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 <[email protected]> * chore(ci): drop stablediffusion build Signed-off-by: Ettore Di Giacinto <[email protected]> * chore(tests): add Signed-off-by: Ettore Di Giacinto <[email protected]> * chore(tests): try to fixup current tests Signed-off-by: Ettore Di Giacinto <[email protected]> * Try to fix tests Signed-off-by: Ettore Di Giacinto <[email protected]> * Tests improvements Signed-off-by: Ettore Di Giacinto <[email protected]> * chore(tests): use quality to specify step Signed-off-by: Ettore Di Giacinto <[email protected]> * chore(tests): switch to sd-1.5 also increase prep time for downloading models Signed-off-by: Ettore Di Giacinto <[email protected]> --------- Signed-off-by: Ettore Di Giacinto <[email protected]>
mudler · Jan 22, 2025 · e15d29a · e15d29a
1 parent 10675ac
commit e15d29a
Show file tree

Hide file tree

Showing 22 changed files with 123 additions and 302 deletions.
diff --git a/.devcontainer/docker-compose-devcontainer.yml b/.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:

diff --git a/.env b/.env
@@ -38,12 +38,12 @@
 ## Uncomment and set to true to enable rebuilding from source
 # REBUILD=true
 
-## Enable go tags, available: stablediffusion, tts
-## stablediffusion: image generation with stablediffusion
+## Enable go tags, available: p2p, tts
+## 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

diff --git a/.github/workflows/release.yaml b/.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/[email protected]
-      - 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

diff --git a/.github/workflows/test.yml b/.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/ && \
-          # Pre-build stable diffusion before we install a newer version of abseil (not compatible with stablediffusion-ncn)
-          PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
+          sudo cp -rfv sources/go-piper/piper-phonemize/pi/lib/. /usr/lib/
         env:
           CUDA_VERSION: 12-4
       - name: Cache grpc
@@ -129,7 +127,7 @@ jobs:
           cd grpc && cd cmake/build && sudo make --jobs 5 install
       - name: Test
         run: |
-          PATH="$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/[email protected]

diff --git a/.vscode/launch.json b/.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}"
         }

diff --git a/Dockerfile b/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 \
-        libopencv-dev && \
+        libopenblas-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

diff --git a/Makefile b/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

diff --git a/aio/cpu/image-gen.yaml b/aio/cpu/image-gen.yaml
@@ -1,56 +1,17 @@
 name: stablediffusion
-backend: stablediffusion
-parameters:
-  model: stablediffusion_assets
-
-license: "BSD-3"
-urls:
-- https://github.com/EdVince/Stable-Diffusion-NCNN
-- https://github.com/EdVince/Stable-Diffusion-NCNN/blob/main/LICENSE
+backend: stablediffusion-ggml
+cfg_scale: 4.5
 
-description: |
-     Stable Diffusion in NCNN with c++, supported txt2img and img2img
+options:
+- sampler:euler
+parameters:
+  model: stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf
+step: 25
 
 download_files:
-- filename: "stablediffusion_assets/AutoencoderKL-256-256-fp16-opt.param"
-  sha256: "18ca4b66685e21406bcf64c484b3b680b4949900415536d599cc876579c85c82"
-  uri: "https://raw.githubusercontent.com/EdVince/Stable-Diffusion-NCNN/main/x86/linux/assets/AutoencoderKL-256-256-fp16-opt.param"
-- 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"
+- filename: "stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf"
+  sha256: "b8944e9fe0b69b36ae1b5bb0185b3a7b8ef14347fe0fa9af6c64c4829022261f"
+  uri: "huggingface://second-state/stable-diffusion-v1-5-GGUF/stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf"
 
 usage: |
         curl http://localhost:8080/v1/images/generations \

diff --git a/backend/go/image/stablediffusion/main.go b/backend/go/image/stablediffusion/main.go
diff --git a/backend/go/image/stablediffusion/stablediffusion.go b/backend/go/image/stablediffusion/stablediffusion.go