, threads, ctx int, f16 bool) (*config.BackendConfig, *schema.OpenAIRequest, error) {

core/schema/openai.go

@@ -191,8 +191,9 @@ type OpenAIRequest struct {
 	Stream bool `json:"stream"`
 
 	// Image (not supported by OpenAI)
-	Mode int `json:"mode"`
+	Mode    int    `json:"mode"`
-	Step int `json:"step"`
+	Quality string `json:"quality"`
+	Step    int    `json:"step"`
 
 	// A grammar to constrain the LLM output
 	Grammar string `json:"grammar" yaml:"grammar"`

pkg/model/initializers.go

@@ -29,6 +29,7 @@ var Aliases map[string]string = map[string]string{
 	"langchain-huggingface":  LCHuggingFaceBackend,
 	"transformers-musicgen":  TransformersBackend,
 	"sentencetransformers":   TransformersBackend,
+	"stablediffusion":        StableDiffusionGGMLBackend,
 }
 
 var TypeAlias map[string]string = map[string]string{
@@ -54,10 +55,10 @@ const (
 
 	LLamaCPPGRPC = "llama-cpp-grpc"
 
-	WhisperBackend         = "whisper"
+	WhisperBackend             = "whisper"
-	StableDiffusionBackend = "stablediffusion"
+	StableDiffusionGGMLBackend = "stablediffusion-ggml"
-	PiperBackend           = "piper"
+	PiperBackend               = "piper"
-	LCHuggingFaceBackend   = "huggingface"
+	LCHuggingFaceBackend       = "huggingface"
 
 	TransformersBackend = "transformers"
 	LocalStoreBackend   = "local-store"

pkg/stablediffusion/generate.go

@@ -1,35 +0,0 @@
-//go:build stablediffusion
-// +build stablediffusion
-
-package stablediffusion
-
-import (
-	stableDiffusion "github.com/mudler/go-stable-diffusion"
-)
-
-func GenerateImage(height, width, mode, step, seed int, positive_prompt, negative_prompt, dst, asset_dir string) error {
-	if height > 512 || width > 512 {
-		return stableDiffusion.GenerateImageUpscaled(
-			height,
-			width,
-			step,
-			seed,
-			positive_prompt,
-			negative_prompt,
-			dst,
-			asset_dir,
-		)
-	}
-	return stableDiffusion.GenerateImage(
-		height,
-		width,
-		mode,
-		step,
-		seed,
-		positive_prompt,
-		negative_prompt,
-		dst,
-		"",
-		asset_dir,
-	)
-}

pkg/stablediffusion/generate_unsupported.go

@@ -1,10 +0,0 @@
-//go:build !stablediffusion
-// +build !stablediffusion
-
-package stablediffusion
-
-import "fmt"
-
-func GenerateImage(height, width, mode, step, seed int, positive_prompt, negative_prompt, dst, asset_dir string) error {
-	return fmt.Errorf("This version of LocalAI was built without the stablediffusion tag")
-}

pkg/stablediffusion/stablediffusion.go

@@ -1,20 +0,0 @@
-package stablediffusion
-
-import "os"
-
-type StableDiffusion struct {
-	assetDir string
-}
-
-func New(assetDir string) (*StableDiffusion, error) {
-	if _, err := os.Stat(assetDir); err != nil {
-		return nil, err
-	}
-	return &StableDiffusion{
-		assetDir: assetDir,
-	}, nil
-}
-
-func (s *StableDiffusion) GenerateImage(height, width, mode, step, seed int, positive_prompt, negative_prompt, dst string) error {
-	return GenerateImage(height, width, mode, step, seed, positive_prompt, negative_prompt, dst, s.assetDir)
-}

tests/e2e-aio/e2e_suite_test.go

@@ -54,7 +54,7 @@ var _ = BeforeSuite(func() {
 	Eventually(func() error {
 		_, err := client.ListModels(context.TODO())
 		return err
-	}, "20m").ShouldNot(HaveOccurred())
+	}, "50m").ShouldNot(HaveOccurred())
 })
 
 var _ = AfterSuite(func() {

tests/e2e-aio/e2e_test.go

@@ -123,8 +123,9 @@ var _ = Describe("E2E test", func() {
 			It("correctly", func() {
 				resp, err := client.CreateImage(context.TODO(),
 					openai.ImageRequest{
-						Prompt: "test",
+						Prompt:  "test",
-						Size:   openai.CreateImageSize512x512,
+						Quality: "1",
+						Size:    openai.CreateImageSize256x256,
 					},
 				)
 				Expect(err).ToNot(HaveOccurred())
@@ -135,7 +136,8 @@ var _ = Describe("E2E test", func() {
 				resp, err := client.CreateImage(context.TODO(),
 					openai.ImageRequest{
 						Prompt:         "test",
-						Size:           openai.CreateImageSize512x512,
+						Size:           openai.CreateImageSize256x256,
+						Quality:        "1",
 						ResponseFormat: openai.CreateImageResponseFormatURL,
 					},
 				)
@@ -147,7 +149,8 @@ var _ = Describe("E2E test", func() {
 				resp, err := client.CreateImage(context.TODO(),
 					openai.ImageRequest{
 						Prompt:         "test",
-						Size:           openai.CreateImageSize512x512,
+						Size:           openai.CreateImageSize256x256,
+						Quality:        "1",
 						ResponseFormat: openai.CreateImageResponseFormatB64JSON,
 					},
 				)