docs(transformers): add docs section about transformers

Fixes: #1276

.dockerignore

@@ -3,4 +3,4 @@ models
 examples/chatbot-ui/models
 examples/rwkv/models
 examples/**/models
-Dockerfile*
+Dockerfile

.github/workflows/image-pr.yml

@@ -22,7 +22,6 @@ jobs:
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
       base-image: ${{ matrix.base-image }}
-      makeflags: "-j3"
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -81,7 +80,6 @@ jobs:
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
       base-image: ${{ matrix.base-image }}
-      makeflags: "-j3"
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}

.github/workflows/image.yml

@@ -26,7 +26,6 @@ jobs:
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
       base-image: ${{ matrix.base-image }}
-      makeflags: "-j3"
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -200,7 +199,6 @@ jobs:
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
       base-image: ${{ matrix.base-image }}
-      makeflags: "-j3"
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}

.github/workflows/image_build.yml

@@ -46,11 +46,6 @@ on:
         required: true
         default: ''
         type: string
-      makeflags:
-        description: 'Make Flags'
-        required: false
-        default: ''
-        type: string
     secrets:
       dockerUsername:
         required: true
@@ -165,7 +160,6 @@ jobs:
             FFMPEG=${{ inputs.ffmpeg }}
             IMAGE_TYPE=${{ inputs.image-type }}
             BASE_IMAGE=${{ inputs.base-image }}
-            MAKEFLAGS=${{ inputs.makeflags }}
           context: .
           file: ./Dockerfile
           platforms: ${{ inputs.platforms }}

.github/workflows/test.yml

@@ -105,13 +105,9 @@ jobs:
       - name: Test
         run: |
           GO_TAGS="stablediffusion tts" make test
-      - name: Setup tmate session if tests fail
-        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3
-        timeout-minutes: 5
 
   tests-apple:
-    runs-on: macOS-14
+    runs-on: macOS-latest
     strategy:
       matrix:
         go-version: ['1.21.x']
@@ -134,8 +130,4 @@ jobs:
         run: |
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
-          BUILD_TYPE="GITHUB_CI_HAS_BROKEN_METAL" CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF" make test
-      - name: Setup tmate session if tests fail
-        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3
-        timeout-minutes: 5
+          CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF" make test

Dockerfile

@@ -63,9 +63,7 @@ WORKDIR /build
 RUN test -n "$TARGETARCH" \
     || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
 
-###################################
-###################################
-
+# Extras requirements
 FROM requirements-core as requirements-extras
 
 RUN curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
@@ -95,11 +93,8 @@ FROM requirements-${IMAGE_TYPE} as builder
 ARG GO_TAGS="stablediffusion tts"
 ARG GRPC_BACKENDS
 ARG BUILD_GRPC=true
-ARG MAKEFLAGS
-
 ENV GRPC_BACKENDS=${GRPC_BACKENDS}
 ENV GO_TAGS=${GO_TAGS}
-ENV MAKEFLAGS=${MAKEFLAGS}
 ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
 ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
 ENV NVIDIA_VISIBLE_DEVICES=all
@@ -108,7 +103,6 @@ WORKDIR /build
 
 COPY . .
 COPY .git .
-RUN echo "GO_TAGS: $GO_TAGS"
 RUN make prepare
 
 # If we are building with clblas support, we need the libraries for the builds
@@ -122,10 +116,10 @@ RUN if [ "${BUILD_TYPE}" = "clblas" ]; then \
 RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
 
 RUN if [ "${BUILD_GRPC}" = "true" ]; then \
-    git clone --recurse-submodules --jobs 4 -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+    git clone --recurse-submodules -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
     cd grpc && mkdir -p cmake/build && cd cmake/build && cmake -DgRPC_INSTALL=ON \
       -DgRPC_BUILD_TESTS=OFF \
-       ../.. && make install \
+       ../.. && make -j12 install \
     ; fi
 
 # Rebuild with defaults backends
@@ -145,12 +139,10 @@ ARG FFMPEG
 ARG BUILD_TYPE
 ARG TARGETARCH
 ARG IMAGE_TYPE=extras
-ARG MAKEFLAGS
 
 ENV BUILD_TYPE=${BUILD_TYPE}
 ENV REBUILD=false
 ENV HEALTHCHECK_ENDPOINT=http://localhost:8080/readyz
-ENV MAKEFLAGS=${MAKEFLAGS}
 
 ARG CUDA_MAJOR_VERSION=11
 ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
@@ -194,43 +186,43 @@ COPY --from=builder /build/backend-assets/grpc/stablediffusion ./backend-assets/
 
 ## Duplicated from Makefile to avoid having a big layer that's hard to push
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/autogptq \
+	 make -C backend/python/autogptq \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/bark \
+	 make -C backend/python/bark \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/diffusers \
+	 make -C backend/python/diffusers \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/vllm \
+	 make -C backend/python/vllm \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/mamba \
+	 make -C backend/python/mamba \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/sentencetransformers \
+	 make -C backend/python/sentencetransformers \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/transformers \
+	 make -C backend/python/transformers \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/vall-e-x \
+	 make -C backend/python/vall-e-x \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/exllama \
+	 make -C backend/python/exllama \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/exllama2 \
+     make -C backend/python/exllama2 \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/petals \
+	 make -C backend/python/petals \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/transformers-musicgen \
+	 make -C backend/python/transformers-musicgen \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    make -C backend/python/coqui \
+	 make -C backend/python/coqui \
     ; fi
 
 # Make sure the models directory exists

Makefile

@@ -4,8 +4,11 @@ GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
 
 # llama.cpp versions
-GOLLAMA_STABLE_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=d01b3c4c32357567f3531d4e6ceffc5d23e87583
+GOLLAMA_VERSION?=aeba71ee842819da681ea537e78846dc75949ac0
+
+GOLLAMA_STABLE_VERSION?=50cee7712066d9e38306eccadcfbb44ea87df4b7
+
+CPPLLAMA_VERSION?=19885d205e768579ab090d1e99281cae58c21b54
 
 # gpt4all version
 GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
@@ -16,13 +19,13 @@ RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
 RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
 
 # whisper.cpp version
-WHISPER_CPP_VERSION?=a56f435fd475afd7edf02bfbf9f8c77f527198c2
+WHISPER_CPP_VERSION?=37a709f6558c6d9783199e2b8cbb136e1c41d346
 
 # bert.cpp version
 BERT_VERSION?=6abe312cded14042f6b7c3cd8edf082713334a4d
 
 # go-piper version
-PIPER_VERSION?=9d0100873a7dbb0824dfea40e8cec70a1b110759
+PIPER_VERSION?=d6b6275ba037dabdba4a8b65dfdf6b2a73a67f07
 
 # stablediffusion version
 STABLEDIFFUSION_VERSION?=362df9da29f882dbf09ade61972d16a1f53c3485
@@ -35,7 +38,6 @@ export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
 export CMAKE_ARGS?=
 
 CGO_LDFLAGS?=
-CGO_LDFLAGS_WHISPER?=
 CUDA_LIBPATH?=/usr/local/cuda/lib64/
 GO_TAGS?=
 BUILD_ID?=git
@@ -70,7 +72,7 @@ UNAME_S := $(shell uname -s)
 endif
 
 ifeq ($(OS),Darwin)
-	
+	CGO_LDFLAGS += -lcblas -framework Accelerate
 	ifeq ($(OSX_SIGNING_IDENTITY),)
 		OSX_SIGNING_IDENTITY := $(shell security find-identity -v -p codesigning | grep '"' | head -n 1 | sed -E 's/.*"(.*)"/\1/')
 	endif
@@ -81,12 +83,6 @@ ifeq ($(OS),Darwin)
 	# disable metal if on Darwin and any other value is explicitly passed.
 	else ifneq ($(BUILD_TYPE),metal)
 		CMAKE_ARGS+=-DLLAMA_METAL=OFF
-		export LLAMA_NO_ACCELERATE=1
-	endif
-
-	ifeq ($(BUILD_TYPE),metal)
-#			-lcblas 	removed: it seems to always be listed as a duplicate flag.
-		CGO_LDFLAGS += -framework Accelerate
 	endif
 endif
 
@@ -95,12 +91,10 @@ ifeq ($(BUILD_TYPE),openblas)
 	export WHISPER_OPENBLAS=1
 endif
 
-
 ifeq ($(BUILD_TYPE),cublas)
 	CGO_LDFLAGS+=-lcublas -lcudart -L$(CUDA_LIBPATH)
 	export LLAMA_CUBLAS=1
 	export WHISPER_CUBLAS=1
-	CGO_LDFLAGS_WHISPER+=-L$(CUDA_LIBPATH)/stubs/ -lcuda
 endif
 
 ifeq ($(BUILD_TYPE),hipblas)
@@ -154,6 +148,7 @@ endif
 
 ALL_GRPC_BACKENDS=backend-assets/grpc/langchain-huggingface
 ALL_GRPC_BACKENDS+=backend-assets/grpc/bert-embeddings
+ALL_GRPC_BACKENDS+=backend-assets/grpc/llama
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
 ALL_GRPC_BACKENDS+=backend-assets/grpc/gpt4all
@@ -173,41 +168,40 @@ ifeq ($(BUILD_API_ONLY),true)
 	GRPC_BACKENDS=
 endif
 
-.PHONY: all test build vendor get-sources prepare-sources prepare
+.PHONY: all test build vendor
 
 all: help
 
-## BERT embeddings
-sources/go-bert:
-	git clone --recurse-submodules https://github.com/go-skynet/go-bert.cpp sources/go-bert
-	cd sources/go-bert && git checkout -b build $(BERT_VERSION) && git submodule update --init --recursive --depth 1
-
-sources/go-bert/libgobert.a: sources/go-bert
-	$(MAKE) -C sources/go-bert libgobert.a
-
-## go-llama-ggml
-sources/go-llama-ggml:
-	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp sources/go-llama-ggml
-	cd sources/go-llama-ggml && git checkout -b build $(GOLLAMA_STABLE_VERSION) && git submodule update --init --recursive --depth 1
-
-sources/go-llama-ggml/libbinding.a: sources/go-llama-ggml
-	$(MAKE) -C sources/go-llama-ggml BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a
-
-## go-piper
-sources/go-piper:
-	git clone --recurse-submodules https://github.com/mudler/go-piper sources/go-piper
-	cd sources/go-piper && git checkout -b build $(PIPER_VERSION) && git submodule update --init --recursive --depth 1
-
-sources/go-piper/libpiper_binding.a: sources/go-piper
-	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o
-
 ## GPT4ALL
 sources/gpt4all:
 	git clone --recurse-submodules $(GPT4ALL_REPO) sources/gpt4all
 	cd sources/gpt4all && git checkout -b build $(GPT4ALL_VERSION) && git submodule update --init --recursive --depth 1
 
-sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a: sources/gpt4all
-	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ libgpt4all.a
+## go-piper
+sources/go-piper:
+	git clone --recurse-submodules https://github.com/mudler/go-piper sources/go-piper
+	cd sources/go-piper && git checkout -b build $(PIPER_VERSION) && git submodule update --init --recursive --depth 1
+
+## BERT embeddings
+sources/go-bert:
+	git clone --recurse-submodules https://github.com/go-skynet/go-bert.cpp sources/go-bert
+	cd sources/go-bert && git checkout -b build $(BERT_VERSION) && git submodule update --init --recursive --depth 1
+
+## stable diffusion
+sources/go-stable-diffusion:
+	git clone --recurse-submodules https://github.com/mudler/go-stable-diffusion sources/go-stable-diffusion
+	cd sources/go-stable-diffusion && git checkout -b build $(STABLEDIFFUSION_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-stable-diffusion/libstablediffusion.a:
+	$(MAKE) -C sources/go-stable-diffusion libstablediffusion.a
+
+## tiny-dream
+sources/go-tiny-dream:
+	git clone --recurse-submodules https://github.com/M0Rf30/go-tiny-dream sources/go-tiny-dream
+	cd sources/go-tiny-dream && git checkout -b build $(TINYDREAM_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-tiny-dream/libtinydream.a:
+	$(MAKE) -C sources/go-tiny-dream libtinydream.a
 
 ## RWKV
 sources/go-rwkv:
@@ -217,23 +211,23 @@ sources/go-rwkv:
 sources/go-rwkv/librwkv.a: sources/go-rwkv
 	cd sources/go-rwkv && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
 
-## stable diffusion
-sources/go-stable-diffusion:
-	git clone --recurse-submodules https://github.com/mudler/go-stable-diffusion sources/go-stable-diffusion
-	cd sources/go-stable-diffusion && git checkout -b build $(STABLEDIFFUSION_VERSION) && git submodule update --init --recursive --depth 1
+sources/go-bert/libgobert.a: sources/go-bert
+	$(MAKE) -C sources/go-bert libgobert.a
 
-sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
-	$(MAKE) -C sources/go-stable-diffusion libstablediffusion.a
+backend-assets/gpt4all: sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a
+	mkdir -p backend-assets/gpt4all
+	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.so backend-assets/gpt4all/ || true
+	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.dylib backend-assets/gpt4all/ || true
+	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.dll backend-assets/gpt4all/ || true
 
-## tiny-dream
-sources/go-tiny-dream:
-	git clone --recurse-submodules https://github.com/M0Rf30/go-tiny-dream sources/go-tiny-dream
-	cd sources/go-tiny-dream && git checkout -b build $(TINYDREAM_VERSION) && git submodule update --init --recursive --depth 1
+backend-assets/espeak-ng-data: sources/go-piper
+	mkdir -p backend-assets/espeak-ng-data
+	$(MAKE) -C sources/go-piper piper.o
+	@cp -rf sources/go-piper/piper-phonemize/pi/share/espeak-ng-data/. backend-assets/espeak-ng-data
 
-sources/go-tiny-dream/libtinydream.a: sources/go-tiny-dream
-	$(MAKE) -C sources/go-tiny-dream libtinydream.a
+sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a: sources/gpt4all
+	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ libgpt4all.a
 
-## whisper
 sources/whisper.cpp:
 	git clone https://github.com/ggerganov/whisper.cpp.git sources/whisper.cpp
 	cd sources/whisper.cpp && git checkout -b build $(WHISPER_CPP_VERSION) && git submodule update --init --recursive --depth 1
@@ -241,34 +235,47 @@ sources/whisper.cpp:
 sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
 	cd sources/whisper.cpp && make libwhisper.a
 
-get-sources: sources/go-llama-ggml sources/gpt4all sources/go-piper sources/go-rwkv sources/whisper.cpp sources/go-bert sources/go-stable-diffusion sources/go-tiny-dream
+sources/go-llama:
+	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp sources/go-llama
+	cd sources/go-llama && git checkout -b build $(GOLLAMA_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-llama-ggml:
+	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp sources/go-llama-ggml
+	cd sources/go-llama-ggml && git checkout -b build $(GOLLAMA_STABLE_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-llama/libbinding.a: sources/go-llama
+	$(MAKE) -C sources/go-llama BUILD_TYPE=$(BUILD_TYPE) libbinding.a
+
+sources/go-llama-ggml/libbinding.a: sources/go-llama-ggml
+	$(MAKE) -C sources/go-llama-ggml BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a
+
+sources/go-piper/libpiper_binding.a: sources/go-piper
+	$(MAKE) -C sources/go-piper libpiper_binding.a example/main
+
+backend/cpp/llama/llama.cpp:
+	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp
+
+get-sources: backend/cpp/llama/llama.cpp sources/go-llama sources/go-llama-ggml sources/gpt4all sources/go-piper sources/go-rwkv sources/whisper.cpp sources/go-bert sources/go-stable-diffusion sources/go-tiny-dream
+	touch $@
 
 replace:
+	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang
 	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(CURDIR)/sources/go-rwkv
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(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
+	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(CURDIR)/sources/go-stable-diffusion
 	$(GOCMD) mod edit -replace github.com/M0Rf30/go-tiny-dream=$(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
-	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang
-
-dropreplace:
-	$(GOCMD) mod edit -dropreplace github.com/donomii/go-rwkv.cpp
-	$(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
-	$(GOCMD) mod edit -dropreplace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang
 
 prepare-sources: get-sources replace
 	$(GOCMD) mod download
+	touch $@
 
 ## GENERIC
 rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
+	$(MAKE) -C sources/go-llama clean
 	$(MAKE) -C sources/go-llama-ggml clean
 	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ clean
 	$(MAKE) -C sources/go-rwkv clean
@@ -280,6 +287,7 @@ rebuild: ## Rebuilds the project
 	$(MAKE) build
 
 prepare: prepare-sources $(OPTIONAL_TARGETS)
+	touch $@
 
 clean: ## Remove build related file
 	$(GOCMD) clean -cache
@@ -290,15 +298,10 @@ clean: ## Remove build related file
 	rm -rf backend-assets
 	$(MAKE) -C backend/cpp/grpc clean
 	$(MAKE) -C backend/cpp/llama clean
-	$(MAKE) dropreplace
-
-clean-tests:
-	rm -rf test-models
-	rm -rf test-dir
-	rm -rf core/http/backend-assets
 
 ## Build:
-build: prepare backend-assets grpcs ## Build the project
+
+build: backend-assets grpcs prepare ## Build the project
 	$(info ${GREEN}I local-ai build info:${RESET})
 	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
 	$(info ${GREEN}I GO_TAGS: ${YELLOW}$(GO_TAGS)${RESET})
@@ -316,10 +319,10 @@ osx-signed: build
 run: prepare ## run local-ai
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) run ./
 
-test-models/testmodel.ggml:
+test-models/testmodel:
 	mkdir test-models
 	mkdir test-dir
-	wget -q https://huggingface.co/TheBloke/orca_mini_3B-GGML/resolve/main/orca-mini-3b.ggmlv3.q4_0.bin -O test-models/testmodel.ggml
+	wget -q https://huggingface.co/TheBloke/orca_mini_3B-GGML/resolve/main/orca-mini-3b.ggmlv3.q4_0.bin -O test-models/testmodel
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
@@ -331,7 +334,7 @@ prepare-test: grpcs
 	cp -rf backend-assets core/http
 	cp tests/models_fixtures/* test-models
 
-test: prepare test-models/testmodel.ggml grpcs
+test: prepare test-models/testmodel grpcs
 	@echo 'Running tests'
 	export GO_TAGS="tts stablediffusion"
 	$(MAKE) prepare-test
@@ -451,85 +454,87 @@ ifeq ($(BUILD_API_ONLY),true)
 	touch backend-assets/keep
 endif
 
-backend-assets/espeak-ng-data: sources/go-piper sources/go-piper/libpiper_binding.a
-	mkdir -p backend-assets/espeak-ng-data
-	@cp -rf sources/go-piper/piper-phonemize/pi/share/espeak-ng-data/. backend-assets/espeak-ng-data
-
-backend-assets/gpt4all: sources/gpt4all sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a
-	mkdir -p backend-assets/gpt4all
-	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.so backend-assets/gpt4all/ || true
-	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.dylib backend-assets/gpt4all/ || true
-	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.dll backend-assets/gpt4all/ || true
-
-backend-assets/grpc: replace
+backend-assets/grpc:
 	mkdir -p backend-assets/grpc
 
-backend-assets/grpc/bert-embeddings: sources/go-bert sources/go-bert/libgobert.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-bert LIBRARY_PATH=$(CURDIR)/sources/go-bert \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
-
-backend-assets/grpc/gpt4all: sources/gpt4all sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a backend-assets/gpt4all backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ LIBRARY_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
-
-backend-assets/grpc/langchain-huggingface: backend-assets/grpc
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/langchain-huggingface ./backend/go/llm/langchain/
-
-backend/cpp/llama/llama.cpp:
-	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp
+backend-assets/grpc/llama: backend-assets/grpc sources/go-llama/libbinding.a
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama LIBRARY_PATH=$(CURDIR)/sources/go-llama \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama ./backend/go/llm/llama/
+# TODO: every binary should have its own folder instead, so can have different  implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp backend/cpp/llama/llama.cpp/ggml-metal.metal backend-assets/grpc/
+endif
 
+## BACKEND CPP LLAMA START
+# Sets the variables in case it has to build the gRPC locally.
 INSTALLED_PACKAGES=$(CURDIR)/backend/cpp/grpc/installed_packages
 INSTALLED_LIB_CMAKE=$(INSTALLED_PACKAGES)/lib/cmake
 ADDED_CMAKE_ARGS=-Dabsl_DIR=${INSTALLED_LIB_CMAKE}/absl \
-				 -DProtobuf_DIR=${INSTALLED_LIB_CMAKE}/protobuf \
-				 -Dutf8_range_DIR=${INSTALLED_LIB_CMAKE}/utf8_range \
-				 -DgRPC_DIR=${INSTALLED_LIB_CMAKE}/grpc \
-				 -DCMAKE_CXX_STANDARD_INCLUDE_DIRECTORIES=${INSTALLED_PACKAGES}/include
+                 -DProtobuf_DIR=${INSTALLED_LIB_CMAKE}/protobuf \
+                 -Dutf8_range_DIR=${INSTALLED_LIB_CMAKE}/utf8_range \
+                 -DgRPC_DIR=${INSTALLED_LIB_CMAKE}/grpc \
+                 -DCMAKE_CXX_STANDARD_INCLUDE_DIRECTORIES=${INSTALLED_PACKAGES}/include
+
 backend/cpp/llama/grpc-server:
-# Conditionally build grpc for the llama backend to use if needed
 ifdef BUILD_GRPC_FOR_BACKEND_LLAMA
 	$(MAKE) -C backend/cpp/grpc build
-	_PROTOBUF_PROTOC=${INSTALLED_PACKAGES}/bin/proto \
-	_GRPC_CPP_PLUGIN_EXECUTABLE=${INSTALLED_PACKAGES}/bin/grpc_cpp_plugin \
-	PATH="${INSTALLED_PACKAGES}/bin:${PATH}" \
-	CMAKE_ARGS="${CMAKE_ARGS} ${ADDED_CMAKE_ARGS}" \
-	LLAMA_VERSION=$(CPPLLAMA_VERSION) \
-	$(MAKE) -C backend/cpp/llama grpc-server
+	export _PROTOBUF_PROTOC=${INSTALLED_PACKAGES}/bin/proto && \
+	export _GRPC_CPP_PLUGIN_EXECUTABLE=${INSTALLED_PACKAGES}/bin/grpc_cpp_plugin && \
+	export PATH="${INSTALLED_PACKAGES}/bin:${PATH}" && \
+	CMAKE_ARGS="${CMAKE_ARGS} ${ADDED_CMAKE_ARGS}" LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server
 else
 	echo "BUILD_GRPC_FOR_BACKEND_LLAMA is not defined."
 	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server
 endif
+## BACKEND CPP LLAMA END
 
+##
 backend-assets/grpc/llama-cpp: backend-assets/grpc backend/cpp/llama/grpc-server
 	cp -rfv backend/cpp/llama/grpc-server backend-assets/grpc/llama-cpp
 # TODO: every binary should have its own folder instead, so can have different metal implementations
 ifeq ($(BUILD_TYPE),metal)
-	cp backend/cpp/llama/llama.cpp/build/bin/default.metallib backend-assets/grpc/
+	cp backend/cpp/llama/llama.cpp/build/bin/ggml-metal.metal backend-assets/grpc/
 endif
 
-backend-assets/grpc/llama-ggml: sources/go-llama-ggml sources/go-llama-ggml/libbinding.a backend-assets/grpc
+backend-assets/grpc/llama-ggml: backend-assets/grpc sources/go-llama-ggml/libbinding.a
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama-ggml
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama-ggml LIBRARY_PATH=$(CURDIR)/sources/go-llama-ggml \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-ggml ./backend/go/llm/llama-ggml/
 
-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 "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
+backend-assets/grpc/gpt4all: backend-assets/grpc backend-assets/gpt4all sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ LIBRARY_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
 
-backend-assets/grpc/rwkv: sources/go-rwkv sources/go-rwkv/librwkv.a backend-assets/grpc
+backend-assets/grpc/rwkv: backend-assets/grpc sources/go-rwkv/librwkv.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv LIBRARY_PATH=$(CURDIR)/sources/go-rwkv \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
 
-backend-assets/grpc/stablediffusion: sources/go-stable-diffusion sources/go-stable-diffusion/libstablediffusion.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-stable-diffusion/ LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion
+backend-assets/grpc/bert-embeddings: backend-assets/grpc sources/go-bert/libgobert.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-bert LIBRARY_PATH=$(CURDIR)/sources/go-bert \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
 
-backend-assets/grpc/tinydream: sources/go-tiny-dream sources/go-tiny-dream/libtinydream.a backend-assets/grpc
+backend-assets/grpc/langchain-huggingface: backend-assets/grpc
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/langchain-huggingface ./backend/go/llm/langchain/
+
+backend-assets/grpc/stablediffusion: backend-assets/grpc
+	if [ ! -f backend-assets/grpc/stablediffusion ]; then \
+		$(MAKE) sources/go-stable-diffusion; \
+		$(MAKE) sources/go-stable-diffusion/libstablediffusion.a; \
+		CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-stable-diffusion/ LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
+		$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion; \
+	fi
+
+backend-assets/grpc/tinydream: backend-assets/grpc sources/go-tiny-dream/libtinydream.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/go-tiny-dream \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/tinydream ./backend/go/image/tinydream
 
-backend-assets/grpc/whisper: sources/whisper.cpp sources/whisper.cpp/libwhisper.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS) $(CGO_LDFLAGS_WHISPER)" C_INCLUDE_PATH=$(CURDIR)/sources/whisper.cpp LIBRARY_PATH=$(CURDIR)/sources/whisper.cpp \
+backend-assets/grpc/piper: backend-assets/grpc backend-assets/espeak-ng-data sources/go-piper/libpiper_binding.a
+	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/sources/go-piper \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
+
+backend-assets/grpc/whisper: backend-assets/grpc sources/whisper.cpp/libwhisper.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/whisper.cpp LIBRARY_PATH=$(CURDIR)/sources/whisper.cpp \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
 
 grpcs: prepare $(GRPC_BACKENDS)

backend/cpp/grpc/Makefile

@@ -48,7 +48,7 @@ $(INSTALLED_PACKAGES): grpc_build
 
 $(GRPC_REPO):
 	git clone --depth $(GIT_CLONE_DEPTH) -b $(TAG_LIB_GRPC) $(GIT_REPO_LIB_GRPC) $(GRPC_REPO)/grpc
-	cd $(GRPC_REPO)/grpc && git submodule update --jobs 2 --init --recursive --depth $(GIT_CLONE_DEPTH)
+	cd $(GRPC_REPO)/grpc && git submodule update --init --recursive --depth $(GIT_CLONE_DEPTH)
 
 $(GRPC_BUILD): $(GRPC_REPO)
 	mkdir -p $(GRPC_BUILD)

backend/cpp/llama/Makefile

@@ -18,12 +18,6 @@ else ifeq ($(BUILD_TYPE),clblas)
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
 else ifeq ($(BUILD_TYPE),hipblas)
 	CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON
-# If it's OSX, DO NOT embed the metal library - -DLLAMA_METAL_EMBED_LIBRARY=ON requires further investigation
-# But if it's OSX without metal, disable it here
-else ifeq ($(OS),darwin)
-	ifneq ($(BUILD_TYPE),metal)
-		CMAKE_ARGS+=-DLLAMA_METAL=OFF
-	endif
 endif
 
 ifeq ($(BUILD_TYPE),sycl_f16)
@@ -41,7 +35,7 @@ llama.cpp:
 	fi
 	cd llama.cpp && git checkout -b build $(LLAMA_VERSION) && git submodule update --init --recursive --depth 1
 
-llama.cpp/examples/grpc-server: llama.cpp
+llama.cpp/examples/grpc-server:
 	mkdir -p llama.cpp/examples/grpc-server
 	cp -r $(abspath ./)/CMakeLists.txt llama.cpp/examples/grpc-server/
 	cp -r $(abspath ./)/grpc-server.cpp llama.cpp/examples/grpc-server/

backend/cpp/llama/grpc-server.cpp

@@ -1084,7 +1084,7 @@ struct llama_server_context
             slot.has_next_token = false;
         }
 
-        if (result.tok == llama_token_eos(model))
+        if (!slot.cache_tokens.empty() && result.tok == llama_token_eos(model))
         {
             slot.stopped_eos = true;
             slot.has_next_token = false;

backend/python/common-env/transformers/transformers-nvidia.yml

@@ -30,7 +30,6 @@ dependencies:
       - async-timeout==4.0.3
       - attrs==23.1.0
       - bark==0.1.5
-      - bitsandbytes==0.43.0
       - boto3==1.28.61
       - botocore==1.31.61
       - certifi==2023.7.22

backend/python/transformers/transformers_server.py

@@ -23,7 +23,7 @@ if XPU:
     from intel_extension_for_transformers.transformers.modeling import AutoModelForCausalLM
     from transformers import AutoTokenizer, AutoModel, set_seed
 else:
-    from transformers import AutoTokenizer, AutoModel, AutoModelForCausalLM, set_seed, BitsAndBytesConfig
+    from transformers import AutoTokenizer, AutoModel, AutoModelForCausalLM, set_seed
 
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
@@ -75,50 +75,18 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             A Result object that contains the result of the LoadModel operation.
         """
         model_name = request.Model
-
-        compute = "auto"
-        if request.F16Memory == True:
-            compute=torch.bfloat16
-
-        self.CUDA = request.CUDA
-
-        device_map="cpu"
-
-        quantization = None
-
-        if self.CUDA:
-            if request.Device:
-                device_map=request.Device
-            else:
-                device_map="cuda:0"
-            if request.Quantization == "bnb_4bit":
-                quantization = BitsAndBytesConfig(
-                    load_in_4bit = True,
-                    bnb_4bit_compute_dtype = compute,
-                    bnb_4bit_quant_type = "nf4",
-                    bnb_4bit_use_double_quant = True,
-                    load_in_8bit = False,
-                )
-            elif request.Quantization == "bnb_8bit":
-                quantization = BitsAndBytesConfig(
-                    load_in_4bit=False,
-                    bnb_4bit_compute_dtype = None,
-                    load_in_8bit=True,                                   
-                )
-                                                   
-    
         try:
             if request.Type == "AutoModelForCausalLM":
                 if XPU:
-                    if quantization == "xpu_4bit":
-                        xpu_4bit = True
                     self.model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode,
-                                              device_map="xpu", load_in_4bit=xpu_4bit)
+                                              device_map="xpu", load_in_4bit=True)
                 else:
-                    self.model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode, use_safetensors=True, quantization_config=quantization, device_map=device_map, torch_dtype=compute)
+                    self.model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode)
             else:
-                self.model = AutoModel.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode,  use_safetensors=True,  quantization_config=quantization, device_map=device_map, torch_dtype=compute)
-            self.tokenizer = AutoTokenizer.from_pretrained(model_name, use_safetensors=True)
+                self.model = AutoModel.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode)
+
+            self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+            self.CUDA = False
             self.XPU = False
 
             if XPU:
@@ -129,6 +97,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 except Exception as err:
                     print("Not using XPU:", err, file=sys.stderr)
 
+            if request.CUDA or torch.cuda.is_available():
+                try:
+                    print("Loading model", model_name, "to CUDA.", file=sys.stderr)
+                    self.model = self.model.to("cuda")
+                    self.CUDA = True
+                except Exception as err:
+                    print("Not using CUDA:", err, file=sys.stderr)
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
         # Implement your logic here for the LoadModel service
@@ -155,17 +130,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         encoded_input = self.tokenizer(request.Embeddings, padding=True, truncation=True, max_length=max_length, return_tensors="pt")    
 
         # Create word embeddings
-        if self.CUDA:
-            encoded_input = encoded_input.to("cuda")
-
-        with torch.no_grad():    
-            model_output = self.model(**encoded_input)
+        model_output = self.model(**encoded_input)
 
         # Pool to get sentence embeddings; i.e. generate one 1024 vector for the entire sentence
-        sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+        sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask']).detach().numpy()
         print("Calculated embeddings for: " + request.Embeddings, file=sys.stderr)
         print("Embeddings:", sentence_embeddings, file=sys.stderr)
-        return backend_pb2.EmbeddingResult(embeddings=sentence_embeddings[0])
+        return backend_pb2.EmbeddingResult(embeddings=sentence_embeddings)
 
     def Predict(self, request, context):
         """
@@ -192,8 +163,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if XPU:
             inputs = inputs.to("xpu")
 
-        outputs = self.model.generate(inputs,max_new_tokens=max_tokens, temperature=request.Temperature, top_p=request.TopP, do_sample=True, pad_token_id=self.tokenizer.eos_token_id)
-        generated_text = self.tokenizer.batch_decode(outputs[:, inputs.shape[1]:], skip_special_tokens=True)[0]
+        outputs = self.model.generate(inputs,max_new_tokens=max_tokens, temperature=request.Temperature, top_p=request.TopP)
+
+        generated_text = self.tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
+        # Remove prompt from response if present
+        if request.Prompt in generated_text:
+            generated_text = generated_text.replace(request.Prompt, "")
 
         return backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
 

core/backend/embeddings.go

@@ -10,6 +10,10 @@ import (
 )
 
 func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() ([]float32, error), error) {
+	if !backendConfig.Embeddings {
+		return nil, fmt.Errorf("endpoint disabled for this model by API configuration")
+	}
+
 	modelFile := backendConfig.Model
 
 	grpcOpts := gRPCModelOpts(backendConfig)

core/config/backend_config.go

@@ -276,12 +276,8 @@ func (cfg *BackendConfig) SetDefaults(debug bool, threads, ctx int, f16 bool) {
 		cfg.F16 = &f16
 	}
 
-	if cfg.Debug == nil {
-		cfg.Debug = &falseV
-	}
-
 	if debug {
-		cfg.Debug = &trueV
+		cfg.Debug = &debug
 	}
 }
 

core/http/api_test.go

@@ -666,15 +666,15 @@ var _ = Describe("API test", func() {
 			Expect(err).ToNot(HaveOccurred())
 			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})
+		It("can generate completions", func() {
+			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel", Prompt: testPrompt})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
 		})
 
-		It("can generate chat completions via ggml", func() {
-			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "testmodel.ggml", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: testPrompt}}})
+		It("can generate chat completions ", func() {
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "testmodel", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: testPrompt}}})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())

core/http/endpoints/openai/request.go

@@ -185,14 +185,6 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 		config.RepeatPenalty = input.RepeatPenalty
 	}
 
-	if input.FrequencyPenalty!= 0 {
-		config.FrequencyPenalty = input.FrequencyPenalty
-	}
-
-	if input.PresencePenalty!= 0 {
-		config.PresencePenalty = input.PresencePenalty
-	}
-
 	if input.Keep != 0 {
 		config.Keep = input.Keep
 	}

core/schema/openai.go

@@ -108,7 +108,7 @@ type ChatCompletionResponseFormat struct {
 type OpenAIRequest struct {
 	PredictionOptions
 
-	Context context.Context  `json:"-"`
+	Context context.Context    `json:"-"`
 	Cancel  context.CancelFunc `json:"-"`
 
 	// whisper

core/schema/prediction.go

@@ -25,7 +25,6 @@ type PredictionOptions struct {
 	Keep          int     `json:"n_keep" yaml:"n_keep"`
 
 	FrequencyPenalty float64 `json:"frequency_penalty" yaml:"frequency_penalty"`
-	PresencePenalty  float64 `json:"presence_penalty" yaml:"presence_penalty"`
 	TFZ              float64 `json:"tfz" yaml:"tfz"`
 
 	TypicalP float64 `json:"typical_p" yaml:"typical_p"`

docs/data/version.json

@@ -1,3 +1,3 @@
 {
-  "version": "v2.10.0"
+  "version": "v2.9.0"
 }

main.go

@@ -306,16 +306,11 @@ For a list of compatible model, check out: https://localai.io/model-compatibilit
 				return fmt.Errorf("failed basic startup tasks with error %s", err.Error())
 			}
 
-			configdir := ctx.String("localai-config-dir")
-			// Watch the configuration directory
-			// If the directory does not exist, we don't watch it
-			if _, err := os.Stat(configdir); err == nil {
-				closeConfigWatcherFn, err := startup.WatchConfigDirectory(ctx.String("localai-config-dir"), options)
-				defer closeConfigWatcherFn()
+			closeConfigWatcherFn, err := startup.WatchConfigDirectory(ctx.String("localai-config-dir"), options)
+			defer closeConfigWatcherFn()
 
-				if err != nil {
-					return fmt.Errorf("failed while watching configuration directory %s", ctx.String("localai-config-dir"))
-				}
+			if err != nil {
+				return fmt.Errorf("failed while watching configuration directory %s", ctx.String("localai-config-dir"))
 			}
 
 			appHTTP, err := http.App(cl, ml, options)

pkg/model/initializers.go

@@ -15,11 +15,12 @@ import (
 )
 
 var Aliases map[string]string = map[string]string{
-	"go-llama": LLamaCPP,
+	"go-llama": GoLlamaBackend,
 	"llama":    LLamaCPP,
 }
 
 const (
+	GoLlamaBackend      = "llama"
 	LlamaGGML           = "llama-ggml"
 	LLamaCPP            = "llama-cpp"
 	Gpt4AllLlamaBackend = "gpt4all-llama"
@@ -34,11 +35,15 @@ const (
 	TinyDreamBackend       = "tinydream"
 	PiperBackend           = "piper"
 	LCHuggingFaceBackend   = "langchain-huggingface"
+
+	// External Backends that need special handling within LocalAI:
+	TransformersMusicGen = "transformers-musicgen"
 )
 
 var AutoLoadBackends []string = []string{
 	LLamaCPP,
 	LlamaGGML,
+	GoLlamaBackend,
 	Gpt4All,
 	BertEmbeddingsBackend,
 	RwkvBackend,

tests/models_fixtures/config.yaml

@@ -1,6 +1,6 @@
 - name: list1
   parameters:
-    model: testmodel.ggml
+    model: testmodel
     top_p: 80
     top_k: 0.9
     temperature: 0.1
@@ -19,7 +19,7 @@
     top_p: 80
     top_k: 0.9
     temperature: 0.1
-    model: testmodel.ggml
+    model: testmodel
   context_size: 200
   stopwords:
   - "HUMAN:"

tests/models_fixtures/gpt4.yaml

@@ -1,6 +1,6 @@
 name: gpt4all
 parameters:
-  model: testmodel.ggml
+  model: testmodel
   top_p: 80
   top_k: 0.9
   temperature: 0.1

tests/models_fixtures/gpt4_2.yaml

@@ -1,6 +1,6 @@
 name: gpt4all-2
 parameters:
-  model: testmodel.ggml
+  model: testmodel
   top_p: 80
   top_k: 0.9
   temperature: 0.1