chore: ⬆️ Update ggerganov/llama.cpp (#3102 )

⬆️ Update ggerganov/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>
models(gallery): add tifa (#3099 )
2026-02-03 03:02:38 -05:00 · 2024-08-01 00:55:09 +00:00 · 2024-07-31 17:14:46 +02:00 · 2024-07-31 17:10:31 +02:00 · 2024-07-31 17:06:06 +02:00 · 2024-07-31 16:09:58 +02:00
72 changed files with 2386 additions and 566 deletions
--- a/.github/workflows/checksum_checker.yaml
+++ b/.github/workflows/checksum_checker.yaml
@@ -41,7 +41,7 @@ jobs:
          token: ${{ secrets.UPDATE_BOT_TOKEN }}
          push-to-fork: ci-forks/LocalAI
          commit-message: ':arrow_up: Checksum updates in gallery/index.yaml'
-          title: 'models(gallery): :arrow_up: update checksum'
+          title: 'chore(model-gallery): :arrow_up: update checksum'
          branch: "update/checksum"
          body: Updating checksums in gallery/index.yaml
          signoff: true
--- a/.github/workflows/disabled/comment-pr.yaml
+++ b/.github/workflows/disabled/comment-pr.yaml
--- a/.github/workflows/image-pr.yml
+++ b/.github/workflows/image-pr.yml
@@ -47,7 +47,7 @@ jobs:
          #   makeflags: "--jobs=3 --output-sync=target"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "4"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-cublas-cuda12-ffmpeg'
@@ -120,7 +120,7 @@ jobs:
          #   makeflags: "--jobs=3 --output-sync=target"
          # - build-type: 'cublas'
          #   cuda-major-version: "12"
-          #   cuda-minor-version: "4"
+          #   cuda-minor-version: "0"
          #   platforms: 'linux/amd64'
          #   tag-latest: 'false'
          #   tag-suffix: '-cublas-cuda12-ffmpeg-core'
--- a/.github/workflows/image.yml
+++ b/.github/workflows/image.yml
@@ -75,7 +75,7 @@ jobs:
            makeflags: "--jobs=3 --output-sync=target"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "4"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-cublas-cuda12'
@@ -100,7 +100,7 @@ jobs:
            makeflags: "--jobs=3 --output-sync=target"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "4"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-cublas-cuda12-ffmpeg'
@@ -285,7 +285,7 @@ jobs:
            makeflags: "--jobs=4 --output-sync=target"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "4"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-cublas-cuda12-core'
@@ -307,7 +307,7 @@ jobs:
            makeflags: "--jobs=4 --output-sync=target"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "4"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-cublas-cuda12-ffmpeg-core'
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -4,6 +4,8 @@ on:
  push:
    branches:
      - master
+    tags:
+      - 'v*'
  pull_request:

 env:
@@ -29,11 +31,10 @@ jobs:
        with:
          go-version: '1.21.x'
          cache: false
-
      - name: Dependencies
        run: |
          sudo apt-get update
-          sudo apt-get install build-essential ffmpeg protobuf-compiler ccache gawk
+          sudo apt-get install build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk
          sudo apt-get install -qy binutils-aarch64-linux-gnu gcc-aarch64-linux-gnu g++-aarch64-linux-gnu libgmock-dev
      - name: Install CUDA Dependencies
        run: |
@@ -149,7 +150,7 @@ jobs:
      - name: Dependencies
        run: |
          sudo apt-get update
-          sudo apt-get install -y wget curl build-essential ffmpeg protobuf-compiler ccache gawk cmake libgmock-dev
+          sudo apt-get install -y wget curl build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk cmake libgmock-dev
      - name: Intel Dependencies
        run: |
          wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
@@ -250,7 +251,7 @@ jobs:
      - name: Dependencies
        run: |
          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends libopencv-dev protobuf-compiler ccache
+          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
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -70,7 +70,7 @@ jobs:
      - name: Dependencies
        run: |
          sudo apt-get update
-          sudo apt-get install build-essential curl ffmpeg
+          sudo apt-get install build-essential ccache upx-ucl curl ffmpeg
          sudo apt-get install -y libgmock-dev
          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
--- a/4
+++ b/4
@@ -24,7 +24,7 @@ RUN apt-get update && \
        cmake \
        curl \
        git \
-        unzip && \
+        unzip upx-ucl && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

@@ -99,7 +99,7 @@ FROM requirements-${IMAGE_TYPE} AS requirements-drivers

 ARG BUILD_TYPE
 ARG CUDA_MAJOR_VERSION=12
-ARG CUDA_MINOR_VERSION=4
+ARG CUDA_MINOR_VERSION=0

 ENV BUILD_TYPE=${BUILD_TYPE}

--- a/49
+++ b/49
@@ -8,7 +8,7 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=705b7ecf60e667ced57c15d67aa86865e3cc7aa7
+CPPLLAMA_VERSION?=ed9d2854c9de4ae1f448334294e61167b04bec2a

 # gpt4all version
 GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
@@ -20,7 +20,7 @@ RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6

 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=f68298ce06ca3edd6e6f3f21c3d0bb5f073942c3
+WHISPER_CPP_VERSION?=6739eb83c3ca5cf40d24c6fe8442a761a1eb6248

 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -58,7 +58,7 @@ RANDOM := $(shell bash -c 'echo $$RANDOM')

 VERSION?=$(shell git describe --always --tags || echo "dev" )
 # go tool nm ./local-ai | grep Commit
-LD_FLAGS?=
+LD_FLAGS?=-s -w
 override LD_FLAGS += -X "github.com/mudler/LocalAI/internal.Version=$(VERSION)"
 override LD_FLAGS += -X "github.com/mudler/LocalAI/internal.Commit=$(shell git rev-parse HEAD)"

@@ -72,6 +72,14 @@ WHITE  := $(shell tput -Txterm setaf 7)
 CYAN   := $(shell tput -Txterm setaf 6)
 RESET  := $(shell tput -Txterm sgr0)

+UPX?=
+# check if upx exists
+ifeq (, $(shell which upx))
+	UPX=
+else
+	UPX=$(shell which upx)
+endif
+
 # Default Docker bridge IP
 E2E_BRIDGE_IP?=172.17.0.1

@@ -377,6 +385,7 @@ build: prepare backend-assets grpcs ## Build the project
 	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
 	$(info ${GREEN}I GO_TAGS: ${YELLOW}$(GO_TAGS)${RESET})
 	$(info ${GREEN}I LD_FLAGS: ${YELLOW}$(LD_FLAGS)${RESET})
+	$(info ${GREEN}I UPX: ${YELLOW}$(UPX)${RESET})
 ifneq ($(BACKEND_LIBS),)
 	$(MAKE) backend-assets/lib
 	cp -f $(BACKEND_LIBS) backend-assets/lib/
@@ -421,7 +430,7 @@ else
 endif

 dist-cross-linux-arm64:
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_NATIVE=off" GRPC_BACKENDS="backend-assets/grpc/llama-cpp-fallback backend-assets/grpc/llama-cpp-grpc backend-assets/util/llama-cpp-rpc-server" \
+	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_NATIVE=off" GRPC_BACKENDS="backend-assets/grpc/llama-cpp-fallback backend-assets/grpc/llama-cpp-grpc backend-assets/util/llama-cpp-rpc-server" GO_TAGS="p2p" \
 	STATIC=true $(MAKE) build
 	mkdir -p release
 # if BUILD_ID is empty, then we don't append it to the binary name
@@ -471,7 +480,7 @@ prepare-e2e:
 	mkdir -p $(TEST_DIR)
 	cp -rfv $(abspath ./tests/e2e-fixtures)/gpu.yaml $(TEST_DIR)/gpu.yaml
 	test -e $(TEST_DIR)/ggllm-test-model.bin || wget -q https://huggingface.co/TheBloke/CodeLlama-7B-Instruct-GGUF/resolve/main/codellama-7b-instruct.Q2_K.gguf -O $(TEST_DIR)/ggllm-test-model.bin
-	docker build --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=4 --build-arg FFMPEG=true -t localai-tests .
+	docker build --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=0 --build-arg FFMPEG=true -t localai-tests .

 run-e2e-image:
 	ls -liah $(abspath ./tests/e2e-fixtures)
@@ -733,13 +742,22 @@ backend-assets/grpc: protogen-go replace
 backend-assets/grpc/bert-embeddings: sources/go-bert.cpp sources/go-bert.cpp/libgobert.a backend-assets/grpc
 	CGO_LDFLAGS="$(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/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 "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/gpt4all
+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),)
+	$(UPX) backend-assets/grpc/huggingface
+endif

 backend/cpp/llama/llama.cpp:
 	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp
@@ -841,29 +859,50 @@ backend-assets/util/llama-cpp-rpc-server: backend-assets/grpc/llama-cpp-grpc
 backend-assets/grpc/llama-ggml: sources/go-llama.cpp sources/go-llama.cpp/libbinding.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama.cpp LIBRARY_PATH=$(CURDIR)/sources/go-llama.cpp \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-ggml ./backend/go/llm/llama-ggml/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/llama-ggml
+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 "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/piper
+endif

 backend-assets/grpc/rwkv: sources/go-rwkv.cpp sources/go-rwkv.cpp/librwkv.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv.cpp LIBRARY_PATH=$(CURDIR)/sources/go-rwkv.cpp \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/rwkv
+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/tinydream: sources/go-tiny-dream sources/go-tiny-dream/libtinydream.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/go-tiny-dream \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/tinydream ./backend/go/image/tinydream
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/tinydream
+endif

 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/include:$(CURDIR)/sources/whisper.cpp/ggml/include" LIBRARY_PATH=$(CURDIR)/sources/whisper.cpp \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/whisper
+endif

 backend-assets/grpc/local-store: backend-assets/grpc
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/local-store ./backend/go/stores/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/local-store
+endif

 grpcs: prepare $(GRPC_BACKENDS)

--- a/backend/cpp/llama/grpc-server.cpp
+++ b/backend/cpp/llama/grpc-server.cpp
@@ -2259,7 +2259,6 @@ static void params_parse(const backend::ModelOptions* request,
     // get the directory of modelfile
     std::string model_dir = params.model.substr(0, params.model.find_last_of("/\\"));
     params.lora_adapter.push_back(std::make_tuple(model_dir + "/"+request->loraadapter(), scale_factor));
-     params.lora_base  =  model_dir + "/"+request->lorabase();
    }
    params.use_mlock = request->mlock();
    params.use_mmap = request->mmap();
--- a/backend/python/autogptq/requirements-intel.txt
+++ b/backend/python/autogptq/requirements-intel.txt
@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/autogptq/requirements.txt
+++ b/backend/python/autogptq/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 torch
 certifi
--- a/backend/python/bark/requirements.txt
+++ b/backend/python/bark/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 bark==0.1.5
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
 transformers
--- a/backend/python/common/template/requirements.txt
+++ b/backend/python/common/template/requirements.txt
@@ -1,2 +1,2 @@
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
--- a/backend/python/coqui/requirements-intel.txt
+++ b/backend/python/coqui/requirements-intel.txt
@@ -3,4 +3,4 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/coqui/requirements.txt
+++ b/backend/python/coqui/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 TTS==0.22.0
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
 transformers
--- a/backend/python/diffusers/requirements.txt
+++ b/backend/python/diffusers/requirements.txt
@@ -3,7 +3,7 @@ accelerate
 compel
 peft
 diffusers
-grpcio==1.65.0
+grpcio==1.65.1
 opencv-python
 pillow
 protobuf
--- a/backend/python/exllama2/requirements.txt
+++ b/backend/python/exllama2/requirements.txt
@@ -1,5 +1,5 @@
 accelerate
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
 torch
--- a/backend/python/mamba/requirements.txt
+++ b/backend/python/mamba/requirements.txt
@@ -1,6 +1,6 @@
 causal-conv1d==1.4.0
 mamba-ssm==2.2.2
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
 transformers
--- 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==1.64.1
+grpcio==1.65.1
 protobuf
 librosa==0.9.1
 faster-whisper==1.0.3
--- a/backend/python/openvoice/requirements.txt
+++ b/backend/python/openvoice/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 librosa
 faster-whisper
--- a/backend/python/openvoice/test.sh
+++ b/backend/python/openvoice/test.sh
@@ -5,7 +5,7 @@ source $(dirname $0)/../common/libbackend.sh

 # Download checkpoints if not present
 if [ ! -d "checkpoints_v2" ]; then
-    wget https://myshell-public-repo-hosting.s3.amazonaws.com/openvoice/checkpoints_v2_0417.zip -O checkpoints_v2.zip
+    wget https://myshell-public-repo-host.s3.amazonaws.com/openvoice/checkpoints_v2_0417.zip -O checkpoints_v2.zip
    unzip checkpoints_v2.zip
 fi

--- a/backend/python/parler-tts/requirements-intel.txt
+++ b/backend/python/parler-tts/requirements-intel.txt
@@ -3,4 +3,4 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/parler-tts/requirements.txt
+++ b/backend/python/parler-tts/requirements.txt
@@ -1,5 +1,5 @@
 accelerate
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 torch
 git+https://github.com/huggingface/parler-tts.git@10016fb0300c0dc31a0fb70e26f3affee7b62f16
--- a/backend/python/petals/requirements-intel.txt
+++ b/backend/python/petals/requirements-intel.txt
@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/rerankers/requirements-intel.txt
+++ b/backend/python/rerankers/requirements-intel.txt
@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/rerankers/requirements.txt
+++ b/backend/python/rerankers/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 rerankers[transformers]
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
 transformers
--- a/backend/python/sentencetransformers/requirements-intel.txt
+++ b/backend/python/sentencetransformers/requirements-intel.txt
@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/sentencetransformers/requirements.txt
+++ b/backend/python/sentencetransformers/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 sentence-transformers==3.0.1
 transformers
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
--- a/backend/python/transformers-musicgen/requirements-intel.txt
+++ b/backend/python/transformers-musicgen/requirements-intel.txt
@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/transformers-musicgen/requirements.txt
+++ b/backend/python/transformers-musicgen/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 transformers
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 torch
 scipy==1.14.0
--- a/backend/python/transformers/requirements-intel.txt
+++ b/backend/python/transformers/requirements-intel.txt
@@ -2,4 +2,3 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/transformers/requirements.txt
+++ b/backend/python/transformers/requirements.txt
@@ -1,9 +1,9 @@
 accelerate
 transformers
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 torch
 certifi
 intel-extension-for-transformers
 bitsandbytes
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/vall-e-x/requirements-intel.txt
+++ b/backend/python/vall-e-x/requirements-intel.txt
@@ -3,4 +3,4 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/vall-e-x/requirements.txt
+++ b/backend/python/vall-e-x/requirements.txt
@@ -1,4 +1,4 @@
 accelerate
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
--- a/backend/python/vllm/requirements.txt
+++ b/backend/python/vllm/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 vllm
-grpcio==1.65.0
+grpcio==1.65.1
 protobuf
 certifi
 transformers
--- a/core/gallery/gallery.go
+++ b/core/gallery/gallery.go
@@ -204,35 +204,34 @@ func DeleteModelFromSystem(basePath string, name string, additionalFiles []strin
 		log.Error().Err(err).Msgf("failed to read gallery file %s", configFile)
 	}

+	var filesToRemove []string
+
 	// Remove additional files
 	if galleryconfig != nil {
 		for _, f := range galleryconfig.Files {
 			fullPath := filepath.Join(basePath, f.Filename)
-			log.Debug().Msgf("Removing file %s", fullPath)
-			if e := os.Remove(fullPath); e != nil {
-				err = errors.Join(err, fmt.Errorf("failed to remove file %s: %w", f.Filename, e))
-			}
+			filesToRemove = append(filesToRemove, fullPath)
 		}
 	}

 	for _, f := range additionalFiles {
 		fullPath := filepath.Join(filepath.Join(basePath, f))
-		log.Debug().Msgf("Removing additional file %s", fullPath)
-		if e := os.Remove(fullPath); e != nil {
+		filesToRemove = append(filesToRemove, fullPath)
+	}
+
+	filesToRemove = append(filesToRemove, configFile)
+	filesToRemove = append(filesToRemove, galleryFile)
+
+	// skip duplicates
+	filesToRemove = utils.Unique(filesToRemove)
+
+	// Removing files
+	for _, f := range filesToRemove {
+		if e := os.Remove(f); e != nil {
 			err = errors.Join(err, fmt.Errorf("failed to remove file %s: %w", f, e))
 		}
 	}

-	log.Debug().Msgf("Removing model config file %s", configFile)
-
-	// Delete the model config file
-	if e := os.Remove(configFile); e != nil {
-		err = errors.Join(err, fmt.Errorf("failed to remove file %s: %w", configFile, e))
-	}
-
-	// Delete gallery config file
-	os.Remove(galleryFile)
-
 	return err
 }

--- a/core/http/elements/gallery.go
+++ b/core/http/elements/gallery.go
@@ -9,7 +9,6 @@ import (
 	"github.com/mudler/LocalAI/core/gallery"
 	"github.com/mudler/LocalAI/core/p2p"
 	"github.com/mudler/LocalAI/core/services"
-	"github.com/mudler/LocalAI/pkg/xsync"
 )

 const (
@@ -372,7 +371,12 @@ func dropBadChars(s string) string {
 	return strings.ReplaceAll(s, "@", "__")
 }

-func ListModels(models []*gallery.GalleryModel, processing *xsync.SyncedMap[string, string], galleryService *services.GalleryService) string {
+type ProcessTracker interface {
+	Exists(string) bool
+	Get(string) string
+}
+
+func ListModels(models []*gallery.GalleryModel, processTracker ProcessTracker, galleryService *services.GalleryService) string {
 	modelsElements := []elem.Node{}
 	descriptionDiv := func(m *gallery.GalleryModel) elem.Node {
 		return elem.Div(
@@ -396,7 +400,7 @@ func ListModels(models []*gallery.GalleryModel, processing *xsync.SyncedMap[stri

 	actionDiv := func(m *gallery.GalleryModel) elem.Node {
 		galleryID := fmt.Sprintf("%s@%s", m.Gallery.Name, m.Name)
-		currentlyProcessing := processing.Exists(galleryID)
+		currentlyProcessing := processTracker.Exists(galleryID)
 		jobID := ""
 		isDeletionOp := false
 		if currentlyProcessing {
@@ -404,7 +408,7 @@ func ListModels(models []*gallery.GalleryModel, processing *xsync.SyncedMap[stri
 			if status != nil && status.Deletion {
 				isDeletionOp = true
 			}
-			jobID = processing.Get(galleryID)
+			jobID = processTracker.Get(galleryID)
 			// TODO:
 			// case not handled, if status == nil : "Waiting"
 		}
--- a/core/http/endpoints/openai/chat.go
+++ b/core/http/endpoints/openai/chat.go
@@ -226,9 +226,15 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup

 			// Update input grammar
 			jsStruct := funcs.ToJSONStructure(config.FunctionsConfig.FunctionNameKey, config.FunctionsConfig.FunctionNameKey)
-			config.Grammar = jsStruct.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
+			g, err := jsStruct.Grammar(config.FunctionsConfig.GrammarOptions()...)
+			if err == nil {
+				config.Grammar = g
+			}
 		case input.JSONFunctionGrammarObject != nil:
-			config.Grammar = input.JSONFunctionGrammarObject.Grammar(config.FunctionsConfig.GrammarConfig.Options()...)
+			g, err := input.JSONFunctionGrammarObject.Grammar(config.FunctionsConfig.GrammarOptions()...)
+			if err == nil {
+				config.Grammar = g
+			}
 		default:
 			// Force picking one of the functions by the request
 			if config.FunctionToCall() != "" {
--- a/core/http/routes/ui.go
+++ b/core/http/routes/ui.go
@@ -21,6 +21,40 @@ import (
 	"github.com/google/uuid"
 )

+type modelOpCache struct {
+	status *xsync.SyncedMap[string, string]
+}
+
+func NewModelOpCache() *modelOpCache {
+	return &modelOpCache{
+		status: xsync.NewSyncedMap[string, string](),
+	}
+}
+
+func (m *modelOpCache) Set(key string, value string) {
+	m.status.Set(key, value)
+}
+
+func (m *modelOpCache) Get(key string) string {
+	return m.status.Get(key)
+}
+
+func (m *modelOpCache) DeleteUUID(uuid string) {
+	for _, k := range m.status.Keys() {
+		if m.status.Get(k) == uuid {
+			m.status.Delete(k)
+		}
+	}
+}
+
+func (m *modelOpCache) Map() map[string]string {
+	return m.status.Map()
+}
+
+func (m *modelOpCache) Exists(key string) bool {
+	return m.status.Exists(key)
+}
+
 func RegisterUIRoutes(app *fiber.App,
 	cl *config.BackendConfigLoader,
 	ml *model.ModelLoader,
@@ -29,7 +63,7 @@ func RegisterUIRoutes(app *fiber.App,
 	auth func(*fiber.Ctx) error) {

 	// keeps the state of models that are being installed from the UI
-	var processingModels = xsync.NewSyncedMap[string, string]()
+	var processingModels = NewModelOpCache()

 	// modelStatus returns the current status of the models being processed (installation or deletion)
 	// it is called asynchonously from the UI
@@ -232,6 +266,8 @@ func RegisterUIRoutes(app *fiber.App,
 			return c.SendString(elements.ProgressBar("100"))
 		}
 		if status.Error != nil {
+			// TODO: instead of deleting the job, we should keep it in the cache and make it dismissable
+			processingModels.DeleteUUID(jobUID)
 			return c.SendString(elements.ErrorProgress(status.Error.Error(), status.GalleryModelName))
 		}

@@ -246,12 +282,7 @@ func RegisterUIRoutes(app *fiber.App,
 		status := galleryService.GetStatus(jobUID)

 		galleryID := ""
-		for _, k := range processingModels.Keys() {
-			if processingModels.Get(k) == jobUID {
-				galleryID = k
-				processingModels.Delete(k)
-			}
-		}
+		processingModels.DeleteUUID(jobUID)
 		if galleryID == "" {
 			log.Debug().Msgf("no processing model found for job : %+v\n", jobUID)
 		}
--- a/core/http/views/p2p.html
+++ b/core/http/views/p2p.html
@@ -16,7 +16,16 @@
                </a> 
            </h2> 
            <h5 class="mb-4 text-justify">LocalAI uses P2P technologies to enable distribution of work between peers. It is possible to share an instance with Federation and/or split the weights of a model across peers (only available with llama.cpp models). You can now share computational resources between your devices or your friends!</h5>
-            
+            <!-- Warning box if p2p token is empty and p2p is enabled -->
+            {{ if and .IsP2PEnabled (eq .P2PToken "") }}
+            <div class="bg-red-500 p-4 rounded-lg shadow-lg mb-12 text-left">
+                <p class="text-xl font-semibold text-white"> <i class="fa-solid fa-exclamation-triangle"></i> Warning: P2P mode is disabled or no token was specified</p>
+                <p class="mb-4">You have to enable P2P mode by starting LocalAI with <code>--p2p</code>. Please restart the server with <code>--p2p</code> to generate a new token automatically that can be used to automatically discover other nodes. If you already have a token specify it with <code>export TOKEN=".."</code> <a href="https://localai.io/features/distribute/" target="_blank">
+                    Check out the documentation for more information.
+                </a> </p>
+            </div>
+            {{ else }}
+
            <!-- Federation Box -->
            <div class="bg-gray-800 p-6 rounded-lg shadow-lg mb-12 text-left">

@@ -128,7 +137,8 @@
                    </div>
                </div>
            </div>
-            <!-- Llama.cpp Box END -->       
+            <!-- Llama.cpp Box END -->    
+            {{ end }}   
        </div>
    </div>

--- a/docs/content/docs/features/distributed_inferencing.md
+++ b/docs/content/docs/features/distributed_inferencing.md
@@ -5,17 +5,65 @@ weight = 15
 url = "/features/distribute/"
 +++

+
+This functionality enables LocalAI to distribute inference requests across multiple worker nodes, improving efficiency and performance. Nodes are automatically discovered and connect via p2p by using a shared token which makes sure the communication is secure and private between the nodes of the network.
+
+LocalAI supports two modes of distributed inferencing via p2p:
+
+- **Federated Mode**: Requests are shared between the cluster and routed to a single worker node in the network based on the load balancer's decision.
+- **Worker Mode** (aka "model sharding" or "splitting weights"): Requests are processed by all the workers which contributes to the final inference result (by sharing the model weights).
+
+## Usage
+
+Starting LocalAI with `--p2p` generates a shared token for connecting multiple instances: and that's all you need to create AI clusters, eliminating the need for intricate network setups. 
+
+Simply navigate to the "Swarm" section in the WebUI and follow the on-screen instructions.
+
+For fully shared instances, initiate LocalAI with --p2p --federated and adhere to the Swarm section's guidance. This feature, while still experimental, offers a tech preview quality experience.
+
+### Federated mode
+
+Federated mode allows to launch multiple LocalAI instances and connect them together in a federated network. This mode is useful when you want to distribute the load of the inference across multiple nodes, but you want to have a single point of entry for the API. In the Swarm section of the WebUI, you can see the instructions to connect multiple instances together.
+
+![346663124-1d2324fd-8b55-4fa2-9856-721a467969c2](https://github.com/user-attachments/assets/19ebd44a-20ff-412c-b92f-cfb8efbe4b21)
+
+To start a LocalAI server in federated mode, run:
+
+```bash
+local-ai run --p2p --federated
+```
+
+This will generate a token that you can use to connect other LocalAI instances to the network or others can use to join the network. If you already have a token, you can specify it using the `TOKEN` environment variable.
+
+To start a load balanced server that routes the requests to the network, run with the `TOKEN`:
+
+```bash
+local-ai federated
+```
+
+To see all the available options, run `local-ai federated --help`.
+
+The instructions are displayed in the "Swarm" section of the WebUI, guiding you through the process of connecting multiple instances.
+
+### Workers mode
+
 {{% alert note %}}
 This feature is available exclusively with llama-cpp compatible models.

 This feature was introduced in [LocalAI pull request #2324](https://github.com/mudler/LocalAI/pull/2324) and is based on the upstream work in [llama.cpp pull request #6829](https://github.com/ggerganov/llama.cpp/pull/6829).
 {{% /alert %}}

-This functionality enables LocalAI to distribute inference requests across multiple worker nodes, improving efficiency and performance.
+To connect multiple workers to a single LocalAI instance, start first a server in p2p mode:

-## Usage
+```bash
+local-ai run --p2p
+```

-### Starting Workers
+And navigate the WebUI to the "Swarm" section to see the instructions to connect multiple workers to the network.
+
+![346663124-1d2324fd-8b55-4fa2-9856-721a467969c2](https://github.com/user-attachments/assets/b8cadddf-a467-49cf-a1ed-8850de95366d)
+
+### Without P2P

 To start workers for distributing the computational load, run:

@@ -23,48 +71,27 @@ To start workers for distributing the computational load, run:
 local-ai worker llama-cpp-rpc <listening_address> <listening_port>
 ```

-Alternatively, you can build the RPC server following the llama.cpp [README](https://github.com/ggerganov/llama.cpp/blob/master/examples/rpc/README.md), which is compatible with LocalAI.
-
-### Starting LocalAI
-
-To start the LocalAI server, which handles API requests, specify the worker addresses using the `LLAMACPP_GRPC_SERVERS` environment variable:
+And you can specify the address of the workers when starting LocalAI with the `LLAMACPP_GRPC_SERVERS` environment variable:

 ```bash
 LLAMACPP_GRPC_SERVERS="address1:port,address2:port" local-ai run
 ```
-
 The workload on the LocalAI server will then be distributed across the specified nodes.

-## Peer-to-Peer Networking
+Alternatively, you can build the RPC workers/server following the llama.cpp [README](https://github.com/ggerganov/llama.cpp/blob/master/examples/rpc/README.md), which is compatible with LocalAI.

-![output](https://github.com/mudler/LocalAI/assets/2420543/8ca277cf-c208-4562-8929-808b2324b584)
+## Manual example (worker)

-Workers can also connect to each other in a peer-to-peer network, distributing the workload in a decentralized manner.
-
-A shared token between the server and the workers is required for communication within the peer-to-peer network. This feature supports both local network (using mDNS discovery) and DHT for communication across different networks.
-
-The token is automatically generated when starting the server with the `--p2p` flag. Workers can be started with the token using `local-ai worker p2p-llama-cpp-rpc` and specifying the token via the environment variable `TOKEN` or with the `--token` argument.
-
-A network is established between the server and workers using DHT and mDNS discovery protocols. The llama.cpp RPC server is automatically started and exposed to the peer-to-peer network, allowing the API server to connect.
-
-When the HTTP server starts, it discovers workers in the network and creates port forwards to the local service. Llama.cpp is configured to use these services. For more details on the implementation, refer to [LocalAI pull request #2343](https://github.com/mudler/LocalAI/pull/2343).
-
-### Usage
+Use the WebUI to guide you in the process of starting new workers. This example shows the manual steps to highlight the process.

 1. Start the server with `--p2p`:

 ```bash
 ./local-ai run --p2p
-# 1:02AM INF loading environment variables from file envFile=.env
-# 1:02AM INF Setting logging to info
-# 1:02AM INF P2P mode enabled
-# 1:02AM INF No token provided, generating one
-# 1:02AM INF Generated Token:
-# XXXXXXXXXXX
-# 1:02AM INF Press a button to proceed
+# Get the token in the Swarm section of the WebUI
 ```

-Copy the displayed token and press Enter.
+Copy the token from the WebUI or via API call (e.g., `curl http://localhost:8000/p2p/token`) and save it for later use.

 To reuse the same token later, restart the server with `--p2ptoken` or `P2P_TOKEN`.

@@ -93,11 +120,7 @@ The server logs should indicate that new workers are being discovered.

 3. Start inference as usual on the server initiated in step 1.

-## Notes
-
- If running in p2p mode with container images, make sure you start the container with `--net host` or `network_mode: host` in the docker-compose file.
- Only a single model is supported currently.
- Ensure the server detects new workers before starting inference. Currently, additional workers cannot be added once inference has begun.
+![output](https://github.com/mudler/LocalAI/assets/2420543/8ca277cf-c208-4562-8929-808b2324b584)


 ## Environment Variables
@@ -109,3 +132,20 @@ There are options that can be tweaked or parameters that can be set using enviro
 | **LOCALAI_P2P_DISABLE_DHT** | Set to "true" to disable DHT and enable p2p layer to be local only (mDNS) |
 | **LOCALAI_P2P_DISABLE_LIMITS** | Set to "true" to disable connection limits and resources management |
 | **LOCALAI_P2P_TOKEN** | Set the token for the p2p network |
+
+## Architecture
+
+LocalAI uses https://github.com/libp2p/go-libp2p under the hood, the same project powering IPFS. Differently from other frameworks, LocalAI uses peer2peer without a single master server, but rather it uses sub/gossip and ledger functionalities to achieve consensus across different peers. 
+
+[EdgeVPN](https://github.com/mudler/edgevpn) is used as a library to establish the network and expose the ledger functionality under a shared token to ease out automatic discovery and have separated, private peer2peer networks.
+
+The weights are split proportional to the memory when running into worker mode, when in federation mode each request is split to every node which have to load the model fully.
+
+## Notes
+
+- If running in p2p mode with container images, make sure you start the container with `--net host` or `network_mode: host` in the docker-compose file.
+- Only a single model is supported currently.
+- Ensure the server detects new workers before starting inference. Currently, additional workers cannot be added once inference has begun.
+- For more details on the implementation, refer to [LocalAI pull request #2343](https://github.com/mudler/LocalAI/pull/2343)
+
+
--- a/docs/data/version.json
+++ b/docs/data/version.json
@@ -1,3 +1,3 @@
 {
-  "version": "v2.18.1"
+  "version": "v2.19.3"
 }
--- a/docs/static/install.sh
+++ b/docs/static/install.sh
@@ -194,7 +194,7 @@ install_container_toolkit_yum() {
    curl -s -L https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo | \
    $SUDO  tee /etc/yum.repos.d/nvidia-container-toolkit.repo

-    if [ "$PACKAGE_MANAGER" == "dnf" ]; then
+    if [ "$PACKAGE_MANAGER" = "dnf" ]; then
        $SUDO $PACKAGE_MANAGER config-manager --enable nvidia-container-toolkit-experimental
    else 
        $SUDO $PACKAGE_MANAGER -y install yum-utils
@@ -629,7 +629,7 @@ case "$ARCH" in
    *) fatal "Unsupported architecture: $ARCH" ;;
 esac

-if [ "$OS" == "Darwin" ]; then
+if [ "$OS" = "Darwin" ]; then
    install_binary_darwin
    exit 0
 fi
--- a/docs/themes/hugo-theme-relearn
+++ b/docs/themes/hugo-theme-relearn
--- a/examples/chainlit/requirements.txt
+++ b/examples/chainlit/requirements.txt
@@ -1,6 +1,6 @@
-llama_index==0.10.55
+llama_index==0.10.56
 requests==2.32.3
-weaviate_client==4.6.5
+weaviate_client==4.6.7
 transformers
 torch
 chainlit
--- a/examples/functions/requirements.txt
+++ b/examples/functions/requirements.txt
@@ -1,2 +1,2 @@
-langchain==0.2.8
-openai==1.35.13
+langchain==0.2.10
+openai==1.37.0
--- a/examples/langchain-chroma/requirements.txt
+++ b/examples/langchain-chroma/requirements.txt
@@ -1,4 +1,4 @@
-langchain==0.2.8
-openai==1.35.13
-chromadb==0.5.4
-llama-index==0.10.55
+langchain==0.2.10
+openai==1.37.0
+chromadb==0.5.5
+llama-index==0.10.56
--- a/examples/langchain/langchainpy-localai-example/requirements.txt
+++ b/examples/langchain/langchainpy-localai-example/requirements.txt
@@ -10,21 +10,21 @@ debugpy==1.8.2
 frozenlist==1.4.1
 greenlet==3.0.3
 idna==3.7
-langchain==0.2.8
-langchain-community==0.2.7
+langchain==0.2.11
+langchain-community==0.2.9
 marshmallow==3.21.3
 marshmallow-enum==1.5.1
 multidict==6.0.5
 mypy-extensions==1.0.0
 numexpr==2.10.1
-numpy==1.26.4
-openai==1.35.13
+numpy==2.0.1
+openai==1.37.1
 openapi-schema-pydantic==1.2.4
 packaging>=23.2
 pydantic==2.8.2
 PyYAML==6.0.1
 requests==2.32.3
-SQLAlchemy==2.0.30
+SQLAlchemy==2.0.31
 tenacity==8.5.0
 tqdm==4.66.4
 typing-inspect==0.9.0
--- a/examples/streamlit-bot/requirements.txt
+++ b/examples/streamlit-bot/requirements.txt
@@ -1,2 +1,2 @@
-streamlit==1.36.0
+streamlit==1.37.0
 requests
--- a/gallery/alpaca.yaml
+++ b/gallery/alpaca.yaml
@@ -0,0 +1,17 @@
+---
+name: "alpaca"
+
+config_file: |
+  context_size: 4096
+  f16: true
+  mmap: true
+  template:
+    chat: |
+        Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+        ### Instruction:
+        {{.Input}}
+
+        ### Response:
+    completion: |
+        {{.Input}}
--- a/gallery/index.yaml
+++ b/gallery/index.yaml
--- a/gallery/llama3-instruct.yaml
+++ b/gallery/llama3-instruct.yaml
@@ -31,7 +31,7 @@ config_file: |
      {'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
      Function call:
    chat: |
-      <|begin_of_text|>{{.Input }}
+      {{.Input }}
      <|start_header_id|>assistant<|end_header_id|>
    completion: |
      {{.Input}}
--- a/gallery/llama3.1-instruct-grammar.yaml
+++ b/gallery/llama3.1-instruct-grammar.yaml
@@ -0,0 +1,64 @@
+---
+name: "llama3-instruct-grammar"
+
+config_file: |
+  mmap: true
+  function:
+    disable_no_action: true
+    grammar:
+      no_mixed_free_string: true
+      mixed_mode: true
+      schema_type: llama3.1 # or JSON is supported too (json)
+    response_regex:
+    - <function=(?P<name>\w+)>(?P<arguments>.*)</function>
+  template:
+    chat_message: |
+      <|start_header_id|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}<|end_header_id|>
+
+      {{ if .FunctionCall -}}
+      Function call:
+      {{ else if eq .RoleName "tool" -}}
+      Function response:
+      {{ end -}}
+      {{ if .Content -}}
+      {{.Content -}}
+      {{ else if .FunctionCall -}}
+      {{ toJson .FunctionCall -}}
+      {{ end -}}
+      <|eot_id|>
+    function: |
+      <|start_header_id|>system<|end_header_id|>
+
+      You have access to the following functions:
+
+      {{range .Functions}}
+      Use the function '{{.Name}}' to '{{.Description}}'
+      {{toJson .Parameters}}
+      {{end}}
+
+      Think very carefully before calling functions.
+      If a you choose to call a function ONLY reply in the following format with no prefix or suffix:
+
+      <function=example_function_name>{{`{{"example_name": "example_value"}}`}}</function>
+
+      Reminder:
+      - If looking for real time information use relevant functions before falling back to searching on internet
+      - Function calls MUST follow the specified format, start with <function= and end with </function>
+      - Required parameters MUST be specified
+      - Only call one function at a time
+      - Put the entire function call reply on one line
+      <|eot_id|>
+      {{.Input }}
+      <|start_header_id|>assistant<|end_header_id|>
+    chat: |
+      {{.Input }}
+      <|start_header_id|>assistant<|end_header_id|>
+    completion: |
+      {{.Input}}
+  context_size: 8192
+  f16: true
+  stopwords:
+  - <|im_end|>
+  - <dummy32000>
+  - "<|eot_id|>"
+  - <|end_of_text|>
--- a/gallery/llama3.1-instruct.yaml
+++ b/gallery/llama3.1-instruct.yaml
@@ -0,0 +1,62 @@
+---
+name: "llama3-instruct"
+
+config_file: |
+  mmap: true
+  function:
+    disable_no_action: true
+    grammar:
+      disable: true
+    response_regex:
+    - <function=(?P<name>\w+)>(?P<arguments>.*)</function>
+  template:
+    chat_message: |
+      <|start_header_id|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}<|end_header_id|>
+
+      {{ if .FunctionCall -}}
+      Function call:
+      {{ else if eq .RoleName "tool" -}}
+      Function response:
+      {{ end -}}
+      {{ if .Content -}}
+      {{.Content -}}
+      {{ else if .FunctionCall -}}
+      {{ toJson .FunctionCall -}}
+      {{ end -}}
+      <|eot_id|>
+    function: |
+      <|start_header_id|>system<|end_header_id|>
+
+      You have access to the following functions:
+
+      {{range .Functions}}
+      Use the function '{{.Name}}' to '{{.Description}}'
+      {{toJson .Parameters}}
+      {{end}}
+
+      Think very carefully before calling functions.
+      If a you choose to call a function ONLY reply in the following format with no prefix or suffix:
+
+      <function=example_function_name>{{`{{"example_name": "example_value"}}`}}</function>
+
+      Reminder:
+      - If looking for real time information use relevant functions before falling back to searching on internet
+      - Function calls MUST follow the specified format, start with <function= and end with </function>
+      - Required parameters MUST be specified
+      - Only call one function at a time
+      - Put the entire function call reply on one line
+      <|eot_id|>
+      {{.Input }}
+      <|start_header_id|>assistant<|end_header_id|>
+    chat: |
+      {{.Input }}
+      <|start_header_id|>assistant<|end_header_id|>
+    completion: |
+      {{.Input}}
+  context_size: 8192
+  f16: true
+  stopwords:
+  - <|im_end|>
+  - <dummy32000>
+  - "<|eot_id|>"
+  - <|end_of_text|>
--- a/gallery/tuluv2.yaml
+++ b/gallery/tuluv2.yaml
@@ -0,0 +1,43 @@
+---
+name: "tuluv2"
+
+config_file: |
+  mmap: true
+  template:
+    chat_message: |
+      <|{{ .RoleName }}|>
+      {{ if .FunctionCall -}}
+      Function call:
+      {{ else if eq .RoleName "tool" -}}
+      Function response:
+      {{ end -}}
+      {{ if .Content -}}
+      {{.Content }}
+      {{ end -}}
+      {{ if .FunctionCall -}}
+      {{toJson .FunctionCall}}
+      {{ end -}}
+    function: |
+      <|{{ .RoleName }}|>
+      {{ if .FunctionCall -}}
+      Function call:
+      {{ else if eq .RoleName "tool" -}}
+      Function response:
+      {{ end -}}
+      {{ if .Content -}}
+      {{.Content }}
+      {{ end -}}
+      {{ if .FunctionCall -}}
+      {{toJson .FunctionCall}}
+      {{ end -}}
+    chat: |
+      {{.Input -}}
+      <|assistant|>
+    completion: |
+      {{.Input}}
+  context_size: 4096
+  f16: true
+  stopwords:
+  - '<|im_end|>'
+  - '<dummy32000>'
+  - '<|endoftext|>'
--- a/pkg/functions/function_structure.go
+++ b/pkg/functions/function_structure.go
@@ -0,0 +1,43 @@
+package functions
+
+import (
+	"encoding/json"
+
+	"github.com/mudler/LocalAI/pkg/functions/grammars"
+)
+
+type Item struct {
+	Type       string                 `json:"type"`
+	Properties map[string]interface{} `json:"properties"`
+}
+
+type JSONFunctionStructure struct {
+	OneOf []Item                 `json:"oneOf,omitempty"`
+	AnyOf []Item                 `json:"anyOf,omitempty"`
+	Defs  map[string]interface{} `json:"$defs,omitempty"`
+}
+
+func (j JSONFunctionStructure) Grammar(options ...func(*grammars.GrammarOption)) (string, error) {
+	grammarOpts := &grammars.GrammarOption{}
+	grammarOpts.Apply(options...)
+
+	dat, err := json.Marshal(j)
+	if err != nil {
+		return "", err
+	}
+
+	converter := NewSchemaConverter(*grammarOpts)
+	return converter.GrammarFromBytes(dat, options...)
+}
+
+type SchemaConverter interface {
+	GrammarFromBytes([]byte, ...func(*grammars.GrammarOption)) (string, error)
+}
+
+func NewSchemaConverter(opt grammars.GrammarOption) SchemaConverter {
+	switch {
+	case opt.SchemaType == grammars.LLama31Schema:
+		return grammars.NewLLama31SchemaConverter(opt.FunctionName)
+	}
+	return grammars.NewJSONSchemaConverter(opt.PropOrder)
+}
--- a/pkg/functions/functions.go
+++ b/pkg/functions/functions.go
@@ -18,6 +18,15 @@ type Function struct {
 }
 type Functions []Function

+type FunctionName struct {
+	Const string `json:"const"`
+}
+
+type Argument struct {
+	Type       string                 `json:"type"`
+	Properties map[string]interface{} `json:"properties"`
+}
+
 type Tool struct {
 	Type     string   `json:"type"`
 	Function Function `json:"function,omitempty"`
--- a/pkg/functions/functions_suite_test.go
+++ b/pkg/functions/functions_suite_test.go
@@ -1,4 +1,4 @@
-package functions
+package functions_test

 import (
 	"testing"
@@ -7,7 +7,7 @@ import (
 	. "github.com/onsi/gomega"
 )

-func TestGrammar(t *testing.T) {
+func TestFunctions(t *testing.T) {
 	RegisterFailHandler(Fail)
-	RunSpecs(t, "Grammar test suite")
+	RunSpecs(t, "Functions test suite")
 }
--- a/pkg/functions/grammar_json_schema.go
+++ b/pkg/functions/grammar_json_schema.go
@@ -1,378 +0,0 @@
-package functions
-
-// a golang port of https://github.com/ggerganov/llama.cpp/pull/1887
-
-import (
-	"encoding/json"
-	"fmt"
-	"regexp"
-	"sort"
-	"strings"
-
-	"github.com/mudler/LocalAI/pkg/utils"
-)
-
-const (
-	JSONBNF = `root   ::= object
-value  ::= object | array | string | number | ("true" | "false" | "null") ws
-
-object ::=
-  "{" ws (
-            string ":" ws value
-    ("," ws string ":" ws value)*
-  )? "}" ws
-
-array  ::=
-  "[" ws (
-            value
-    ("," ws value)*
-  )? "]" ws
-
-string ::=
-  "\"" (
-    [^"\\] |
-    "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
-  )* "\"" ws
-
-number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
-
-ws ::= ([ \t\n] ws)?`
-)
-
-var (
-	SPACE_RULE = `" "?`
-
-	PRIMITIVE_RULES = map[string]string{
-		"boolean": `("true" | "false") space`,
-		"number":  `("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? space`,
-		"integer": `("-"? ([0-9] | [1-9] [0-9]*)) space`,
-		"string": `"\"" (
-			[^"\\] |
-			"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
-		  )* "\"" space`,
-		// TODO: we shouldn't forbid \" and \\ or all unicode and have this branch here,
-		// however, if we don't have it, the grammar will be ambiguous and
-		// empirically results are way worse.
-		"freestring": `(
-			[^\x00] |
-			"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
-		  )* space`,
-		"null": `"null" space`,
-	}
-
-	INVALID_RULE_CHARS_RE     = regexp.MustCompile(`[^a-zA-Z0-9-]+`)
-	GRAMMAR_LITERAL_ESCAPE_RE = regexp.MustCompile(`[\r\n"]`)
-	GRAMMAR_LITERAL_ESCAPES   = map[string]string{
-		"\r": `\r`,
-		"\n": `\n`,
-		`"`:  `\"`,
-	}
-)
-
-type JSONSchemaConverter struct {
-	propOrder map[string]int
-	rules     map[string]string
-}
-
-func NewJSONSchemaConverter(propOrder string) *JSONSchemaConverter {
-	propOrderSlice := strings.Split(propOrder, ",")
-	propOrderMap := make(map[string]int)
-	for idx, name := range propOrderSlice {
-		propOrderMap[name] = idx
-	}
-
-	rules := make(map[string]string)
-	rules["space"] = SPACE_RULE
-
-	return &JSONSchemaConverter{
-		propOrder: propOrderMap,
-		rules:     rules,
-	}
-}
-
-func (sc *JSONSchemaConverter) formatLiteral(literal interface{}) string {
-	escaped := GRAMMAR_LITERAL_ESCAPE_RE.ReplaceAllStringFunc(jsonString(literal), func(match string) string {
-		return GRAMMAR_LITERAL_ESCAPES[match]
-	})
-	return fmt.Sprintf(`"%s"`, escaped)
-}
-
-func (sc *JSONSchemaConverter) addRule(name, rule string) string {
-	escName := INVALID_RULE_CHARS_RE.ReplaceAllString(name, "-")
-	key := escName
-	if existingRule, ok := sc.rules[escName]; ok && existingRule != rule {
-		i := 0
-		for {
-			key = fmt.Sprintf("%s%d", escName, i)
-			if _, ok := sc.rules[key]; !ok {
-				break
-			}
-			i++
-		}
-	}
-	sc.rules[key] = rule
-	return key
-}
-
-const arrayNewLines = `arr  ::=
-  "[\n"  (
-		realvalue
-    (",\n"  realvalue)*
-  )? "]"`
-
-const array = `arr  ::=
-  "["  (
-		realvalue
-    (","  realvalue)*
-  )? "]"`
-
-func (sc *JSONSchemaConverter) finalizeGrammar(options ...func(*GrammarOption)) string {
-
-	grammarOpts := &GrammarOption{}
-	grammarOpts.Apply(options...)
-
-	prefix := grammarOpts.Prefix
-	maybeArray := grammarOpts.MaybeArray
-	disableParallelNewLines := grammarOpts.DisableParallelNewLines
-	maybeString := grammarOpts.MaybeString
-	noMixedFreeString := grammarOpts.NoMixedFreeString
-
-	var lines []string
-
-	swapRoot := maybeArray || maybeString || prefix != ""
-
-	// write down the computed rules.
-	// if maybeArray is true, we need to add the array rule and slightly tweak the root rule
-	for name, rule := range sc.rules {
-		if swapRoot && name == "root" {
-			name = "realvalue"
-		}
-		lines = append(lines, fmt.Sprintf("%s ::= %s", name, rule))
-	}
-
-	if !swapRoot {
-		return strings.Join(lines, "\n")
-	}
-
-	newRoot := "realvalue"
-	if maybeArray {
-		newRoot = "arr | realvalue"
-	}
-
-	freestringRule := "mixedstring"
-	if noMixedFreeString {
-		freestringRule = "freestring"
-	}
-
-	if prefix != "" {
-		// quote newlines in suffix
-		prefix = utils.EscapeNewLines(prefix)
-
-		if maybeArray && maybeString {
-			newRoot = "(" + newRoot + ")"
-		}
-
-		if maybeString {
-			//newRoot = "( (\"" + suffix + "\" " + newRoot + ") | freestring ) "
-			newRoot = "( \"" + prefix + "\" " + newRoot + " | " + freestringRule + " ) "
-		} else {
-			newRoot = "\"" + prefix + "\" " + "" + newRoot + ""
-		}
-	} else if maybeString {
-		if maybeArray {
-			//	newRoot = "(" + newRoot + ")"
-		}
-
-		newRoot = freestringRule + " | " + newRoot
-	}
-
-	lines = append(lines, fmt.Sprintf("%s ::= %s", "root", newRoot))
-	if disableParallelNewLines {
-		lines = append(lines, array)
-	} else {
-		lines = append(lines, arrayNewLines)
-	}
-
-	if maybeArray {
-		if grammarOpts.ExpectStringsAfterJSON {
-			lines = append(lines, `mixedstring ::= freestring | freestring arr freestring | (freestring realvalue freestring)* | realvalue | arr`)
-		} else {
-			lines = append(lines, `mixedstring ::= freestring | freestring arr | freestring realvalue | realvalue | arr`)
-		}
-	} else {
-		if grammarOpts.ExpectStringsAfterJSON {
-			lines = append(lines, `mixedstring ::= freestring | (freestring realvalue freestring)* | realvalue`)
-		} else {
-			lines = append(lines, `mixedstring ::= freestring | freestring realvalue | realvalue`)
-		}
-	}
-
-	return strings.Join(lines, "\n")
-}
-
-func (sc *JSONSchemaConverter) visit(schema map[string]interface{}, name string, rootSchema map[string]interface{}) string {
-	st, existType := schema["type"]
-	var schemaType string
-	if existType {
-		schemaType = st.(string)
-	}
-	ruleName := name
-	if name == "" {
-		ruleName = "root"
-	}
-	_, oneOfExists := schema["oneOf"]
-	_, anyOfExists := schema["anyOf"]
-	if oneOfExists || anyOfExists {
-		var alternatives []string
-		oneOfSchemas, oneOfExists := schema["oneOf"].([]interface{})
-		anyOfSchemas, anyOfExists := schema["anyOf"].([]interface{})
-
-		if oneOfExists {
-			for i, altSchema := range oneOfSchemas {
-				alternative := sc.visit(altSchema.(map[string]interface{}), fmt.Sprintf("%s-%d", ruleName, i), rootSchema)
-				alternatives = append(alternatives, alternative)
-			}
-		} else if anyOfExists {
-			for i, altSchema := range anyOfSchemas {
-				alternative := sc.visit(altSchema.(map[string]interface{}), fmt.Sprintf("%s-%d", ruleName, i), rootSchema)
-				alternatives = append(alternatives, alternative)
-			}
-		}
-
-		rule := strings.Join(alternatives, " | ")
-		return sc.addRule(ruleName, rule)
-	} else if ref, exists := schema["$ref"].(string); exists {
-		referencedSchema := sc.resolveReference(ref, rootSchema)
-		return sc.visit(referencedSchema, name, rootSchema)
-	} else if constVal, exists := schema["const"]; exists {
-		return sc.addRule(ruleName, sc.formatLiteral(constVal))
-	} else if enumVals, exists := schema["enum"].([]interface{}); exists {
-		var enumRules []string
-		for _, enumVal := range enumVals {
-			enumRule := sc.formatLiteral(enumVal)
-			enumRules = append(enumRules, enumRule)
-		}
-		rule := strings.Join(enumRules, " | ")
-		return sc.addRule(ruleName, rule)
-	} else if properties, exists := schema["properties"].(map[string]interface{}); schemaType == "object" && exists {
-		propOrder := sc.propOrder
-		var propPairs []struct {
-			propName   string
-			propSchema map[string]interface{}
-		}
-
-		for propName, propSchema := range properties {
-			propPairs = append(propPairs, struct {
-				propName   string
-				propSchema map[string]interface{}
-			}{propName: propName, propSchema: propSchema.(map[string]interface{})})
-		}
-
-		sort.Slice(propPairs, func(i, j int) bool {
-			iOrder := propOrder[propPairs[i].propName]
-			jOrder := propOrder[propPairs[j].propName]
-			if iOrder != 0 && jOrder != 0 {
-				return iOrder < jOrder
-			}
-			return propPairs[i].propName < propPairs[j].propName
-		})
-
-		var rule strings.Builder
-		rule.WriteString(`"{" space`)
-
-		for i, propPair := range propPairs {
-			propName := propPair.propName
-			propSchema := propPair.propSchema
-			propRuleName := sc.visit(propSchema, fmt.Sprintf("%s-%s", ruleName, propName), rootSchema)
-
-			if i > 0 {
-				rule.WriteString(` "," space`)
-			}
-
-			rule.WriteString(fmt.Sprintf(` %s space ":" space %s`, sc.formatLiteral(propName), propRuleName))
-		}
-
-		rule.WriteString(` "}" space`)
-		return sc.addRule(ruleName, rule.String())
-	} else if items, exists := schema["items"].(map[string]interface{}); schemaType == "array" && exists {
-		itemRuleName := sc.visit(items, fmt.Sprintf("%s-item", ruleName), rootSchema)
-		rule := fmt.Sprintf(`"[" space (%s ("," space %s)*)? "]" space`, itemRuleName, itemRuleName)
-		return sc.addRule(ruleName, rule)
-	} else {
-		primitiveRule, exists := PRIMITIVE_RULES[schemaType]
-		if !exists {
-			panic(fmt.Sprintf("Unrecognized schema: %v", schema))
-		}
-		if ruleName == "root" {
-			schemaType = "root"
-		}
-		return sc.addRule(schemaType, primitiveRule)
-	}
-}
-func (sc *JSONSchemaConverter) resolveReference(ref string, rootSchema map[string]interface{}) map[string]interface{} {
-	if !strings.HasPrefix(ref, "#/$defs/") {
-		panic(fmt.Sprintf("Invalid reference format: %s", ref))
-	}
-
-	defKey := strings.TrimPrefix(ref, "#/$defs/")
-	definitions, exists := rootSchema["$defs"].(map[string]interface{})
-	if !exists {
-		fmt.Println(rootSchema)
-
-		panic("No definitions found in the schema")
-	}
-
-	def, exists := definitions[defKey].(map[string]interface{})
-	if !exists {
-		fmt.Println(definitions)
-
-		panic(fmt.Sprintf("Definition not found: %s", defKey))
-	}
-
-	return def
-}
-
-func (sc *JSONSchemaConverter) Grammar(schema map[string]interface{}, options ...func(*GrammarOption)) string {
-	sc.addRule("freestring", PRIMITIVE_RULES["freestring"])
-	sc.visit(schema, "", schema)
-	return sc.finalizeGrammar(options...)
-}
-
-func (sc *JSONSchemaConverter) GrammarFromBytes(b []byte, options ...func(*GrammarOption)) string {
-	var schema map[string]interface{}
-	_ = json.Unmarshal(b, &schema)
-	return sc.Grammar(schema, options...)
-}
-
-func jsonString(v interface{}) string {
-	b, _ := json.Marshal(v)
-	return string(b)
-}
-
-type FunctionName struct {
-	Const string `json:"const"`
-}
-
-type Argument struct {
-	Type       string                 `json:"type"`
-	Properties map[string]interface{} `json:"properties"`
-}
-
-type Item struct {
-	Type       string                 `json:"type"`
-	Properties map[string]interface{} `json:"properties"`
-}
-
-type JSONFunctionStructure struct {
-	OneOf []Item                 `json:"oneOf,omitempty"`
-	AnyOf []Item                 `json:"anyOf,omitempty"`
-	Defs  map[string]interface{} `json:"$defs,omitempty"`
-}
-
-func (j JSONFunctionStructure) Grammar(options ...func(*GrammarOption)) string {
-	grammarOpts := &GrammarOption{}
-	grammarOpts.Apply(options...)
-
-	dat, _ := json.Marshal(j)
-	return NewJSONSchemaConverter(grammarOpts.PropOrder).GrammarFromBytes(dat, options...)
-}
--- a/pkg/functions/grammars/bnf_rules.go
+++ b/pkg/functions/grammars/bnf_rules.go
@@ -0,0 +1,58 @@
+package grammars
+
+import (
+	"encoding/json"
+	"regexp"
+)
+
+var (
+	PRIMITIVE_RULES = map[string]string{
+		"boolean": `("true" | "false") space`,
+		"number":  `("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? space`,
+		"integer": `("-"? ([0-9] | [1-9] [0-9]*)) space`,
+		"string": `"\"" (
+			[^"\\] |
+			"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
+		  )* "\"" space`,
+		// TODO: we shouldn't forbid \" and \\ or all unicode and have this branch here,
+		// however, if we don't have it, the grammar will be ambiguous and
+		// empirically results are way worse.
+		"freestring": `(
+			[^\x00] |
+			"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
+		  )* space`,
+		"null": `"null" space`,
+	}
+
+	INVALID_RULE_CHARS_RE     = regexp.MustCompile(`[^a-zA-Z0-9-]+`)
+	GRAMMAR_LITERAL_ESCAPE_RE = regexp.MustCompile(`[\r\n"]`)
+	GRAMMAR_LITERAL_ESCAPES   = map[string]string{
+		"\r": `\r`,
+		"\n": `\n`,
+		`"`:  `\"`,
+	}
+)
+
+const (
+	SPACE_RULE = `" "?`
+
+	arrayNewLines = `arr  ::=
+  "[\n"  (
+		realvalue
+    (",\n"  realvalue)*
+  )? "]"`
+
+	array = `arr  ::=
+  "["  (
+		realvalue
+    (","  realvalue)*
+  )? "]"`
+)
+
+func jsonString(v interface{}) (string, error) {
+	b, err := json.Marshal(v)
+	if err != nil {
+		return "", err
+	}
+	return string(b), nil
+}
--- a/pkg/functions/grammars/grammars_suite_test.go
+++ b/pkg/functions/grammars/grammars_suite_test.go
@@ -0,0 +1,25 @@
+package grammars_test
+
+import (
+	"testing"
+
+	. "github.com/mudler/LocalAI/pkg/functions"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func TestGrammar(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "Grammar test suite")
+}
+
+func createFunction(field1 string, field2 string, name string, properties map[string]interface{}) map[string]interface{} {
+	property := map[string]interface{}{}
+	property[field1] = FunctionName{Const: name}
+	property[field2] = Argument{
+		Type:       "object",
+		Properties: properties,
+	}
+	return property
+}
--- a/pkg/functions/grammars/json_schema.go
+++ b/pkg/functions/grammars/json_schema.go
@@ -0,0 +1,220 @@
+package grammars
+
+// a golang port of https://github.com/ggerganov/llama.cpp/pull/1887
+
+import (
+	"encoding/json"
+	"fmt"
+	"sort"
+	"strings"
+)
+
+type JSONSchemaConverter struct {
+	propOrder map[string]int
+	rules     Rules
+}
+
+func NewJSONSchemaConverter(propOrder string) *JSONSchemaConverter {
+	propOrderSlice := strings.Split(propOrder, ",")
+	propOrderMap := make(map[string]int)
+	for idx, name := range propOrderSlice {
+		propOrderMap[name] = idx
+	}
+
+	rules := make(map[string]string)
+	rules["space"] = SPACE_RULE
+
+	return &JSONSchemaConverter{
+		propOrder: propOrderMap,
+		rules:     rules,
+	}
+}
+
+func (sc *JSONSchemaConverter) formatLiteral(literal interface{}) (string, error) {
+	jLiteral, err := jsonString(literal)
+	if err != nil {
+		return "", err
+	}
+	escaped := GRAMMAR_LITERAL_ESCAPE_RE.ReplaceAllStringFunc(jLiteral, func(match string) string {
+		return GRAMMAR_LITERAL_ESCAPES[match]
+	})
+	return fmt.Sprintf(`"%s"`, escaped), nil
+}
+
+func (sc *JSONSchemaConverter) addRule(name, rule string) string {
+	escName := INVALID_RULE_CHARS_RE.ReplaceAllString(name, "-")
+	key := escName
+	if existingRule, ok := sc.rules[escName]; ok && existingRule != rule {
+		i := 0
+		for {
+			key = fmt.Sprintf("%s%d", escName, i)
+			if _, ok := sc.rules[key]; !ok {
+				break
+			}
+			i++
+		}
+	}
+	sc.rules[key] = rule
+	return key
+}
+
+func (sc *JSONSchemaConverter) visit(schema map[string]interface{}, name string, rootSchema map[string]interface{}) (string, error) {
+	st, existType := schema["type"]
+	var schemaType string
+	if existType {
+		schemaType = st.(string)
+	}
+	ruleName := name
+	if name == "" {
+		ruleName = "root"
+	}
+	_, oneOfExists := schema["oneOf"]
+	_, anyOfExists := schema["anyOf"]
+	if oneOfExists || anyOfExists {
+		var alternatives []string
+		oneOfSchemas, oneOfExists := schema["oneOf"].([]interface{})
+		anyOfSchemas, anyOfExists := schema["anyOf"].([]interface{})
+
+		if oneOfExists {
+			for i, altSchema := range oneOfSchemas {
+				alternative, err := sc.visit(altSchema.(map[string]interface{}), fmt.Sprintf("%s-%d", ruleName, i), rootSchema)
+				if err != nil {
+					return "", err
+				}
+				alternatives = append(alternatives, alternative)
+			}
+		} else if anyOfExists {
+			for i, altSchema := range anyOfSchemas {
+				alternative, err := sc.visit(altSchema.(map[string]interface{}), fmt.Sprintf("%s-%d", ruleName, i), rootSchema)
+				if err != nil {
+					return "", err
+				}
+				alternatives = append(alternatives, alternative)
+			}
+		}
+
+		rule := strings.Join(alternatives, " | ")
+		return sc.addRule(ruleName, rule), nil
+	} else if ref, exists := schema["$ref"].(string); exists {
+		referencedSchema, err := sc.resolveReference(ref, rootSchema)
+		if err != nil {
+			return "", err
+		}
+		return sc.visit(referencedSchema, name, rootSchema)
+	} else if constVal, exists := schema["const"]; exists {
+		literal, err := sc.formatLiteral((constVal))
+		if err != nil {
+			return "", err
+		}
+		return sc.addRule(ruleName, literal), nil
+	} else if enumVals, exists := schema["enum"].([]interface{}); exists {
+		var enumRules []string
+		for _, enumVal := range enumVals {
+			enumRule, err := sc.formatLiteral(enumVal)
+			if err != nil {
+				return "", err
+			}
+			enumRules = append(enumRules, enumRule)
+		}
+		rule := strings.Join(enumRules, " | ")
+		return sc.addRule(ruleName, rule), nil
+	} else if properties, exists := schema["properties"].(map[string]interface{}); schemaType == "object" && exists {
+		propOrder := sc.propOrder
+		var propPairs []struct {
+			propName   string
+			propSchema map[string]interface{}
+		}
+
+		for propName, propSchema := range properties {
+			propPairs = append(propPairs, struct {
+				propName   string
+				propSchema map[string]interface{}
+			}{propName: propName, propSchema: propSchema.(map[string]interface{})})
+		}
+
+		sort.Slice(propPairs, func(i, j int) bool {
+			iOrder := propOrder[propPairs[i].propName]
+			jOrder := propOrder[propPairs[j].propName]
+			if iOrder != 0 && jOrder != 0 {
+				return iOrder < jOrder
+			}
+			return propPairs[i].propName < propPairs[j].propName
+		})
+
+		var rule strings.Builder
+		rule.WriteString(`"{" space`)
+
+		for i, propPair := range propPairs {
+			propName := propPair.propName
+			propSchema := propPair.propSchema
+			propRuleName, err := sc.visit(propSchema, fmt.Sprintf("%s-%s", ruleName, propName), rootSchema)
+			if err != nil {
+				return "", err
+			}
+			lPropName, err := sc.formatLiteral(propName)
+			if err != nil {
+				return "", err
+			}
+			if i > 0 {
+				rule.WriteString(` "," space`)
+			}
+
+			rule.WriteString(fmt.Sprintf(` %s space ":" space %s`, lPropName, propRuleName))
+		}
+
+		rule.WriteString(` "}" space`)
+		return sc.addRule(ruleName, rule.String()), nil
+	} else if items, exists := schema["items"].(map[string]interface{}); schemaType == "array" && exists {
+		itemRuleName, err := sc.visit(items, fmt.Sprintf("%s-item", ruleName), rootSchema)
+		if err != nil {
+			return "", err
+		}
+		rule := fmt.Sprintf(`"[" space (%s ("," space %s)*)? "]" space`, itemRuleName, itemRuleName)
+		return sc.addRule(ruleName, rule), nil
+	} else {
+		primitiveRule, exists := PRIMITIVE_RULES[schemaType]
+		if !exists {
+			return "", fmt.Errorf("unrecognized schema: %v", schema)
+		}
+		if ruleName == "root" {
+			schemaType = "root"
+		}
+		return sc.addRule(schemaType, primitiveRule), nil
+	}
+}
+func (sc *JSONSchemaConverter) resolveReference(ref string, rootSchema map[string]interface{}) (map[string]interface{}, error) {
+	if !strings.HasPrefix(ref, "#/$defs/") {
+		return nil, fmt.Errorf("invalid reference format: %s", ref)
+	}
+
+	defKey := strings.TrimPrefix(ref, "#/$defs/")
+	definitions, exists := rootSchema["$defs"].(map[string]interface{})
+	if !exists {
+		return nil, fmt.Errorf("no definitions found in the schema: %s", rootSchema)
+	}
+
+	def, exists := definitions[defKey].(map[string]interface{})
+	if !exists {
+		return nil, fmt.Errorf("definition not found: %s %+v", defKey, definitions)
+	}
+
+	return def, nil
+}
+
+func (sc *JSONSchemaConverter) Grammar(schema map[string]interface{}, options ...func(*GrammarOption)) (string, error) {
+	sc.addRule("freestring", PRIMITIVE_RULES["freestring"])
+	_, err := sc.visit(schema, "", schema)
+	if err != nil {
+		return "", err
+	}
+	return sc.rules.ToGrammar(options...), nil
+}
+
+func (sc *JSONSchemaConverter) GrammarFromBytes(b []byte, options ...func(*GrammarOption)) (string, error) {
+	var schema map[string]interface{}
+	err := json.Unmarshal(b, &schema)
+	if err != nil {
+		return "", err
+	}
+	return sc.Grammar(schema, options...)
+}
--- a/pkg/functions/grammars/json_schema_test.go
+++ b/pkg/functions/grammars/json_schema_test.go
@@ -1,24 +1,14 @@
-package functions_test
+package grammars_test

 import (
 	"strings"

-	"github.com/mudler/LocalAI/pkg/functions"
 	. "github.com/mudler/LocalAI/pkg/functions"
+	. "github.com/mudler/LocalAI/pkg/functions/grammars"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 )

-func createFunction(field1 string, field2 string, name string, properties map[string]interface{}) map[string]interface{} {
-	property := map[string]interface{}{}
-	property[field1] = FunctionName{Const: name}
-	property[field2] = Argument{
-		Type:       "object",
-		Properties: properties,
-	}
-	return property
-}
-
 var testFunctions = []Item{
 	{
 		Type: "object",
@@ -245,7 +235,8 @@ root-1-name ::= "\"search\""`
 var _ = Describe("JSON schema grammar tests", func() {
 	Context("JSON", func() {
 		It("generates a valid grammar from JSON schema", func() {
-			grammar := NewJSONSchemaConverter("").GrammarFromBytes([]byte(testInput1))
+			grammar, err := NewJSONSchemaConverter("").GrammarFromBytes([]byte(testInput1))
+			Expect(err).To(BeNil())
 			results := strings.Split(inputResult1, "\n")
 			for _, r := range results {
 				if r != "" {
@@ -255,7 +246,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))))
 		})
 		It("generates a valid grammar from JSON schema", func() {
-			grammar := NewJSONSchemaConverter("").GrammarFromBytes([]byte(testInput2))
+			grammar, err := NewJSONSchemaConverter("").GrammarFromBytes([]byte(testInput2))
+			Expect(err).To(BeNil())
 			results := strings.Split(inputResult3, "\n")
 			for _, r := range results {
 				if r != "" {
@@ -269,7 +261,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctions}

-			grammar := structuredGrammar.Grammar()
+			grammar, err := structuredGrammar.Grammar()
+			Expect(err).To(BeNil())
 			results := strings.Split(inputResult1, "\n")
 			for _, r := range results {
 				if r != "" {
@@ -283,7 +276,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctions}

-			grammar := structuredGrammar.Grammar(functions.EnableMaybeArray)
+			grammar, err := structuredGrammar.Grammar(EnableMaybeArray)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					inputResult2,
@@ -301,7 +295,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(functions.EnableMaybeArray)
+			grammar, err := structuredGrammar.Grammar(EnableMaybeArray)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					inputResult4,
@@ -319,10 +314,11 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(
-				functions.SetPrefix("suffix"),
-				functions.EnableMaybeArray,
+			grammar, err := structuredGrammar.Grammar(
+				SetPrefix("suffix"),
+				EnableMaybeArray,
 			)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					rootResult(`"suffix" arr | realvalue`),
@@ -339,7 +335,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"))
+			grammar, err := structuredGrammar.Grammar(SetPrefix("suffix"))
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					rootResult(`"suffix" realvalue`),
@@ -356,7 +353,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"), functions.EnableMaybeString)
+			grammar, err := structuredGrammar.Grammar(SetPrefix("suffix"), EnableMaybeString)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					rootResult(`( "suffix" realvalue | mixedstring )`),
@@ -373,7 +371,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(functions.SetPrefix("suffix"), functions.EnableMaybeString, functions.EnableMaybeArray)
+			grammar, err := structuredGrammar.Grammar(SetPrefix("suffix"), EnableMaybeString, EnableMaybeArray)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					rootResult(`( "suffix" (arr | realvalue) | mixedstring )`),
@@ -392,7 +391,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray)
+			grammar, err := structuredGrammar.Grammar(EnableMaybeString, EnableMaybeArray)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					rootResult(`mixedstring | arr | realvalue`),
@@ -410,7 +410,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 			structuredGrammar := JSONFunctionStructure{
 				OneOf: testFunctionsName}

-			grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray, functions.NoMixedFreeString)
+			grammar, err := structuredGrammar.Grammar(EnableMaybeString, EnableMaybeArray, NoMixedFreeString)
+			Expect(err).To(BeNil())
 			results := strings.Split(
 				strings.Join([]string{
 					rootResult(`freestring | arr | realvalue`),
@@ -432,7 +433,8 @@ var _ = Describe("JSON schema grammar tests", func() {
 realvalue
 (","  realvalue)*
 )? "]"`
-			grammar := structuredGrammar.Grammar(functions.EnableMaybeString, functions.EnableMaybeArray, functions.DisableParallelNewLines)
+			grammar, err := structuredGrammar.Grammar(EnableMaybeString, EnableMaybeArray, DisableParallelNewLines)
+			Expect(err).To(BeNil())
 			results := strings.Split(content, "\n")
 			for _, r := range results {
 				if r != "" {
--- a/pkg/functions/grammars/llama31_schema.go
+++ b/pkg/functions/grammars/llama31_schema.go
@@ -0,0 +1,281 @@
+package grammars
+
+import (
+	"encoding/json"
+	"fmt"
+	"regexp"
+	"sort"
+	"strings"
+)
+
+type LLama31SchemaConverter struct {
+	fnName string
+	rules  Rules
+}
+
+func NewLLama31SchemaConverter(fnName string) *LLama31SchemaConverter {
+	rules := make(map[string]string)
+	rules["space"] = SPACE_RULE
+	if fnName == "" {
+		fnName = "name"
+	}
+
+	return &LLama31SchemaConverter{
+		rules:  rules,
+		fnName: fnName,
+	}
+}
+
+var GRAMMAR_LITERAL_ESCAPESLlama = map[string]string{
+	"\r": `\r`,
+	"\n": `\n`,
+}
+
+var GRAMMAR_LITERAL_ESCAPE_RELlama = regexp.MustCompile(`[\r\n]`)
+
+func (sc *LLama31SchemaConverter) formatLiteral(literal interface{}) (string, error) {
+	jLiteral, err := jsonString(literal)
+	if err != nil {
+		return "", err
+	}
+	escaped := GRAMMAR_LITERAL_ESCAPE_RELlama.ReplaceAllStringFunc(jLiteral, func(match string) string {
+		return GRAMMAR_LITERAL_ESCAPESLlama[match]
+	})
+	return escaped, nil
+}
+
+func (sc *LLama31SchemaConverter) formatLiteralQuoted(literal interface{}) (string, error) {
+	jLiteral, err := jsonString(literal)
+	if err != nil {
+		return "", err
+	}
+	escaped := GRAMMAR_LITERAL_ESCAPE_RE.ReplaceAllStringFunc(jLiteral, func(match string) string {
+		return GRAMMAR_LITERAL_ESCAPES[match]
+	})
+	return fmt.Sprintf(`"%s"`, escaped), nil
+}
+
+func (sc *LLama31SchemaConverter) addRule(name, rule string) string {
+	escName := INVALID_RULE_CHARS_RE.ReplaceAllString(name, "-")
+	key := escName
+	if existingRule, ok := sc.rules[escName]; ok && existingRule != rule {
+		i := 0
+		for {
+			key = fmt.Sprintf("%s%d", escName, i)
+			if _, ok := sc.rules[key]; !ok {
+				break
+			}
+			i++
+		}
+	}
+	sc.rules[key] = rule
+	return key
+}
+
+func (sc *LLama31SchemaConverter) visit(schema map[string]interface{}, name string, rootSchema map[string]interface{}) (string, error) {
+	st, existType := schema["type"]
+	var schemaType string
+	if existType {
+		schemaType = st.(string)
+	}
+	ruleName := name
+	if name == "" {
+		ruleName = "root"
+	}
+	_, oneOfExists := schema["oneOf"]
+	_, anyOfExists := schema["anyOf"]
+	if oneOfExists || anyOfExists {
+		var alternatives []string
+		oneOfSchemas, oneOfExists := schema["oneOf"].([]interface{})
+		anyOfSchemas, anyOfExists := schema["anyOf"].([]interface{})
+
+		if oneOfExists {
+			for i, altSchema := range oneOfSchemas {
+				alternative, err := sc.visit(altSchema.(map[string]interface{}), fmt.Sprintf("%s-%d", ruleName, i), rootSchema)
+				if err != nil {
+					return "", err
+				}
+				alternatives = append(alternatives, alternative)
+			}
+		} else if anyOfExists {
+			for i, altSchema := range anyOfSchemas {
+				alternative, err := sc.visit(altSchema.(map[string]interface{}), fmt.Sprintf("%s-%d", ruleName, i), rootSchema)
+				if err != nil {
+					return "", err
+				}
+				alternatives = append(alternatives, alternative)
+			}
+		}
+
+		rule := strings.Join(alternatives, " | ")
+		return sc.addRule(ruleName, rule), nil
+	} else if ref, exists := schema["$ref"].(string); exists {
+		referencedSchema, err := sc.resolveReference(ref, rootSchema)
+		if err != nil {
+			return "", err
+		}
+		return sc.visit(referencedSchema, name, rootSchema)
+	} else if constVal, exists := schema["const"]; exists {
+
+		literal, err := sc.formatLiteral((constVal))
+		if err != nil {
+			return "", err
+		}
+		return sc.addRule(ruleName, literal), nil
+	} else if enumVals, exists := schema["enum"].([]interface{}); exists {
+		var enumRules []string
+		for _, enumVal := range enumVals {
+			enumRule, err := sc.formatLiteralQuoted(enumVal)
+			if err != nil {
+				return "", err
+			}
+			enumRules = append(enumRules, enumRule)
+		}
+		rule := strings.Join(enumRules, " | ")
+		return sc.addRule(ruleName, rule), nil
+	} else if properties, exists := schema["properties"].(map[string]interface{}); schemaType == "object" && exists {
+		baseProperty := false
+		depth := strings.Split(name, "-")
+		if len(depth) == 2 {
+			baseProperty = true
+		}
+		type propData []struct {
+			propName   string
+			propSchema map[string]interface{}
+		}
+		var propPairs propData
+
+		for propName, propSchema := range properties {
+			propPairs = append(propPairs, struct {
+				propName   string
+				propSchema map[string]interface{}
+			}{propName: propName, propSchema: propSchema.(map[string]interface{})})
+		}
+
+		sort.Slice(propPairs, func(i, j int) bool {
+			return propPairs[i].propName < propPairs[j].propName
+		})
+
+		var rule strings.Builder
+		if baseProperty {
+			rule.WriteString(`"<function="`)
+		} else {
+			rule.WriteString(`"{" space`)
+		}
+
+		if baseProperty {
+
+			namePair := propData{}
+			for i, propPair := range propPairs {
+				propName := propPair.propName
+				if propName == sc.fnName {
+					namePair = append(namePair, propPair)
+					// remove namePair from propPairs
+					propPairs = append(propPairs[:i], propPairs[i+1:]...)
+					break
+				}
+			}
+			if len(namePair) == 0 {
+				return "", fmt.Errorf("no function name found in the schema: %s", schema)
+			}
+
+			propRuleName, err := sc.visit(namePair[0].propSchema, fmt.Sprintf("%s-%s", ruleName, sc.fnName), rootSchema)
+			if err != nil {
+				return "", err
+			}
+
+			rule.WriteString(fmt.Sprintf(` %s ">{" `, propRuleName))
+
+			for _, propPair := range propPairs {
+				propName := propPair.propName
+				propSchema := propPair.propSchema
+				propRuleName, err := sc.visit(propSchema, fmt.Sprintf("%s-%s", ruleName, propName), rootSchema)
+				if err != nil {
+					return "", err
+				}
+
+				rule.WriteString(propRuleName)
+			}
+
+			rule.WriteString(` "}</function>"`)
+
+		} else {
+			for i, propPair := range propPairs {
+				propName := propPair.propName
+				propSchema := propPair.propSchema
+				propRuleName, err := sc.visit(propSchema, fmt.Sprintf("%s-%s", ruleName, propName), rootSchema)
+				if err != nil {
+					return "", err
+				}
+				lPropName, err := sc.formatLiteralQuoted(propName)
+				if err != nil {
+					return "", err
+				}
+				if i > 0 {
+					rule.WriteString(` "," space`)
+				}
+
+				rule.WriteString(fmt.Sprintf(` %s space ":" space %s`, lPropName, propRuleName))
+			}
+
+		}
+
+		if !baseProperty {
+			rule.WriteString(` "}" space`)
+		}
+
+		return sc.addRule(ruleName, rule.String()), nil
+	} else if items, exists := schema["items"].(map[string]interface{}); schemaType == "array" && exists {
+		itemRuleName, err := sc.visit(items, fmt.Sprintf("%s-item", ruleName), rootSchema)
+		if err != nil {
+			return "", err
+		}
+		rule := fmt.Sprintf(`"[" space (%s ("," space %s)*)? "]" space`, itemRuleName, itemRuleName)
+		return sc.addRule(ruleName, rule), nil
+	} else {
+		primitiveRule, exists := PRIMITIVE_RULES[schemaType]
+		if !exists {
+			return "", fmt.Errorf("unrecognized schema: %v", schema)
+		}
+		if ruleName == "root" {
+			schemaType = "root"
+		}
+		return sc.addRule(schemaType, primitiveRule), nil
+	}
+}
+func (sc *LLama31SchemaConverter) resolveReference(ref string, rootSchema map[string]interface{}) (map[string]interface{}, error) {
+	if !strings.HasPrefix(ref, "#/$defs/") {
+		return nil, fmt.Errorf("invalid reference format: %s", ref)
+	}
+
+	defKey := strings.TrimPrefix(ref, "#/$defs/")
+	definitions, exists := rootSchema["$defs"].(map[string]interface{})
+	if !exists {
+		return nil, fmt.Errorf("no definitions found in the schema: %s", rootSchema)
+	}
+
+	def, exists := definitions[defKey].(map[string]interface{})
+	if !exists {
+		return nil, fmt.Errorf("definition not found: %s %+v", defKey, definitions)
+	}
+
+	return def, nil
+}
+
+func (sc *LLama31SchemaConverter) Grammar(schema map[string]interface{}, options ...func(*GrammarOption)) (string, error) {
+	sc.addRule("freestring", PRIMITIVE_RULES["freestring"])
+	_, err := sc.visit(schema, "", schema)
+	if err != nil {
+		return "", err
+	}
+	return sc.rules.ToGrammar(options...), nil
+}
+
+func (sc *LLama31SchemaConverter) GrammarFromBytes(b []byte, options ...func(*GrammarOption)) (string, error) {
+	var schema map[string]interface{}
+	err := json.Unmarshal(b, &schema)
+	if err != nil {
+		return "", err
+	}
+	return sc.Grammar(schema, options...)
+}
--- a/pkg/functions/grammars/llama31_schema_test.go
+++ b/pkg/functions/grammars/llama31_schema_test.go
@@ -0,0 +1,76 @@
+package grammars_test
+
+import (
+	"strings"
+
+	. "github.com/mudler/LocalAI/pkg/functions/grammars"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+const (
+	testllama31Input1 = `
+	{
+		"oneOf": [
+			{
+				"type": "object",
+				"properties": {
+					"function": {"const": "create_event"},
+					"arguments": {
+						"type": "object",
+						"properties": {
+							"title": {"type": "string"},
+							"date": {"type": "string"},
+							"time": {"type": "string"}
+						}
+					}
+				}
+			},
+			{
+				"type": "object",
+				"properties": {
+					"function": {"const": "search"},
+					"arguments": {
+						"type": "object",
+						"properties": {
+							"query": {"type": "string"}
+						}
+					}
+				}
+			}
+		]
+	}`
+	// <function=example_function_name>{{"example_name": "example_value"}}</function>
+	testllama31inputResult1 = `root-0-function ::= "create_event"
+freestring ::= (
+		[^"\\] |
+		"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
+  )* space
+root-0 ::= "<function=" root-0-function ">{" root-0-arguments "}</function>"
+root-1-arguments ::= "{" space "\"query\"" space ":" space string "}" space
+root ::= root-0 | root-1
+space ::= " "?
+root-0-arguments ::= "{" space "\"date\"" space ":" space string "," space "\"time\"" space ":" space string "," space "\"title\"" space ":" space string "}" space
+root-1 ::= "<function=" root-1-function ">{" root-1-arguments "}</function>"
+string ::= "\"" (
+	[^"\\] |
+	"\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F])
+)* "\"" space
+root-1-function ::= "search"`
+)
+
+var _ = Describe("JSON schema grammar tests", func() {
+	Context("JSON", func() {
+		It("generates a valid grammar from JSON schema", func() {
+			grammar, err := NewLLama31SchemaConverter("function").GrammarFromBytes([]byte(testllama31Input1))
+			Expect(err).ToNot(HaveOccurred())
+			results := strings.Split(testllama31inputResult1, "\n")
+			for _, r := range results {
+				if r != "" {
+					Expect(grammar).To(ContainSubstring(r))
+				}
+			}
+			Expect(len(results)).To(Equal(len(strings.Split(grammar, "\n"))))
+		})
+	})
+})
--- a/pkg/functions/grammars/options.go
+++ b/pkg/functions/grammars/options.go
@@ -1,4 +1,4 @@
-package functions
+package grammars

 type GrammarOption struct {
 	PropOrder               string
@@ -8,6 +8,9 @@ type GrammarOption struct {
 	MaybeString             bool
 	NoMixedFreeString       bool
 	ExpectStringsAfterJSON  bool
+
+	FunctionName string
+	SchemaType   SchemaConverterType
 }

 func (o *GrammarOption) Apply(options ...func(*GrammarOption)) {
@@ -48,3 +51,15 @@ func SetPropOrder(order string) func(*GrammarOption) {
 		o.PropOrder = order
 	}
 }
+
+func WithSchemaType(schemaType SchemaConverterType) func(*GrammarOption) {
+	return func(o *GrammarOption) {
+		o.SchemaType = schemaType
+	}
+}
+
+func WithFunctionName(name string) func(*GrammarOption) {
+	return func(o *GrammarOption) {
+		o.FunctionName = name
+	}
+}
--- a/pkg/functions/grammars/rules.go
+++ b/pkg/functions/grammars/rules.go
@@ -0,0 +1,93 @@
+package grammars
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/mudler/LocalAI/pkg/utils"
+)
+
+type Rules map[string]string
+
+func (rules Rules) ToGrammar(options ...func(*GrammarOption)) string {
+	grammarOpts := &GrammarOption{}
+	grammarOpts.Apply(options...)
+
+	prefix := grammarOpts.Prefix
+	maybeArray := grammarOpts.MaybeArray
+	disableParallelNewLines := grammarOpts.DisableParallelNewLines
+	maybeString := grammarOpts.MaybeString
+	noMixedFreeString := grammarOpts.NoMixedFreeString
+
+	var lines []string
+
+	swapRoot := maybeArray || maybeString || prefix != ""
+
+	// write down the computed rules.
+	// if maybeArray is true, we need to add the array rule and slightly tweak the root rule
+	for name, rule := range rules {
+		if swapRoot && name == "root" {
+			name = "realvalue"
+		}
+		lines = append(lines, fmt.Sprintf("%s ::= %s", name, rule))
+	}
+
+	if !swapRoot {
+		return strings.Join(lines, "\n")
+	}
+
+	newRoot := "realvalue"
+	if maybeArray {
+		newRoot = "arr | realvalue"
+	}
+
+	freestringRule := "mixedstring"
+	if noMixedFreeString {
+		freestringRule = "freestring"
+	}
+
+	if prefix != "" {
+		// quote newlines in suffix
+		prefix = utils.EscapeNewLines(prefix)
+
+		if maybeArray && maybeString {
+			newRoot = "(" + newRoot + ")"
+		}
+
+		if maybeString {
+			//newRoot = "( (\"" + suffix + "\" " + newRoot + ") | freestring ) "
+			newRoot = "( \"" + prefix + "\" " + newRoot + " | " + freestringRule + " ) "
+		} else {
+			newRoot = "\"" + prefix + "\" " + "" + newRoot + ""
+		}
+	} else if maybeString {
+		if maybeArray {
+			//	newRoot = "(" + newRoot + ")"
+		}
+
+		newRoot = freestringRule + " | " + newRoot
+	}
+
+	lines = append(lines, fmt.Sprintf("%s ::= %s", "root", newRoot))
+	if disableParallelNewLines {
+		lines = append(lines, array)
+	} else {
+		lines = append(lines, arrayNewLines)
+	}
+
+	if maybeArray {
+		if grammarOpts.ExpectStringsAfterJSON {
+			lines = append(lines, `mixedstring ::= freestring | freestring arr freestring | (freestring realvalue freestring)* | realvalue | arr`)
+		} else {
+			lines = append(lines, `mixedstring ::= freestring | freestring arr | freestring realvalue | realvalue | arr`)
+		}
+	} else {
+		if grammarOpts.ExpectStringsAfterJSON {
+			lines = append(lines, `mixedstring ::= freestring | (freestring realvalue freestring)* | realvalue`)
+		} else {
+			lines = append(lines, `mixedstring ::= freestring | freestring realvalue | realvalue`)
+		}
+	}
+
+	return strings.Join(lines, "\n")
+}
--- a/pkg/functions/grammars/types.go
+++ b/pkg/functions/grammars/types.go
@@ -0,0 +1,33 @@
+package grammars
+
+type SchemaConverterType int
+
+const (
+	JSONSchema SchemaConverterType = iota
+	LLama31Schema
+)
+
+const (
+	LlamaType string = "llama3.1"
+	JSONType  string = "json"
+)
+
+func (s SchemaConverterType) String() string {
+	switch s {
+	case JSONSchema:
+		return JSONType
+	case LLama31Schema:
+		return LlamaType
+	}
+	return "unknown"
+}
+
+func NewType(t string) SchemaConverterType {
+	switch t {
+	case JSONType:
+		return JSONSchema
+	case LlamaType:
+		return LLama31Schema
+	}
+	return JSONSchema
+}
--- a/pkg/functions/json_mode.go
+++ b/pkg/functions/json_mode.go
@@ -0,0 +1,28 @@
+package functions
+
+const (
+	JSONBNF = `root   ::= object
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\"" (
+    [^"\\] |
+    "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+ws ::= ([ \t\n] ws)?`
+)
--- a/pkg/functions/parse.go
+++ b/pkg/functions/parse.go
@@ -7,6 +7,7 @@ import (
 	"regexp"
 	"strings"

+	"github.com/mudler/LocalAI/pkg/functions/grammars"
 	"github.com/mudler/LocalAI/pkg/utils"
 	"github.com/rs/zerolog/log"
 )
@@ -22,7 +23,9 @@ type GrammarConfig struct {
 	MixedMode bool `yaml:"mixed_mode"`

 	// NoMixedFreeString disables the mixed mode for free strings
-	// In this way if the LLM selects a free string, it won't be mixed necessarly with JSON objects
+	// In this way if the LLM selects a free string, it won't be mixed necessarly with JSON objects.
+	// For example, if enabled the LLM or returns a JSON object or a free string, but not a mix of both
+	// If disabled(default): the LLM can return a JSON object surrounded by free strings (e.g. `this is the JSON result: { "bar": "baz" } for your question`). This forces the LLM to return at least a JSON object, but its not going to be strict
 	NoMixedFreeString bool `yaml:"no_mixed_free_string"`

 	// NoGrammar disables the grammar parsing and parses the responses directly from the LLM
@@ -39,6 +42,10 @@ type GrammarConfig struct {
 	// for instance name,arguments will make print { "name": "foo", "arguments": { "bar": "baz" } }
 	// instead of { "arguments": { "bar": "baz" }, "name": "foo" }
 	PropOrder string `yaml:"properties_order"`
+
+	// SchemaType can be configured to use a specific schema type to force the grammar
+	// available : json, llama3.1
+	SchemaType string `yaml:"schema_type"`
 }

 // FunctionsConfig is the configuration for the tool/function call.
@@ -92,28 +99,36 @@ type FuncCallResults struct {
 	Arguments string
 }

-func (g GrammarConfig) Options() []func(o *GrammarOption) {
-	opts := []func(o *GrammarOption){}
-	if g.MixedMode {
-		opts = append(opts, EnableMaybeString)
+func (g FunctionsConfig) GrammarOptions() []func(o *grammars.GrammarOption) {
+	opts := []func(o *grammars.GrammarOption){}
+	if g.GrammarConfig.MixedMode {
+		opts = append(opts, grammars.EnableMaybeString)
 	}
-	if g.ParallelCalls {
-		opts = append(opts, EnableMaybeArray)
+	if g.GrammarConfig.ParallelCalls {
+		opts = append(opts, grammars.EnableMaybeArray)
 	}
-	if g.DisableParallelNewLines {
-		opts = append(opts, DisableParallelNewLines)
+	if g.GrammarConfig.DisableParallelNewLines {
+		opts = append(opts, grammars.DisableParallelNewLines)
 	}
-	if g.Prefix != "" {
-		opts = append(opts, SetPrefix(g.Prefix))
+	if g.GrammarConfig.Prefix != "" {
+		opts = append(opts, grammars.SetPrefix(g.GrammarConfig.Prefix))
 	}
-	if g.NoMixedFreeString {
-		opts = append(opts, NoMixedFreeString)
+	if g.GrammarConfig.NoMixedFreeString {
+		opts = append(opts, grammars.NoMixedFreeString)
 	}
-	if g.ExpectStringsAfterJSON {
-		opts = append(opts, ExpectStringsAfterJSON)
+	if g.GrammarConfig.ExpectStringsAfterJSON {
+		opts = append(opts, grammars.ExpectStringsAfterJSON)
 	}

-	opts = append(opts, SetPropOrder(g.PropOrder))
+	if g.GrammarConfig.SchemaType != "" {
+		opts = append(opts, grammars.WithSchemaType(grammars.NewType(g.GrammarConfig.SchemaType)))
+	}
+
+	if g.FunctionNameKey != "" {
+		opts = append(opts, grammars.WithFunctionName(g.FunctionNameKey))
+	}
+
+	opts = append(opts, grammars.SetPropOrder(g.GrammarConfig.PropOrder))
 	return opts
 }

--- a/pkg/model/initializers.go
+++ b/pkg/model/initializers.go
@@ -212,7 +212,7 @@ func selectGRPCProcess(backend, assetDir string, f16 bool) string {
 					grpcProcess = p
 					foundCUDA = true
 				} else {
-					log.Info().Msgf("GPU device found but no CUDA backend present")
+					log.Debug().Msgf("Nvidia GPU device found, no embedded CUDA variant found. You can ignore this message if you are using container with CUDA support")
 				}
 			}
 			if strings.Contains(gpu.String(), "amd") {
@@ -222,7 +222,7 @@ func selectGRPCProcess(backend, assetDir string, f16 bool) string {
 					grpcProcess = p
 					foundAMDGPU = true
 				} else {
-					log.Info().Msgf("GPU device found but no HIPBLAS backend present")
+					log.Debug().Msgf("AMD GPU device found, no embedded HIPBLAS variant found. You can ignore this message if you are using container with HIPBLAS support")
 				}
 			}
 			if strings.Contains(gpu.String(), "intel") {
@@ -236,7 +236,7 @@ func selectGRPCProcess(backend, assetDir string, f16 bool) string {
 					grpcProcess = p
 					foundIntelGPU = true
 				} else {
-					log.Info().Msgf("GPU device found but no Intel backend present")
+					log.Debug().Msgf("Intel GPU device found, no embedded SYCL variant found. You can ignore this message if you are using container with SYCL support")
 				}
 			}
 		}
--- a/pkg/utils/strings.go
+++ b/pkg/utils/strings.go
@@ -18,3 +18,15 @@ func RandString(n int) string {
 	}
 	return string(b)
 }
+
+func Unique(arr []string) []string {
+	unique := make(map[string]bool)
+	var result []string
+	for _, item := range arr {
+		if _, ok := unique[item]; !ok {
+			unique[item] = true
+			result = append(result, item)
+		}
+	}
+	return result
+}