fix(ci): install latest git

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

.github/ci/modelslist.go

@@ -6,6 +6,7 @@ import (
 	"io/ioutil"
 	"os"
 
+	"github.com/microcosm-cc/bluemonday"
 	"gopkg.in/yaml.v3"
 )
 
@@ -279,6 +280,12 @@ func main() {
 		return
 	}
 
+	// Ensure that all arbitrary text content is sanitized before display
+	for i, m := range models {
+		models[i].Name = bluemonday.StrictPolicy().Sanitize(m.Name)
+		models[i].Description = bluemonday.StrictPolicy().Sanitize(m.Description)
+	}
+
 	// render the template
 	data := struct {
 		Models          []*GalleryModel

.github/dependabot.yml

@@ -9,6 +9,8 @@ updates:
     directory: "/"
     schedule:
       interval: "weekly"
+    ignore:
+    - dependency-name: "github.com/mudler/LocalAI/pkg/grpc/proto"
   - package-ecosystem: "github-actions"
     # Workflow files stored in the default location of `.github/workflows`. (You don't need to specify `/.github/workflows` for `directory`. You can use `directory: "/"`.)
     directory: "/"

.github/workflows/deploy-explorer.yaml

@@ -33,7 +33,7 @@ jobs:
         run: |
           CGO_ENABLED=0 make build-api
       - name: rm
-        uses: appleboy/ssh-action@v1.0.3
+        uses: appleboy/ssh-action@v1.1.0
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}
@@ -53,7 +53,7 @@ jobs:
             rm: true
             target: ./local-ai
       - name: restarting
-        uses: appleboy/ssh-action@v1.0.3
+        uses: appleboy/ssh-action@v1.1.0
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}

.github/workflows/notify-models.yaml

@@ -79,7 +79,7 @@ jobs:
         args: ${{ steps.summarize.outputs.message }}
     - name: Setup tmate session if fails
       if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3.18
+      uses: mxschmitt/action-tmate@v3.19
       with:
         detached: true
         connect-timeout-seconds: 180
@@ -161,7 +161,7 @@ jobs:
         TWITTER_ACCESS_TOKEN_SECRET: ${{ secrets.TWITTER_ACCESS_TOKEN_SECRET }}
     - name: Setup tmate session if fails
       if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3.18
+      uses: mxschmitt/action-tmate@v3.19
       with:
         detached: true
         connect-timeout-seconds: 180

.github/workflows/release.yaml

@@ -123,7 +123,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -232,7 +232,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -308,7 +308,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -350,7 +350,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180

.github/workflows/secscan.yaml

@@ -18,7 +18,7 @@ jobs:
         if: ${{ github.actor != 'dependabot[bot]' }}
       - name: Run Gosec Security Scanner
         if: ${{ github.actor != 'dependabot[bot]' }}
-        uses: securego/gosec@v2.21.0
+        uses: securego/gosec@v2.21.4
         with:
           # we let the report trigger content trigger a failure using the GitHub Security features.
           args: '-no-fail -fmt sarif -out results.sarif ./...'

.github/workflows/test-extra.yml

@@ -105,6 +105,14 @@ jobs:
   tests-parler-tts:
     runs-on: ubuntu-latest
     steps:
+      - name: Force Install GIT latest
+        run: |
+          sudo apt-get update \
+          && sudo apt-get install -y software-properties-common \
+          && sudo apt-get update \
+          && sudo add-apt-repository -y ppa:git-core/ppa \
+          && sudo apt-get update \
+          && sudo apt-get install -y git
       - name: Clone
         uses: actions/checkout@v4
         with:

.github/workflows/test.yml

@@ -133,7 +133,7 @@ jobs:
           PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" make --jobs 5 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -178,17 +178,26 @@ jobs:
         uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Dependencies
+        run: |
+          # Install protoc
+          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
+          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
+          rm protoc.zip
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          PATH="$PATH:$HOME/go/bin" make protogen-go
       - name: Build images
         run: |
           docker build --build-arg FFMPEG=true --build-arg IMAGE_TYPE=extras --build-arg EXTRA_BACKENDS=rerankers --build-arg MAKEFLAGS="--jobs=5 --output-sync=target" -t local-ai:tests -f Dockerfile .
           BASE_IMAGE=local-ai:tests DOCKER_AIO_IMAGE=local-ai-aio:test make docker-aio
       - name: Test
         run: |
-          LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
+            PATH="$PATH:$HOME/go/bin" LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
             make run-e2e-aio
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -226,7 +235,7 @@ jobs:
           BUILD_TYPE="GITHUB_CI_HAS_BROKEN_METAL" CMAKE_ARGS="-DGGML_F16C=OFF -DGGML_AVX512=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF" make --jobs 4 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180

CONTRIBUTING.md

@@ -15,8 +15,6 @@ Thank you for your interest in contributing to LocalAI! We appreciate your time
 - [Documentation](#documentation)
 - [Community and Communication](#community-and-communication)
 
-
-
 ## Getting Started
 
 ### Prerequisites
@@ -54,7 +52,7 @@ If you find a bug, have a feature request, or encounter any issues, please check
 
 ## Coding Guidelines
 
-- No specific coding guidelines at the moment. Please make sure the code can be tested. The most popular lint tools like []`golangci-lint`](https://golangci-lint.run) can help you here.
+- No specific coding guidelines at the moment. Please make sure the code can be tested. The most popular lint tools like [`golangci-lint`](https://golangci-lint.run) can help you here.
 
 ## Testing
 
@@ -84,5 +82,3 @@ We are welcome the contribution of the documents, please open new PR or create a
 - You can reach out via the Github issue tracker.
 - Open a new discussion at [Discussion](https://github.com/go-skynet/LocalAI/discussions)
 - Join the Discord channel [Discord](https://discord.gg/uJAeKSAGDy)
-
----

Dockerfile

@@ -9,6 +9,8 @@ FROM ${BASE_IMAGE} AS requirements-core
 USER root
 
 ARG GO_VERSION=1.22.6
+ARG CMAKE_VERSION=3.26.4
+ARG CMAKE_FROM_SOURCE=false
 ARG TARGETARCH
 ARG TARGETVARIANT
 
@@ -21,13 +23,25 @@ RUN apt-get update && \
         build-essential \
         ccache \
         ca-certificates \
-        cmake \
-        curl \
+        curl libssl-dev \
         git \
         unzip upx-ucl && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install CMake (the version in 22.04 is too old)
+RUN <<EOT bash
+    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
+    else
+        apt-get update && \
+        apt-get install -y \
+            cmake && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/*
+    fi
+EOT
+
 # Install Go
 RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | tar -C /usr/local -xz
 ENV PATH=$PATH:/root/go/bin:/usr/local/go/bin
@@ -188,6 +202,8 @@ FROM ${GRPC_BASE_IMAGE} AS grpc
 # This is a bit of a hack, but it's required in order to be able to effectively cache this layer in CI
 ARG GRPC_MAKEFLAGS="-j4 -Otarget"
 ARG GRPC_VERSION=v1.65.0
+ARG CMAKE_FROM_SOURCE=false
+ARG CMAKE_VERSION=3.26.4
 
 ENV MAKEFLAGS=${GRPC_MAKEFLAGS}
 
@@ -196,12 +212,24 @@ WORKDIR /build
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         ca-certificates \
-        build-essential \
-        cmake \
+        build-essential curl libssl-dev \
         git && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install CMake (the version in 22.04 is too old)
+RUN <<EOT bash
+    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
+    else
+        apt-get update && \
+        apt-get install -y \
+            cmake && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/*
+    fi
+EOT
+
 # We install GRPC to a different prefix here so that we can copy in only the build artifacts later
 # saves several hundred MB on the final docker image size vs copying in the entire GRPC source tree
 # and running make install in the target container

Makefile

@@ -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?=f0c7b5edf82aa200656fd88c11ae3a805d7130bf
+CPPLLAMA_VERSION?=0a1c750c80147687df267114c81956757cc14382
 
 # go-rwkv version
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
@@ -16,7 +16,7 @@ RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=34972dbe221709323714fc8402f2e24041d48213
+WHISPER_CPP_VERSION?=0fbaac9c891055796456df7b9122a70c220f9ca1
 
 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -468,15 +468,15 @@ run-e2e-image:
 	ls -liah $(abspath ./tests/e2e-fixtures)
 	docker run -p 5390:8080 -e MODELS_PATH=/models -e THREADS=1 -e DEBUG=true -d --rm -v $(TEST_DIR):/models --gpus all --name e2e-tests-$(RANDOM) localai-tests
 
-run-e2e-aio:
+run-e2e-aio: protogen-go
 	@echo 'Running e2e AIO tests'
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e-aio
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e-aio
 
 test-e2e:
 	@echo 'Running e2e tests'
 	BUILD_TYPE=$(BUILD_TYPE) \
 	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390/v1 \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e
 
 teardown-e2e:
 	rm -rf $(TEST_DIR) || true
@@ -484,24 +484,24 @@ teardown-e2e:
 
 test-llama: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-tts: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-stablediffusion: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-stores: backend-assets/grpc/local-store
 	mkdir -p tests/integration/backend-assets/grpc
 	cp -f backend-assets/grpc/local-store tests/integration/backend-assets/grpc/
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts 1 -v -r tests/integration
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts $(TEST_FLAKES) -v -r tests/integration
 
 test-container:
 	docker build --target requirements -t local-ai-test-container .

README.md

@@ -66,11 +66,24 @@ docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 # docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12 
 ```
 
+To load models:
+
+```bash
+# From the model gallery (see available models with `local-ai models list`, in the WebUI from the model tab, or visiting https://models.localai.io)
+local-ai run llama-3.2-1b-instruct:q4_k_m
+# Start LocalAI with the phi-2 model directly from huggingface
+local-ai run huggingface://TheBloke/phi-2-GGUF/phi-2.Q8_0.gguf
+# Install and run a model from the Ollama OCI registry
+local-ai run ollama://gemma:2b
+# Run a model from a configuration file
+local-ai run https://gist.githubusercontent.com/.../phi-2.yaml
+# Install and run a model from a standard OCI registry (e.g., Docker Hub)
+local-ai run oci://localai/phi-2:latest
+```
+
 [💻 Getting started](https://localai.io/basics/getting_started/index.html)
 
-## 🔥🔥 Hot topics / Roadmap
-
-[Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
+## 📰 Latest project news
 
 - Aug 2024:  🆕 FLUX-1, [P2P Explorer](https://explorer.localai.io)
 - July 2024: 🔥🔥 🆕 P2P Dashboard, LocalAI Federated mode and AI Swarms: https://github.com/mudler/LocalAI/pull/2723
@@ -83,8 +96,12 @@ docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 - May 2024: Chat, TTS, and Image generation in the WebUI: https://github.com/mudler/LocalAI/pull/2222
 - April 2024: Reranker API: https://github.com/mudler/LocalAI/pull/2121
 
-Hot topics (looking for contributors):
+Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
+## 🔥🔥 Hot topics (looking for help):
+
+- Multimodal with vLLM and Video understanding: https://github.com/mudler/LocalAI/pull/3729
+- Realtime API https://github.com/mudler/LocalAI/issues/3714
 - 🔥🔥 Distributed, P2P Global community pools: https://github.com/mudler/LocalAI/issues/3113
 - WebUI improvements: https://github.com/mudler/LocalAI/issues/2156
 - Backends v2: https://github.com/mudler/LocalAI/issues/1126

backend/backend.proto

@@ -26,6 +26,19 @@ service Backend {
   rpc StoresFind(StoresFindOptions) returns (StoresFindResult) {}
 
   rpc Rerank(RerankRequest) returns (RerankResult) {}
+
+  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
+}
+
+// Define the empty request
+message MetricsRequest {}
+
+message MetricsResponse {
+  int32 slot_id = 1;
+  string prompt_json_for_slot = 2;  // Stores the prompt as a JSON string.
+  float tokens_per_second = 3;
+  int32 tokens_generated = 4;
+  int32 prompt_tokens_processed = 5;
 }
 
 message RerankRequest {
@@ -136,6 +149,7 @@ message PredictOptions {
   repeated Message Messages = 44;
   repeated string Videos = 45;
   repeated string Audios = 46;
+  string CorrelationId = 47;
 }
 
 // The response message containing the result
@@ -205,6 +219,7 @@ message ModelOptions {
   int32  SwapSpace = 53;
   int32  MaxModelLen = 54;
   int32  TensorParallelSize = 55;
+  string LoadFormat = 58;
 
   string MMProj = 41;
 

backend/cpp/llama/grpc-server.cpp

@@ -113,7 +113,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
     std::string ret;
     for (; begin != end; ++begin)
     {
-        ret += llama_token_to_piece(ctx, *begin);
+        ret += common_token_to_piece(ctx, *begin);
     }
     return ret;
 }
@@ -121,7 +121,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
 // format incomplete utf-8 multibyte character for output
 static std::string tokens_to_output_formatted_string(const llama_context *ctx, const llama_token token)
 {
-    std::string out = token == -1 ? "" : llama_token_to_piece(ctx, token);
+    std::string out = token == -1 ? "" : common_token_to_piece(ctx, token);
     // if the size is 1 and first bit is 1, meaning it's a partial character
     //   (size > 1 meaning it's already a known token)
     if (out.size() == 1 && (out[0] & 0x80) == 0x80)
@@ -203,8 +203,8 @@ struct llama_client_slot
     std::string stopping_word;
 
     // sampling
-    struct gpt_sampler_params sparams;
-    gpt_sampler *ctx_sampling = nullptr;
+    struct common_sampler_params sparams;
+    common_sampler *ctx_sampling = nullptr;
 
     int32_t ga_i = 0;   // group-attention state
     int32_t ga_n = 1;   // group-attention factor
@@ -257,7 +257,7 @@ struct llama_client_slot
         images.clear();
     }
 
-    bool has_budget(gpt_params &global_params) {
+    bool has_budget(common_params &global_params) {
         if (params.n_predict == -1 && global_params.n_predict == -1)
         {
             return true; // limitless
@@ -391,6 +391,39 @@ struct llama_metrics {
     }
 };
 
+struct llava_embd_batch {
+    std::vector<llama_pos>      pos;
+    std::vector<int32_t>        n_seq_id;
+    std::vector<llama_seq_id>   seq_id_0;
+    std::vector<llama_seq_id *> seq_ids;
+    std::vector<int8_t>         logits;
+    llama_batch batch;
+    llava_embd_batch(float * embd, int32_t n_tokens, llama_pos pos_0, llama_seq_id seq_id) {
+        pos     .resize(n_tokens);
+        n_seq_id.resize(n_tokens);
+        seq_ids .resize(n_tokens + 1);
+        logits  .resize(n_tokens);
+        seq_id_0.resize(1);
+        seq_id_0[0] = seq_id;
+        seq_ids [n_tokens] = nullptr;
+        batch = {
+            /*n_tokens       =*/ n_tokens,
+            /*tokens         =*/ nullptr,
+            /*embd           =*/ embd,
+            /*pos            =*/ pos.data(),
+            /*n_seq_id       =*/ n_seq_id.data(),
+            /*seq_id         =*/ seq_ids.data(),
+            /*logits         =*/ logits.data(),
+        };
+        for (int i = 0; i < n_tokens; i++) {
+            batch.pos     [i] = pos_0 + i;
+            batch.n_seq_id[i] = 1;
+            batch.seq_id  [i] = seq_id_0.data();
+            batch.logits  [i] = false;
+        }
+    }
+};
+
 struct llama_server_context
 {
     llama_model *model = nullptr;
@@ -398,7 +431,7 @@ struct llama_server_context
 
     clip_ctx *clp_ctx = nullptr;
 
-    gpt_params params;
+    common_params params;
 
     llama_batch batch;
 
@@ -441,7 +474,7 @@ struct llama_server_context
         }
     }
 
-    bool load_model(const gpt_params &params_)
+    bool load_model(const common_params &params_)
     {
         params = params_;
         if (!params.mmproj.empty()) {
@@ -458,9 +491,9 @@ struct llama_server_context
             }
         }
 
-        llama_init_result llama_init = llama_init_from_gpt_params(params);
-        model = llama_init.model;
-        ctx = llama_init.context;
+        common_init_result common_init = common_init_from_params(params);
+        model = common_init.model;
+        ctx = common_init.context;
         if (model == nullptr)
         {
             LOG_ERR("unable to load model: %s", params.model.c_str());
@@ -495,6 +528,16 @@ struct llama_server_context
         }
     }
 
+    llama_client_slot* get_active_slot() {
+        for (llama_client_slot& slot : slots) {
+            // Check if the slot is currently processing
+            if (slot.is_processing()) {
+                return &slot;  // Return the active slot
+            }
+        }
+        return nullptr;  // No active slot found
+    }
+
     void initialize() {
         // create slots
         all_slots_are_idle = true;
@@ -568,12 +611,12 @@ struct llama_server_context
                     std::vector<llama_token> p;
                     if (first)
                     {
-                        p = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+                        p = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
                         first = false;
                     }
                     else
                     {
-                        p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
+                        p = common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
                     }
                     prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
                 }
@@ -590,7 +633,7 @@ struct llama_server_context
         else
         {
             auto s = json_prompt.template get<std::string>();
-            prompt_tokens = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+            prompt_tokens = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
         }
 
         return prompt_tokens;
@@ -619,7 +662,7 @@ struct llama_server_context
 
     bool launch_slot_with_data(llama_client_slot* &slot, json data) {
         slot_params default_params;
-        gpt_sampler_params default_sparams;
+        common_sampler_params default_sparams;
  
         slot->params.stream             = json_value(data, "stream",            false);
         slot->params.cache_prompt       = json_value(data, "cache_prompt",      false);
@@ -759,7 +802,7 @@ struct llama_server_context
                     }
                     else if (el[0].is_string())
                     {
-                        auto toks = llama_tokenize(model, el[0].get<std::string>(), false);
+                        auto toks = common_tokenize(model, el[0].get<std::string>(), false);
                         for (auto tok : toks)
                         {
                             slot->sparams.logit_bias.push_back({tok, bias});
@@ -791,7 +834,7 @@ struct llama_server_context
                         sampler_names.emplace_back(name);
                     }
                 }
-                slot->sparams.samplers = gpt_sampler_types_from_names(sampler_names, false);
+                slot->sparams.samplers = common_sampler_types_from_names(sampler_names, false);
         }
         else
         {
@@ -875,9 +918,9 @@ struct llama_server_context
 
         if (slot->ctx_sampling != nullptr)
         {
-            gpt_sampler_free(slot->ctx_sampling);
+            common_sampler_free(slot->ctx_sampling);
         }
-        slot->ctx_sampling = gpt_sampler_init(model, slot->sparams);
+        slot->ctx_sampling = common_sampler_init(model, slot->sparams);
         //llama_set_rng_seed(ctx, slot->params.seed);
         slot->command = LOAD_PROMPT;
 
@@ -904,13 +947,13 @@ struct llama_server_context
         system_tokens.clear();
 
         if (!system_prompt.empty()) {
-            system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);
+            system_tokens = common_tokenize(ctx, system_prompt, add_bos_token);
 
-            llama_batch_clear(batch);
+            common_batch_clear(batch);
 
             for (int i = 0; i < (int)system_tokens.size(); ++i)
             {
-                llama_batch_add(batch, system_tokens[i], i, { 0 }, false);
+                common_batch_add(batch, system_tokens[i], i, { 0 }, false);
             }
 
             for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += params.n_batch)
@@ -924,7 +967,6 @@ struct llama_server_context
                     batch.n_seq_id + i,
                     batch.seq_id   + i,
                     batch.logits   + i,
-                    0, 0, 0, // unused
                 };
                 if (llama_decode(ctx, batch_view) != 0)
                 {
@@ -999,7 +1041,7 @@ struct llama_server_context
 
     bool process_token(completion_token_output &result, llama_client_slot &slot) {
         // remember which tokens were sampled - used for repetition penalties during sampling
-        const std::string token_str = llama_token_to_piece(ctx, result.tok);
+        const std::string token_str = common_token_to_piece(ctx, result.tok);
         slot.sampled = result.tok;
 
         // search stop word and delete it
@@ -1150,7 +1192,7 @@ struct llama_server_context
         samplers.reserve(slot.sparams.samplers.size());
         for (const auto & sampler : slot.sparams.samplers)
         {
-            samplers.emplace_back(gpt_sampler_type_to_str(sampler));
+            samplers.emplace_back(common_sampler_type_to_str(sampler));
         }
 
         return json {
@@ -1206,7 +1248,7 @@ struct llama_server_context
         if (slot.sparams.n_probs > 0)
         {
             std::vector<completion_token_output> probs_output = {};
-            const std::vector<llama_token> to_send_toks = llama_tokenize(ctx, tkn.text_to_send, false);
+            const std::vector<llama_token> to_send_toks = common_tokenize(ctx, tkn.text_to_send, false);
             size_t probs_pos      = std::min(slot.sent_token_probs_index,                       slot.generated_token_probs.size());
             size_t probs_stop_pos = std::min(slot.sent_token_probs_index + to_send_toks.size(), slot.generated_token_probs.size());
             if (probs_pos < probs_stop_pos)
@@ -1258,7 +1300,7 @@ struct llama_server_context
             std::vector<completion_token_output> probs = {};
             if (!slot.params.stream && slot.stopped_word)
             {
-                const std::vector<llama_token> stop_word_toks = llama_tokenize(ctx, slot.stopping_word, false);
+                const std::vector<llama_token> stop_word_toks = common_tokenize(ctx, slot.stopping_word, false);
                 probs = std::vector<completion_token_output>(slot.generated_token_probs.begin(), slot.generated_token_probs.end() - stop_word_toks.size());
             }
             else
@@ -1369,7 +1411,6 @@ struct llama_server_context
                     batch.n_seq_id + i,
                     batch.seq_id   + i,
                     batch.logits   + i,
-                    0, 0, 0, // unused
                 };
                 if (llama_decode(ctx, batch_view))
                 {
@@ -1388,8 +1429,9 @@ struct llama_server_context
                 }
 
                 const int n_embd = llama_n_embd(model);
-                llama_batch batch_img = { n_eval, nullptr, (img.image_embedding + i * n_embd), nullptr, nullptr, nullptr, nullptr, slot.n_past, 1, 0, };
-                if (llama_decode(ctx, batch_img))
+                float * embd = img.image_embedding + i * n_embd;
+                llava_embd_batch llava_batch = llava_embd_batch(embd, n_eval, slot.n_past, 0);
+                if (llama_decode(ctx, llava_batch.batch))
                 {
                     LOG("%s : failed to eval image\n", __func__);
                     return false;
@@ -1398,7 +1440,7 @@ struct llama_server_context
             }
             image_idx++;
 
-            llama_batch_clear(batch);
+            common_batch_clear(batch);
 
             // append prefix of next image
             const auto json_prompt = (image_idx >= (int) slot.images.size()) ?
@@ -1408,7 +1450,7 @@ struct llama_server_context
             std::vector<llama_token> append_tokens = tokenize(json_prompt, false); // has next image
             for (int i = 0; i < (int) append_tokens.size(); ++i)
             {
-                llama_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
+                common_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
                 slot.n_past += 1;
             }
         }
@@ -1540,7 +1582,7 @@ struct llama_server_context
             update_system_prompt();
         }
 
-        llama_batch_clear(batch);
+        common_batch_clear(batch);
 
         if (all_slots_are_idle)
         {
@@ -1618,7 +1660,7 @@ struct llama_server_context
 
             // TODO: we always have to take into account the "system_tokens"
             //       this is not great and needs to be improved somehow
-            llama_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
+            common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
             slot.n_past += 1;
         }
 
@@ -1712,7 +1754,7 @@ struct llama_server_context
 
                     if (!slot.params.cache_prompt)
                     {
-                        gpt_sampler_reset(slot.ctx_sampling);
+                        common_sampler_reset(slot.ctx_sampling);
 
                         slot.n_past = 0;
                         slot.n_past_se = 0;
@@ -1724,7 +1766,7 @@ struct llama_server_context
                         // push the prompt into the sampling context (do not apply grammar)
                         for (auto &token : prompt_tokens)
                         {
-                            gpt_sampler_accept(slot.ctx_sampling, token, false);
+                            common_sampler_accept(slot.ctx_sampling, token, false);
                         }
 
                         slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
@@ -1816,7 +1858,7 @@ struct llama_server_context
                                 ga_i += ga_w/ga_n;
                             }
                         }
-                        llama_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
+                        common_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
                         slot_npast++;
                     }
 
@@ -1894,7 +1936,6 @@ struct llama_server_context
                 batch.n_seq_id + i,
                 batch.seq_id   + i,
                 batch.logits   + i,
-                0, 0, 0, // unused
             };
 
             const int ret = llama_decode(ctx, batch_view);
@@ -1933,9 +1974,9 @@ struct llama_server_context
                 }
 
                 completion_token_output result;
-                const llama_token id = gpt_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
+                const llama_token id = common_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
 
-                gpt_sampler_accept(slot.ctx_sampling, id, true);
+                common_sampler_accept(slot.ctx_sampling, id, true);
 
                 slot.n_decoded += 1;
                 if (slot.n_decoded == 1)
@@ -1946,7 +1987,7 @@ struct llama_server_context
                 }
 
                 result.tok = id;
-                const auto * cur_p = gpt_sampler_get_candidates(slot.ctx_sampling);
+                const auto * cur_p = common_sampler_get_candidates(slot.ctx_sampling);
 
                 for (size_t i = 0; i < (size_t) slot.sparams.n_probs; ++i) {
                     result.probs.push_back({
@@ -1999,7 +2040,7 @@ static json format_partial_response(
 struct token_translator
 {
     llama_context * ctx;
-    std::string operator()(llama_token tok)                    const { return llama_token_to_piece(ctx, tok); }
+    std::string operator()(llama_token tok)                    const { return common_token_to_piece(ctx, tok); }
     std::string operator()(const completion_token_output &cto) const { return (*this)(cto.tok); }
 };
 
@@ -2106,6 +2147,9 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     data["ignore_eos"] = predict->ignoreeos();
     data["embeddings"] = predict->embeddings();
 
+    // Add the correlationid to json data
+    data["correlation_id"] = predict->correlationid();
+
     // for each image in the request, add the image data
     //
     for (int i = 0; i < predict->images_size(); i++) {
@@ -2190,7 +2234,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 // }
 
 static void params_parse(const backend::ModelOptions* request,
-                                gpt_params & params) {
+                                common_params & params) {
    
     // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
 
@@ -2298,7 +2342,7 @@ public:
 
   grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
     // Implement LoadModel RPC
-    gpt_params params;
+    common_params params;
     params_parse(request, params);
 
     llama_backend_init();
@@ -2344,6 +2388,11 @@ public:
                 int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
                 reply.set_prompt_tokens(tokens_evaluated);
 
+                // Log Request Correlation Id
+                LOG_VERBOSE("correlation:", {
+                    { "id", data["correlation_id"] }
+                });
+
                 // Send the reply
                 writer->Write(reply);
 
@@ -2367,6 +2416,12 @@ public:
         std::string completion_text;
         task_result result = llama.queue_results.recv(task_id);
         if (!result.error && result.stop) {
+            
+            // Log Request Correlation Id
+            LOG_VERBOSE("correlation:", {
+                { "id", data["correlation_id"] }
+            });
+
             completion_text = result.result_json.value("content", "");
             int32_t tokens_predicted = result.result_json.value("tokens_predicted", 0);
             int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
@@ -2406,6 +2461,31 @@ public:
 
         return grpc::Status::OK;
     }
+
+    grpc::Status GetMetrics(ServerContext* context, const backend::MetricsRequest* request, backend::MetricsResponse* response) {
+        llama_client_slot* active_slot = llama.get_active_slot();
+
+        if (active_slot != nullptr) {
+            // Calculate the tokens per second using existing logic
+            double tokens_per_second = 1e3 / active_slot->t_token_generation * active_slot->n_decoded;
+
+            // Populate the response with metrics
+            response->set_slot_id(active_slot->id);
+            response->set_prompt_json_for_slot(active_slot->prompt.dump());
+            response->set_tokens_per_second(tokens_per_second);
+            response->set_tokens_generated(active_slot->n_decoded);
+            response->set_prompt_tokens_processed(active_slot->num_prompt_tokens_processed);
+        } else {
+            // Handle case when no active slot exists
+            response->set_slot_id(0);
+            response->set_prompt_json_for_slot("");
+            response->set_tokens_per_second(0);
+            response->set_tokens_generated(0);
+            response->set_prompt_tokens_processed(0);
+        }
+
+        return grpc::Status::OK;
+    } 
 };
 
 void RunServer(const std::string& server_address) {

backend/python/autogptq/requirements-cublas11.txt

@@ -1,2 +1,2 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118

backend/python/autogptq/requirements-cublas12.txt

@@ -1 +1 @@
-torch
+torch==2.4.1

backend/python/autogptq/requirements-hipblas.txt

@@ -1,2 +1,2 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
+torch==2.4.1+rocm6.0

backend/python/autogptq/requirements.txt

@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi
 transformers

backend/python/bark/requirements-cpu.txt

@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1

backend/python/bark/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchaudio
+torch==2.4.1+cu118
+torchaudio==2.4.1+cu118
 transformers
 accelerate

backend/python/bark/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1
 transformers
 accelerate

backend/python/bark/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
-torchaudio
+torch==2.4.1+rocm6.0
+torchaudio==2.4.1+rocm6.0
 transformers
 accelerate

backend/python/bark/requirements.txt

@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi

backend/python/common/template/requirements.txt

@@ -1,2 +1,2 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf

backend/python/coqui/requirements-cpu.txt

@@ -1,3 +1,4 @@
 transformers
 accelerate
-torch
+torch==2.4.1
+coqui-tts

backend/python/coqui/requirements-cublas11.txt

@@ -1,5 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchaudio
+torch==2.4.1+cu118
+torchaudio==2.4.1+cu118
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements-cublas12.txt

@@ -1,4 +1,5 @@
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements-hipblas.txt

@@ -1,5 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
-torchaudio
+torch==2.4.1+rocm6.0
+torchaudio==2.4.1+rocm6.0
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements-intel.txt

@@ -5,4 +5,5 @@ torchaudio
 optimum[openvino]
 setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements.txt

@@ -1,4 +1,4 @@
-coqui-tts
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
-certifi
+certifi
+packaging==24.1

backend/python/coqui/test.py

@@ -19,7 +19,7 @@ class TestBackendServicer(unittest.TestCase):
         This method sets up the gRPC service by starting the server
         """
         self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(10)
+        time.sleep(30)
 
     def tearDown(self) -> None:
         """

backend/python/diffusers/requirements-cpu.txt

@@ -5,5 +5,5 @@ accelerate
 compel
 peft
 sentencepiece
-torch
+torch==2.4.1
 optimum-quanto

backend/python/diffusers/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
 diffusers
 opencv-python
 transformers

backend/python/diffusers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
+torch==2.4.1
 diffusers
 opencv-python
 transformers

backend/python/diffusers/requirements.txt

@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.66.1
+grpcio==1.67.0
 pillow
 protobuf
 certifi

backend/python/exllama2/requirements-cpu.txt

@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch
+torch==2.4.1

backend/python/exllama2/requirements-cublas11.txt

@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
 transformers
 accelerate

backend/python/exllama2/requirements-cublas12.txt

@@ -1,3 +1,3 @@
-torch
+torch==2.4.1
 transformers
 accelerate

backend/python/exllama2/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi
 wheel

backend/python/mamba/requirements-cpu.txt

@@ -1,2 +1,2 @@
-torch
+torch==2.4.1
 transformers

backend/python/mamba/requirements-cublas11.txt

@@ -1,3 +1,3 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
 transformers

backend/python/mamba/requirements-cublas12.txt

@@ -1,2 +1,2 @@
-torch
+torch==2.4.1
 transformers

backend/python/mamba/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi

backend/python/openvoice/requirements-cpu.txt

@@ -1 +1,3 @@
-torch
+torch==2.4.1
+git+https://github.com/myshell-ai/MeloTTS.git
+git+https://github.com/myshell-ai/OpenVoice.git

backend/python/openvoice/requirements-cublas11.txt

@@ -1,2 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
+git+https://github.com/myshell-ai/MeloTTS.git
+git+https://github.com/myshell-ai/OpenVoice.git

backend/python/openvoice/requirements-cublas12.txt

@@ -1 +1,3 @@
-torch
+torch==2.4.1
+git+https://github.com/myshell-ai/MeloTTS.git
+git+https://github.com/myshell-ai/OpenVoice.git

backend/python/openvoice/requirements-hipblas.txt

@@ -1,2 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
+torch==2.4.1+rocm6.0
+git+https://github.com/myshell-ai/MeloTTS.git
+git+https://github.com/myshell-ai/OpenVoice.git

backend/python/openvoice/requirements-intel.txt

@@ -2,22 +2,22 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 librosa==0.9.1
-faster-whisper==1.0.3
+faster-whisper==0.9.0
 pydub==0.25.1
 wavmark==0.0.3
-numpy==1.26.4
+numpy==1.22.0
 eng_to_ipa==0.0.2
 inflect==7.0.0
 unidecode==1.3.7
-whisper-timestamped==1.15.4
+whisper-timestamped==1.14.2
 openai
 python-dotenv
 pypinyin==0.50.0
 cn2an==0.5.22
 jieba==0.42.1
-gradio==4.38.1
 langid==1.1.6
 git+https://github.com/myshell-ai/MeloTTS.git
+git+https://github.com/myshell-ai/OpenVoice.git

backend/python/openvoice/requirements.txt

@@ -1,10 +1,10 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 librosa
 faster-whisper
 pydub==0.25.1
 wavmark==0.0.3
-numpy
+numpy==1.22.0
 eng_to_ipa==0.0.2
 inflect
 unidecode
@@ -13,8 +13,8 @@ openai
 python-dotenv
 pypinyin
 cn2an==0.5.22
+networkx==2.8.8
 jieba==0.42.1
-gradio
+gradio==3.48.0
 langid==1.1.6
-git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
+llvmlite==0.43.0

backend/python/openvoice/test.py

@@ -19,7 +19,7 @@ class TestBackendServicer(unittest.TestCase):
         This method sets up the gRPC service by starting the server
         """
         self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(10)
+        time.sleep(30)
 
     def tearDown(self) -> None:
         """

backend/python/parler-tts/install.sh

@@ -15,12 +15,12 @@ installRequirements
 
 # https://github.com/descriptinc/audiotools/issues/101
 # incompatible protobuf versions.
-PYDIR=python3.10
-pyenv="${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/"
+# PYDIR=python3.10
+# pyenv="${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/"
 
-if [ ! -d ${pyenv} ]; then
-    echo "(parler-tts/install.sh): Error: ${pyenv} does not exist"
-    exit 1
-fi
+# if [ ! -d ${pyenv} ]; then
+#     echo "(parler-tts/install.sh): Error: ${pyenv} does not exist"
+#     exit 1
+# fi
 
-curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${pyenv}/builder.py
+# curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${pyenv}/builder.py

backend/python/parler-tts/requirements-after.txt

@@ -1,3 +1,4 @@
 git+https://github.com/huggingface/parler-tts.git@8e465f1b5fcd223478e07175cb40494d19ffbe17
 llvmlite==0.43.0
 numba==0.60.0
+git+https://github.com/descriptinc/audiotools

backend/python/parler-tts/requirements-cpu.txt

@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch
+torch==2.4.1

backend/python/parler-tts/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchaudio
+torch==2.4.1+cu118
+torchaudio==2.4.1+cu118
 transformers
 accelerate

backend/python/parler-tts/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1
 transformers
 accelerate

backend/python/parler-tts/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi
 llvmlite==0.43.0

backend/python/rerankers/requirements-cpu.txt

@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch
+torch==2.4.1
 rerankers[transformers]

backend/python/rerankers/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 transformers
 accelerate
-torch
+torch==2.4.1+cu118
 rerankers[transformers]

backend/python/rerankers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch
+torch==2.4.1
 rerankers[transformers]

backend/python/rerankers/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 transformers
 accelerate
-torch
+torch==2.4.1+rocm6.0
 rerankers[transformers]

backend/python/rerankers/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi

backend/python/sentencetransformers/requirements-cpu.txt

@@ -1,6 +1,6 @@
-torch
+torch==2.4.1
 accelerate
 transformers
 bitsandbytes
-sentence-transformers==3.1.0
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
 accelerate
-sentence-transformers==3.1.0
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
+torch==2.4.1
 accelerate
-sentence-transformers==3.1.0
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
+torch==2.4.1+rocm6.0
 accelerate
-sentence-transformers==3.1.0
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements-intel.txt

@@ -4,5 +4,5 @@ torch
 optimum[openvino]
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
 accelerate
-sentence-transformers==3.1.0
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi
 datasets

backend/python/transformers-musicgen/requirements-cpu.txt

@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch
+torch==2.4.1

backend/python/transformers-musicgen/requirements-cublas11.txt

@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 transformers
 accelerate
-torch
+torch==2.4.1+cu118

backend/python/transformers-musicgen/requirements-cublas12.txt

@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch
+torch==2.4.1

backend/python/transformers-musicgen/requirements-hipblas.txt

@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 transformers
 accelerate
-torch
+torch==2.4.1+rocm6.0

backend/python/transformers-musicgen/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 scipy==1.14.0
 certifi

backend/python/transformers/backend.py

@@ -72,7 +72,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         Returns:
             A Result object that contains the result of the LoadModel operation.
         """
+
         model_name = request.Model
+        
+        # Check to see if the Model exists in the filesystem already.
+        if os.path.exists(request.ModelFile):
+            model_name = request.ModelFile
 
         compute = torch.float16
         if request.F16Memory == True:

backend/python/transformers/requirements-cpu.txt

@@ -1,4 +1,4 @@
-torch
+torch==2.4.1
 accelerate
 transformers
 bitsandbytes

backend/python/transformers/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
 accelerate
 transformers
 bitsandbytes

backend/python/transformers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
+torch==2.4.1
 accelerate
 transformers
 bitsandbytes

backend/python/transformers/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
+torch==2.4.1+rocm6.0
 accelerate
 transformers
 bitsandbytes

backend/python/transformers/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406

backend/python/vall-e-x/requirements-cpu.txt

@@ -1,3 +1,3 @@
 accelerate
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1

backend/python/vall-e-x/requirements-cublas11.txt

@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 accelerate
-torch
-torchaudio
+torch==2.4.1+cu118
+torchaudio==2.4.1+cu118

backend/python/vall-e-x/requirements-cublas12.txt

@@ -1,3 +1,3 @@
 accelerate
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1

backend/python/vall-e-x/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi

backend/python/vllm/backend.py

@@ -5,6 +5,8 @@ import argparse
 import signal
 import sys
 import os
+from typing import List
+from PIL import Image
 
 import backend_pb2
 import backend_pb2_grpc
@@ -15,6 +17,10 @@ from vllm.engine.async_llm_engine import AsyncLLMEngine
 from vllm.sampling_params import SamplingParams
 from vllm.utils import random_uuid
 from vllm.transformers_utils.tokenizer import get_tokenizer
+from vllm.multimodal.utils import fetch_image
+from vllm.assets.video import VideoAsset
+import base64
+import io
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 
@@ -89,6 +95,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
         if request.Quantization != "":
             engine_args.quantization = request.Quantization
+        if request.LoadFormat != "":
+            engine_args.load_format = request.LoadFormat
         if request.GPUMemoryUtilization != 0:
             engine_args.gpu_memory_utilization = request.GPUMemoryUtilization
         if request.TrustRemoteCode:
@@ -105,6 +113,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         try:
             self.llm = AsyncLLMEngine.from_engine_args(engine_args)
         except Exception as err:
+            print(f"Unexpected {err=}, {type(err)=}", file=sys.stderr)
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
 
         try:
@@ -117,7 +126,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            )
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-
+        print("Model loaded successfully", file=sys.stderr)
         return backend_pb2.Result(message="Model loaded successfully", success=True)
 
     async def Predict(self, request, context):
@@ -196,15 +205,35 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if request.Seed != 0:
             sampling_params.seed = request.Seed
 
+        # Extract image paths and process images
         prompt = request.Prompt
-        
-        # If tokenizer template is enabled and messages are provided instead of prompt apply the tokenizer template
+
+        image_paths = request.Images
+        image_data = [self.load_image(img_path) for img_path in image_paths]
+
+        videos_path = request.Videos
+        video_data = [self.load_video(video_path) for video_path in videos_path]
+
+        # If tokenizer template is enabled and messages are provided instead of prompt, apply the tokenizer template
         if not request.Prompt and request.UseTokenizerTemplate and request.Messages:
             prompt = self.tokenizer.apply_chat_template(request.Messages, tokenize=False, add_generation_prompt=True)
 
-        # Generate text
+        # Generate text using the LLM engine
         request_id = random_uuid()
-        outputs = self.llm.generate(prompt, sampling_params, request_id)
+        print(f"Generating text with request_id: {request_id}", file=sys.stderr)
+        multi_modal_data = {}
+        if image_data:
+            multi_modal_data["image"] = image_data
+        if video_data:
+            multi_modal_data["video"] = video_data
+        outputs = self.llm.generate(
+            {
+            "prompt": prompt,
+            "multi_modal_data": multi_modal_data if multi_modal_data else None,
+            },
+            sampling_params=sampling_params,
+            request_id=request_id,
+        )
 
         # Stream the results
         generated_text = ""
@@ -227,9 +256,57 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if streaming:
             return
 
+        # Remove the image files from /tmp folder
+        for img_path in image_paths:
+            try:
+                os.remove(img_path)
+            except Exception as e:
+                print(f"Error removing image file: {img_path}, {e}", file=sys.stderr)
+
         # Sending the final generated text
         yield backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
 
+    def load_image(self, image_path: str):
+        """
+        Load an image from the given file path or base64 encoded data.
+        
+        Args:
+            image_path (str): The path to the image file or base64 encoded data.
+
+        Returns:
+            Image: The loaded image.
+        """
+        try:
+
+            image_data = base64.b64decode(image_path)
+            image = Image.open(io.BytesIO(image_data))
+            return image
+        except Exception as e:
+            print(f"Error loading image {image_path}: {e}", file=sys.stderr)
+            return None
+
+    def load_video(self, video_path: str):
+        """
+        Load a video from the given file path.
+        
+        Args:
+            video_path (str): The path to the image file.
+
+        Returns:
+            Video: The loaded video.
+        """
+        try:
+            timestamp = str(int(time.time() * 1000))  # Generate timestamp
+            p = f"/tmp/vl-{timestamp}.data"  # Use timestamp in filename
+            with open(p, "wb") as f:
+                f.write(base64.b64decode(video_path))
+            video = VideoAsset(name=p).np_ndarrays
+            os.remove(p)
+            return video
+        except Exception as e:
+            print(f"Error loading video {video_path}: {e}", file=sys.stderr)
+            return None
+
 async def serve(address):
     # Start asyncio gRPC server
     server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))

backend/python/vllm/install.sh

@@ -13,4 +13,20 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
-installRequirements
+# We don't embed this into the images as it is a large dependency and not always needed.
+# Besides, the speed inference are not actually usable in the current state for production use-cases.
+if [ "x${BUILD_TYPE}" == "x" ] && [ "x${FROM_SOURCE}" == "xtrue" ]; then
+        ensureVenv
+        # https://docs.vllm.ai/en/v0.6.1/getting_started/cpu-installation.html
+        if [ ! -d vllm ]; then
+            git clone https://github.com/vllm-project/vllm
+        fi
+        pushd vllm
+            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.67.0 protobuf bitsandbytes
+            uv pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
+            VLLM_TARGET_DEVICE=cpu python setup.py install
+        popd
+        rm -rf vllm
+    else
+        installRequirements
+fi

backend/python/vllm/requirements-cpu.txt

@@ -1,3 +1,3 @@
 accelerate
-torch
+torch==2.4.1
 transformers

backend/python/vllm/requirements-cublas11.txt

@@ -1,4 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 accelerate
-torch
-transformers
+torch==2.4.1+cu118
+transformers
+bitsandbytes

backend/python/vllm/requirements-cublas12.txt

@@ -1,3 +1,4 @@
 accelerate
-torch
-transformers
+torch==2.4.1
+transformers
+bitsandbytes

backend/python/vllm/requirements-hipblas.txt

@@ -1,4 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 accelerate
-torch
-transformers
+torch==2.4.1+rocm6.0
+transformers
+bitsandbytes

backend/python/vllm/requirements-intel.txt

@@ -4,4 +4,5 @@ accelerate
 torch
 transformers
 optimum[openvino]
-setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
+bitsandbytes

backend/python/vllm/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.67.0
 protobuf
 certifi
 setuptools

core/backend/embeddings.go

@@ -10,20 +10,11 @@ import (
 )
 
 func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() ([]float32, error), error) {
-	modelFile := backendConfig.Model
-
-	grpcOpts := GRPCModelOpts(backendConfig)
 
 	var inferenceModel interface{}
 	var err error
 
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(*backendConfig.Threads)),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-	})
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	if backendConfig.Backend == "" {
 		inferenceModel, err = loader.GreedyLoader(opts...)

core/backend/image.go

@@ -8,19 +8,8 @@ import (
 )
 
 func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, src, dst string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() error, error) {
-	threads := backendConfig.Threads
-	if *threads == 0 && appConfig.Threads != 0 {
-		threads = &appConfig.Threads
-	}
-	gRPCOpts := GRPCModelOpts(backendConfig)
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithBackendString(backendConfig.Backend),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithThreads(uint32(*threads)),
-		model.WithContext(appConfig.Context),
-		model.WithModel(backendConfig.Model),
-		model.WithLoadGRPCLoadModelOpts(gRPCOpts),
-	})
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	inferenceModel, err := loader.BackendLoader(
 		opts...,

core/backend/llm.go

@@ -2,6 +2,7 @@ package backend
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"os"
 	"regexp"
@@ -33,22 +34,11 @@ type TokenUsage struct {
 
 func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model
-	threads := c.Threads
-	if *threads == 0 && o.Threads != 0 {
-		threads = &o.Threads
-	}
-	grpcOpts := GRPCModelOpts(c)
 
 	var inferenceModel grpc.Backend
 	var err error
 
-	opts := modelOpts(c, o, []model.Option{
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(*threads)), // some models uses this to allocate threads during startup
-		model.WithAssetDir(o.AssetsDestination),
-		model.WithModel(modelFile),
-		model.WithContext(o.Context),
-	})
+	opts := ModelOptions(c, o, []model.Option{})
 
 	if c.Backend != "" {
 		opts = append(opts, model.WithBackendString(c.Backend))
@@ -88,6 +78,16 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 			switch ct := message.Content.(type) {
 			case string:
 				protoMessages[i].Content = ct
+			case []interface{}:
+				// If using the tokenizer template, in case of multimodal we want to keep the multimodal content as and return only strings here
+				data, _ := json.Marshal(ct)
+				resultData := []struct {
+					Text string `json:"text"`
+				}{}
+				json.Unmarshal(data, &resultData)
+				for _, r := range resultData {
+					protoMessages[i].Content += r.Text
+				}
 			default:
 				return nil, fmt.Errorf("unsupported type for schema.Message.Content for inference: %T", ct)
 			}

core/backend/options.go

@@ -11,32 +11,65 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
-func modelOpts(c config.BackendConfig, so *config.ApplicationConfig, opts []model.Option) []model.Option {
+func ModelOptions(c config.BackendConfig, so *config.ApplicationConfig, opts []model.Option) []model.Option {
+	name := c.Name
+	if name == "" {
+		name = c.Model
+	}
+
+	defOpts := []model.Option{
+		model.WithBackendString(c.Backend),
+		model.WithModel(c.Model),
+		model.WithAssetDir(so.AssetsDestination),
+		model.WithContext(so.Context),
+		model.WithModelID(name),
+	}
+
+	threads := 1
+
+	if c.Threads != nil {
+		threads = *c.Threads
+	}
+
+	if so.Threads != 0 {
+		threads = so.Threads
+	}
+
+	c.Threads = &threads
+
+	grpcOpts := grpcModelOpts(c)
+	defOpts = append(defOpts, model.WithLoadGRPCLoadModelOpts(grpcOpts))
+
 	if so.SingleBackend {
-		opts = append(opts, model.WithSingleActiveBackend())
+		defOpts = append(defOpts, model.WithSingleActiveBackend())
 	}
 
 	if so.ParallelBackendRequests {
-		opts = append(opts, model.EnableParallelRequests)
+		defOpts = append(defOpts, model.EnableParallelRequests)
 	}
 
 	if c.GRPC.Attempts != 0 {
-		opts = append(opts, model.WithGRPCAttempts(c.GRPC.Attempts))
+		defOpts = append(defOpts, model.WithGRPCAttempts(c.GRPC.Attempts))
 	}
 
 	if c.GRPC.AttemptsSleepTime != 0 {
-		opts = append(opts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
+		defOpts = append(defOpts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
 	}
 
 	for k, v := range so.ExternalGRPCBackends {
-		opts = append(opts, model.WithExternalBackend(k, v))
+		defOpts = append(defOpts, model.WithExternalBackend(k, v))
 	}
 
-	return opts
+	return append(defOpts, opts...)
 }
 
 func getSeed(c config.BackendConfig) int32 {
-	seed := int32(*c.Seed)
+	var seed int32 = config.RAND_SEED
+
+	if c.Seed != nil {
+		seed = int32(*c.Seed)
+	}
+
 	if seed == config.RAND_SEED {
 		seed = rand.Int31()
 	}
@@ -44,11 +77,47 @@ func getSeed(c config.BackendConfig) int32 {
 	return seed
 }
 
-func GRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
+func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 	b := 512
 	if c.Batch != 0 {
 		b = c.Batch
 	}
+
+	f16 := false
+	if c.F16 != nil {
+		f16 = *c.F16
+	}
+
+	embeddings := false
+	if c.Embeddings != nil {
+		embeddings = *c.Embeddings
+	}
+
+	lowVRAM := false
+	if c.LowVRAM != nil {
+		lowVRAM = *c.LowVRAM
+	}
+
+	mmap := false
+	if c.MMap != nil {
+		mmap = *c.MMap
+	}
+
+	ctxSize := 1024
+	if c.ContextSize != nil {
+		ctxSize = *c.ContextSize
+	}
+
+	mmlock := false
+	if c.MMlock != nil {
+		mmlock = *c.MMlock
+	}
+
+	nGPULayers := 9999999
+	if c.NGPULayers != nil {
+		nGPULayers = *c.NGPULayers
+	}
+
 	return &pb.ModelOptions{
 		CUDA:                 c.CUDA || c.Diffusers.CUDA,
 		SchedulerType:        c.Diffusers.SchedulerType,
@@ -56,20 +125,21 @@ func GRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		CFGScale:             c.Diffusers.CFGScale,
 		LoraAdapter:          c.LoraAdapter,
 		LoraScale:            c.LoraScale,
-		F16Memory:            *c.F16,
+		F16Memory:            f16,
 		LoraBase:             c.LoraBase,
 		IMG2IMG:              c.Diffusers.IMG2IMG,
 		CLIPModel:            c.Diffusers.ClipModel,
 		CLIPSubfolder:        c.Diffusers.ClipSubFolder,
 		CLIPSkip:             int32(c.Diffusers.ClipSkip),
 		ControlNet:           c.Diffusers.ControlNet,
-		ContextSize:          int32(*c.ContextSize),
+		ContextSize:          int32(ctxSize),
 		Seed:                 getSeed(c),
 		NBatch:               int32(b),
 		NoMulMatQ:            c.NoMulMatQ,
 		DraftModel:           c.DraftModel,
 		AudioPath:            c.VallE.AudioPath,
 		Quantization:         c.Quantization,
+		LoadFormat:           c.LoadFormat,
 		GPUMemoryUtilization: c.GPUMemoryUtilization,
 		TrustRemoteCode:      c.TrustRemoteCode,
 		EnforceEager:         c.EnforceEager,
@@ -85,16 +155,16 @@ func GRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		YarnBetaSlow:         c.YarnBetaSlow,
 		NGQA:                 c.NGQA,
 		RMSNormEps:           c.RMSNormEps,
-		MLock:                *c.MMlock,
+		MLock:                mmlock,
 		RopeFreqBase:         c.RopeFreqBase,
 		RopeScaling:          c.RopeScaling,
 		Type:                 c.ModelType,
 		RopeFreqScale:        c.RopeFreqScale,
 		NUMA:                 c.NUMA,
-		Embeddings:           *c.Embeddings,
-		LowVRAM:              *c.LowVRAM,
-		NGPULayers:           int32(*c.NGPULayers),
-		MMap:                 *c.MMap,
+		Embeddings:           embeddings,
+		LowVRAM:              lowVRAM,
+		NGPULayers:           int32(nGPULayers),
+		MMap:                 mmap,
 		MainGPU:              c.MainGPU,
 		Threads:              int32(*c.Threads),
 		TensorSplit:          c.TensorSplit,

core/backend/rerank.go

@@ -9,21 +9,9 @@ import (
 	model "github.com/mudler/LocalAI/pkg/model"
 )
 
-func Rerank(backend, modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
-	bb := backend
-	if bb == "" {
-		return nil, fmt.Errorf("backend is required")
-	}
+func Rerank(modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
 
-	grpcOpts := GRPCModelOpts(backendConfig)
-
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
-		model.WithBackendString(bb),
-		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-	})
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
 	rerankModel, err := loader.BackendLoader(opts...)
 	if err != nil {
 		return nil, err

core/backend/soundgeneration.go

@@ -13,7 +13,6 @@ import (
 )
 
 func SoundGeneration(
-	backend string,
 	modelFile string,
 	text string,
 	duration *float32,
@@ -25,18 +24,8 @@ func SoundGeneration(
 	appConfig *config.ApplicationConfig,
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {
-	if backend == "" {
-		return "", nil, fmt.Errorf("backend is a required parameter")
-	}
 
-	grpcOpts := GRPCModelOpts(backendConfig)
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
-		model.WithBackendString(backend),
-		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-	})
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
 
 	soundGenModel, err := loader.BackendLoader(opts...)
 	if err != nil {

core/backend/token_metrics.go

@@ -0,0 +1,33 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/pkg/grpc/proto"
+	model "github.com/mudler/LocalAI/pkg/model"
+)
+
+func TokenMetrics(
+	modelFile string,
+	loader *model.ModelLoader,
+	appConfig *config.ApplicationConfig,
+	backendConfig config.BackendConfig) (*proto.MetricsResponse, error) {
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{
+		model.WithModel(modelFile),
+	})
+	model, err := loader.BackendLoader(opts...)
+	if err != nil {
+		return nil, err
+	}
+
+	if model == nil {
+		return nil, fmt.Errorf("could not loadmodel model")
+	}
+
+	res, err := model.GetTokenMetrics(context.Background(), &proto.MetricsRequest{})
+
+	return res, err
+}

core/backend/tokenize.go

@@ -0,0 +1,44 @@
+package backend
+
+import (
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/grpc"
+	model "github.com/mudler/LocalAI/pkg/model"
+)
+
+func ModelTokenize(s string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (schema.TokenizeResponse, error) {
+
+	modelFile := backendConfig.Model
+
+	var inferenceModel grpc.Backend
+	var err error
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{
+		model.WithModel(modelFile),
+	})
+
+	if backendConfig.Backend == "" {
+		inferenceModel, err = loader.GreedyLoader(opts...)
+	} else {
+		opts = append(opts, model.WithBackendString(backendConfig.Backend))
+		inferenceModel, err = loader.BackendLoader(opts...)
+	}
+	if err != nil {
+		return schema.TokenizeResponse{}, err
+	}
+
+	predictOptions := gRPCPredictOpts(backendConfig, loader.ModelPath)
+	predictOptions.Prompt = s
+
+	// tokenize the string
+	resp, err := inferenceModel.TokenizeString(appConfig.Context, predictOptions)
+	if err != nil {
+		return schema.TokenizeResponse{}, err
+	}
+
+	return schema.TokenizeResponse{
+		Tokens: resp.Tokens,
+	}, nil
+
+}

core/backend/transcript.go

@@ -14,13 +14,11 @@ import (
 
 func ModelTranscription(audio, language string, translate bool, ml *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (*schema.TranscriptionResult, error) {
 
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithBackendString(model.WhisperBackend),
-		model.WithModel(backendConfig.Model),
-		model.WithContext(appConfig.Context),
-		model.WithThreads(uint32(*backendConfig.Threads)),
-		model.WithAssetDir(appConfig.AssetsDestination),
-	})
+	if backendConfig.Backend == "" {
+		backendConfig.Backend = model.WhisperBackend
+	}
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	transcriptionModel, err := ml.BackendLoader(opts...)
 	if err != nil {

core/backend/tts.go

@@ -28,14 +28,9 @@ func ModelTTS(
 		bb = model.PiperBackend
 	}
 
-	grpcOpts := GRPCModelOpts(backendConfig)
-
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
+	opts := ModelOptions(config.BackendConfig{}, appConfig, []model.Option{
 		model.WithBackendString(bb),
 		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
 	})
 	ttsModel, err := loader.BackendLoader(opts...)
 	if err != nil {