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/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
@@ -197,7 +197,7 @@ jobs:
             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
@@ -235,7 +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

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?=d5ed2b929d85bbd7dbeecb690880f07d9d7a6077
+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?=ccc2547210e09e3a1785817383ab770389bb442b
+WHISPER_CPP_VERSION?=0fbaac9c891055796456df7b9122a70c220f9ca1
 
 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -470,13 +470,13 @@ run-e2e-image:
 
 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

@@ -66,6 +66,21 @@ 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)
 
 ## 📰 Latest project news

backend/backend.proto

@@ -219,6 +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());
@@ -578,12 +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());
                 }
@@ -600,7 +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;
@@ -629,7 +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);
@@ -769,7 +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});
@@ -801,7 +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
         {
@@ -885,9 +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;
 
@@ -914,13 +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)
@@ -934,7 +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)
                 {
@@ -1009,7 +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
@@ -1160,7 +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 {
@@ -1216,7 +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)
@@ -1268,7 +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
@@ -1379,7 +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))
                 {
@@ -1398,8 +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;
@@ -1408,7 +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()) ?
@@ -1418,7 +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;
             }
         }
@@ -1550,7 +1582,7 @@ struct llama_server_context
             update_system_prompt();
         }
 
-        llama_batch_clear(batch);
+        common_batch_clear(batch);
 
         if (all_slots_are_idle)
         {
@@ -1628,7 +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;
         }
 
@@ -1722,7 +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;
@@ -1734,7 +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);
@@ -1826,7 +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++;
                     }
 
@@ -1904,7 +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);
@@ -1943,9 +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)
@@ -1956,7 +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({
@@ -2009,7 +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); }
 };
 
@@ -2203,7 +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
 
@@ -2311,7 +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();

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==1.66.2
+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==1.66.2
+grpcio==1.67.0
 protobuf
 certifi

backend/python/common/template/requirements.txt

@@ -1,2 +1,2 @@
-grpcio==1.66.2
+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==1.66.2
+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==1.66.2
+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==1.66.2
+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==1.66.2
+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==1.66.2
+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==4.44.1
 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==1.66.2
+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==1.66.2
+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==1.66.2
+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==3.1.1
+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==3.1.1
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
+torch==2.4.1
 accelerate
-sentence-transformers==3.1.1
+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==3.1.1
+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==3.1.1
+sentence-transformers==3.2.0
 transformers

backend/python/sentencetransformers/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.2
+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==1.66.2
+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==1.66.2
+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==1.66.2
+grpcio==1.67.0
 protobuf
 certifi

backend/python/vllm/backend.py

@@ -19,6 +19,8 @@ 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
 
@@ -93,6 +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:
@@ -217,13 +221,15 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         # Generate text using the LLM engine
         request_id = random_uuid()
         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": {
-                    "image": image_data if image_data else None,
-                    "video": video_data if video_data else None,
-                } if image_data or video_data else None,
+            "prompt": prompt,
+            "multi_modal_data": multi_modal_data if multi_modal_data else None,
             },
             sampling_params=sampling_params,
             request_id=request_id,
@@ -262,19 +268,22 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
     def load_image(self, image_path: str):
         """
-        Load an image from the given file path.
+        Load an image from the given file path or base64 encoded data.
         
         Args:
-            image_path (str): The path to the image file.
+            image_path (str): The path to the image file or base64 encoded data.
 
         Returns:
             Image: The loaded image.
         """
         try:
-            return Image.open(image_path)
+
+            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 self.load_video(image_path)
+            return None
 
     def load_video(self, video_path: str):
         """
@@ -287,10 +296,15 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             Video: The loaded video.
         """
         try:
-            video = VideoAsset(name=video_path).np_ndarrays
+            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 {image_path}: {e}", file=sys.stderr)
+            print(f"Error loading video {video_path}: {e}", file=sys.stderr)
             return None
 
 async def serve(address):

backend/python/vllm/install.sh

@@ -13,14 +13,16 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
-if [ "x${BUILD_TYPE}" == "x" ]; then
+# We don't embed this into the images as it is a large dependency and not always needed.
+# Besides, the speed inference are not actually usable in the current state for production use-cases.
+if [ "x${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==1.66.2 protobuf bitsandbytes
+            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

backend/python/vllm/requirements-cpu.txt

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

backend/python/vllm/requirements-cublas11.txt

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

backend/python/vllm/requirements-cublas12.txt

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

backend/python/vllm/requirements-hipblas.txt

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

backend/python/vllm/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.2
+grpcio==1.67.0
 protobuf
 certifi
 setuptools

core/backend/llm.go

@@ -2,6 +2,7 @@ package backend
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"os"
 	"regexp"
@@ -77,6 +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

@@ -139,6 +139,7 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		DraftModel:           c.DraftModel,
 		AudioPath:            c.VallE.AudioPath,
 		Quantization:         c.Quantization,
+		LoadFormat:           c.LoadFormat,
 		GPUMemoryUtilization: c.GPUMemoryUtilization,
 		TrustRemoteCode:      c.TrustRemoteCode,
 		EnforceEager:         c.EnforceEager,

core/cli/run.go

@@ -53,6 +53,7 @@ type RunCMD struct {
 	OpaqueErrors                       bool     `env:"LOCALAI_OPAQUE_ERRORS" default:"false" help:"If true, all error responses are replaced with blank 500 errors. This is intended only for hardening against information leaks and is normally not recommended." group:"hardening"`
 	UseSubtleKeyComparison             bool     `env:"LOCALAI_SUBTLE_KEY_COMPARISON" default:"false" help:"If true, API Key validation comparisons will be performed using constant-time comparisons rather than simple equality. This trades off performance on each request for resiliancy against timing attacks." group:"hardening"`
 	DisableApiKeyRequirementForHttpGet bool     `env:"LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET" default:"false" help:"If true, a valid API key is not required to issue GET requests to portions of the web ui. This should only be enabled in secure testing environments" group:"hardening"`
+	DisableMetricsEndpoint             bool     `env:"LOCALAI_DISABLE_METRICS_ENDPOINT,DISABLE_METRICS_ENDPOINT" default:"false" help:"Disable the /metrics endpoint" group:"api"`
 	HttpGetExemptedEndpoints           []string `env:"LOCALAI_HTTP_GET_EXEMPTED_ENDPOINTS" default:"^/$,^/browse/?$,^/talk/?$,^/p2p/?$,^/chat/?$,^/text2image/?$,^/tts/?$,^/static/.*$,^/swagger.*$" help:"If LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET is overriden to true, this is the list of endpoints to exempt. Only adjust this in case of a security incident or as a result of a personal security posture review" group:"hardening"`
 	Peer2Peer                          bool     `env:"LOCALAI_P2P,P2P" name:"p2p" default:"false" help:"Enable P2P mode" group:"p2p"`
 	Peer2PeerDHTInterval               int      `env:"LOCALAI_P2P_DHT_INTERVAL,P2P_DHT_INTERVAL" default:"360" name:"p2p-dht-interval" help:"Interval for DHT refresh (used during token generation)" group:"p2p"`
@@ -108,6 +109,10 @@ func (r *RunCMD) Run(ctx *cliContext.Context) error {
 		config.WithLoadToMemory(r.LoadToMemory),
 	}
 
+	if r.DisableMetricsEndpoint {
+		opts = append(opts, config.DisableMetricsEndpoint)
+	}
+
 	token := ""
 	if r.Peer2Peer || r.Peer2PeerToken != "" {
 		log.Info().Msg("P2P mode enabled")

core/config/application_config.go

@@ -39,6 +39,7 @@ type ApplicationConfig struct {
 	OpaqueErrors                       bool
 	UseSubtleKeyComparison             bool
 	DisableApiKeyRequirementForHttpGet bool
+	DisableMetrics                     bool
 	HttpGetExemptedEndpoints           []*regexp.Regexp
 	DisableGalleryEndpoint             bool
 	LoadToMemory                       []string
@@ -350,6 +351,10 @@ func WithDisableApiKeyRequirementForHttpGet(required bool) AppOption {
 	}
 }
 
+var DisableMetricsEndpoint AppOption = func(o *ApplicationConfig) {
+	o.DisableMetrics = true
+}
+
 func WithHttpGetExemptedEndpoints(endpoints []string) AppOption {
 	return func(o *ApplicationConfig) {
 		o.HttpGetExemptedEndpoints = []*regexp.Regexp{}

core/config/backend_config.go

@@ -143,6 +143,7 @@ type LLMConfig struct {
 	DraftModel           string  `yaml:"draft_model"`
 	NDraft               int32   `yaml:"n_draft"`
 	Quantization         string  `yaml:"quantization"`
+	LoadFormat           string  `yaml:"load_format"`
 	GPUMemoryUtilization float32 `yaml:"gpu_memory_utilization"` // vLLM
 	TrustRemoteCode      bool    `yaml:"trust_remote_code"`      // vLLM
 	EnforceEager         bool    `yaml:"enforce_eager"`          // vLLM
@@ -197,9 +198,7 @@ type TemplateConfig struct {
 	// It defaults to \n
 	JoinChatMessagesByCharacter *string `yaml:"join_chat_messages_by_character"`
 
-	Video string `yaml:"video"`
-	Image string `yaml:"image"`
-	Audio string `yaml:"audio"`
+	Multimodal string `yaml:"multimodal"`
 }
 
 func (c *BackendConfig) UnmarshalYAML(value *yaml.Node) error {

core/http/app.go

@@ -109,19 +109,21 @@ func App(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *confi
 		app.Use(recover.New())
 	}
 
-	metricsService, err := services.NewLocalAIMetricsService()
-	if err != nil {
-		return nil, err
-	}
+	if !appConfig.DisableMetrics {
+		metricsService, err := services.NewLocalAIMetricsService()
+		if err != nil {
+			return nil, err
+		}
 
-	if metricsService != nil {
-		app.Use(localai.LocalAIMetricsAPIMiddleware(metricsService))
-		app.Hooks().OnShutdown(func() error {
-			return metricsService.Shutdown()
-		})
-	}
+		if metricsService != nil {
+			app.Use(localai.LocalAIMetricsAPIMiddleware(metricsService))
+			app.Hooks().OnShutdown(func() error {
+				return metricsService.Shutdown()
+			})
+		}
 
- // Health Checks should always be exempt from auth, so register these first
+	}
+	// Health Checks should always be exempt from auth, so register these first
 	routes.HealthRoutes(app)
 
 	kaConfig, err := middleware.GetKeyAuthConfig(appConfig)

core/http/app_test.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+	"strings"
 
 	"github.com/mudler/LocalAI/core/config"
 	. "github.com/mudler/LocalAI/core/http"
@@ -950,7 +951,7 @@ var _ = Describe("API test", func() {
 					openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
 				Expect(err).ToNot(HaveOccurred())
 				Expect(len(resp.Choices) > 0).To(BeTrue())
-				Expect(resp.Choices[0].Message.Content).To(Or(ContainSubstring("Sure"), ContainSubstring("five")))
+				Expect(strings.ToLower(resp.Choices[0].Message.Content)).To(Or(ContainSubstring("sure"), ContainSubstring("five")))
 
 				stream, err := client.CreateChatCompletionStream(context.TODO(), openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
 				Expect(err).ToNot(HaveOccurred())
@@ -969,7 +970,7 @@ var _ = Describe("API test", func() {
 					tokens++
 				}
 				Expect(text).ToNot(BeEmpty())
-				Expect(text).To(Or(ContainSubstring("Sure"), ContainSubstring("five")))
+				Expect(strings.ToLower(text)).To(Or(ContainSubstring("sure"), ContainSubstring("five")))
 
 				Expect(tokens).ToNot(Or(Equal(1), Equal(0)))
 			})

core/http/elements/gallery.go

@@ -6,6 +6,7 @@ import (
 
 	"github.com/chasefleming/elem-go"
 	"github.com/chasefleming/elem-go/attrs"
+	"github.com/microcosm-cc/bluemonday"
 	"github.com/mudler/LocalAI/core/gallery"
 	"github.com/mudler/LocalAI/core/p2p"
 	"github.com/mudler/LocalAI/core/services"
@@ -41,7 +42,7 @@ func DoneProgress(galleryID, text string, showDelete bool) string {
 				"tabindex":  "-1",
 				"autofocus": "",
 			},
-			elem.Text(text),
+			elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 		),
 		elem.If(showDelete, deleteButton(galleryID, modelName), reInstallButton(galleryID)),
 	).Render()
@@ -57,7 +58,7 @@ func ErrorProgress(err, galleryName string) string {
 				"tabindex":  "-1",
 				"autofocus": "",
 			},
-			elem.Text("Error "+err),
+			elem.Text("Error "+bluemonday.StrictPolicy().Sanitize(err)),
 		),
 		installButton(galleryName),
 	).Render()
@@ -170,7 +171,7 @@ func P2PNodeBoxes(nodes []p2p.NodeData) string {
 						attrs.Props{
 							"class": "text-gray-200 font-semibold ml-2 mr-1",
 						},
-						elem.Text(n.ID),
+						elem.Text(bluemonday.StrictPolicy().Sanitize(n.ID)),
 					),
 					elem.Text("Status: "),
 					elem.If(
@@ -227,7 +228,7 @@ func StartProgressBar(uid, progress, text string) string {
 				"tabindex":  "-1",
 				"autofocus": "",
 			},
-			elem.Text(text),
+			elem.Text(bluemonday.StrictPolicy().Sanitize(text)), //Perhaps overly defensive
 			elem.Div(attrs.Props{
 				"hx-get":     "/browse/job/progress/" + uid,
 				"hx-trigger": "every 600ms",
@@ -249,9 +250,7 @@ func cardSpan(text, icon string) elem.Node {
 			"class": icon + " pr-2",
 		}),
 
-		elem.Text(text),
-
-		//elem.Text(text),
+		elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 	)
 }
 
@@ -285,11 +284,9 @@ func searchableElement(text, icon string) elem.Node {
 				elem.I(attrs.Props{
 					"class": icon + " pr-2",
 				}),
-				elem.Text(text),
+				elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 			),
 		),
-
-		//elem.Text(text),
 	)
 }
 
@@ -303,7 +300,7 @@ func link(text, url string) elem.Node {
 		elem.I(attrs.Props{
 			"class": "fas fa-link pr-2",
 		}),
-		elem.Text(text),
+		elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 	)
 }
 func installButton(galleryName string) elem.Node {
@@ -387,13 +384,13 @@ func ListModels(models []*gallery.GalleryModel, processTracker ProcessTracker, g
 				attrs.Props{
 					"class": "mb-2 text-xl font-bold leading-tight",
 				},
-				elem.Text(m.Name),
+				elem.Text(bluemonday.StrictPolicy().Sanitize(m.Name)),
 			),
 			elem.P(
 				attrs.Props{
 					"class": "mb-4 text-sm [&:not(:hover)]:truncate text-base",
 				},
-				elem.Text(m.Description),
+				elem.Text(bluemonday.StrictPolicy().Sanitize(m.Description)),
 			),
 		)
 	}

core/http/endpoints/localai/welcome.go

@@ -13,15 +13,10 @@ import (
 func WelcomeEndpoint(appConfig *config.ApplicationConfig,
 	cl *config.BackendConfigLoader, ml *model.ModelLoader, modelStatus func() (map[string]string, map[string]string)) func(*fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		models, _ := services.ListModels(cl, ml, config.NoFilterFn, services.SKIP_IF_CONFIGURED)
 		backendConfigs := cl.GetAllBackendConfigs()
-
 		galleryConfigs := map[string]*gallery.Config{}
-		modelsWithBackendConfig := map[string]interface{}{}
 
 		for _, m := range backendConfigs {
-			modelsWithBackendConfig[m.Name] = nil
-
 			cfg, err := gallery.GetLocalModelConfiguration(ml.ModelPath, m.Name)
 			if err != nil {
 				continue
@@ -29,13 +24,15 @@ func WelcomeEndpoint(appConfig *config.ApplicationConfig,
 			galleryConfigs[m.Name] = cfg
 		}
 
+		modelsWithoutConfig, _ := services.ListModels(cl, ml, config.NoFilterFn, services.LOOSE_ONLY)
+
 		// Get model statuses to display in the UI the operation in progress
 		processingModels, taskTypes := modelStatus()
 
 		summary := fiber.Map{
 			"Title":             "LocalAI API - " + internal.PrintableVersion(),
 			"Version":           internal.PrintableVersion(),
-			"Models":            models,
+			"Models":            modelsWithoutConfig,
 			"ModelsConfig":      backendConfigs,
 			"GalleryConfig":     galleryConfigs,
 			"IsP2PEnabled":      p2p.IsP2PEnabled(),

core/http/endpoints/openai/assistant.go

@@ -10,6 +10,7 @@ import (
 	"time"
 
 	"github.com/gofiber/fiber/v2"
+	"github.com/microcosm-cc/bluemonday"
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/core/services"
@@ -83,7 +84,7 @@ func CreateAssistantEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoad
 
 		if !modelExists(cl, ml, request.Model) {
 			log.Warn().Msgf("Model: %s was not found in list of models.", request.Model)
-			return c.Status(fiber.StatusBadRequest).SendString("Model " + request.Model + " not found")
+			return c.Status(fiber.StatusBadRequest).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Model %q not found", request.Model)))
 		}
 
 		if request.Tools == nil {
@@ -147,7 +148,7 @@ func ListAssistantsEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoade
 		// Convert string limit to integer
 		limit, err := strconv.Atoi(limitQuery)
 		if err != nil {
-			return c.Status(http.StatusBadRequest).SendString(fmt.Sprintf("Invalid limit query value: %s", limitQuery))
+			return c.Status(http.StatusBadRequest).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Invalid limit query value: %s", limitQuery)))
 		}
 
 		// Sort assistants
@@ -288,7 +289,7 @@ func GetAssistantEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader,
 			}
 		}
 
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant with id: %s", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant with id: %s", assistantID)))
 	}
 }
 
@@ -337,11 +338,11 @@ func CreateAssistantFileEndpoint(cl *config.BackendConfigLoader, ml *model.Model
 					}
 				}
 
-				return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find file_id: %s", request.FileID))
+				return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find file_id: %s", request.FileID)))
 			}
 		}
 
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find %q", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find %q", assistantID)))
 	}
 }
 
@@ -442,7 +443,7 @@ func ModifyAssistantEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoad
 				return c.Status(fiber.StatusOK).JSON(newAssistant)
 			}
 		}
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant with id: %s", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant with id: %s", assistantID)))
 	}
 }
 
@@ -513,9 +514,9 @@ func GetAssistantFileEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoa
 				if assistantFile.ID == fileId {
 					return c.Status(fiber.StatusOK).JSON(assistantFile)
 				}
-				return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant file with file_id: %s", fileId))
+				return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant file with file_id: %s", fileId)))
 			}
 		}
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant file with assistant_id: %s", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant file with assistant_id: %s", assistantID)))
 	}
 }

core/http/endpoints/openai/files.go

@@ -8,6 +8,7 @@ import (
 	"sync/atomic"
 	"time"
 
+	"github.com/microcosm-cc/bluemonday"
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 
@@ -49,7 +50,7 @@ func UploadFilesEndpoint(cm *config.BackendConfigLoader, appConfig *config.Appli
 
 		err = c.SaveFile(file, savePath)
 		if err != nil {
-			return c.Status(fiber.StatusInternalServerError).SendString("Failed to save file: " + err.Error())
+			return c.Status(fiber.StatusInternalServerError).SendString("Failed to save file: " + bluemonday.StrictPolicy().Sanitize(err.Error()))
 		}
 
 		f := schema.File{
@@ -121,7 +122,7 @@ func GetFilesEndpoint(cm *config.BackendConfigLoader, appConfig *config.Applicat
 	return func(c *fiber.Ctx) error {
 		file, err := getFileFromRequest(c)
 		if err != nil {
-			return c.Status(fiber.StatusInternalServerError).SendString(err.Error())
+			return c.Status(fiber.StatusInternalServerError).SendString(bluemonday.StrictPolicy().Sanitize(err.Error()))
 		}
 
 		return c.JSON(file)
@@ -143,14 +144,14 @@ func DeleteFilesEndpoint(cm *config.BackendConfigLoader, appConfig *config.Appli
 	return func(c *fiber.Ctx) error {
 		file, err := getFileFromRequest(c)
 		if err != nil {
-			return c.Status(fiber.StatusInternalServerError).SendString(err.Error())
+			return c.Status(fiber.StatusInternalServerError).SendString(bluemonday.StrictPolicy().Sanitize(err.Error()))
 		}
 
 		err = os.Remove(filepath.Join(appConfig.UploadDir, file.Filename))
 		if err != nil {
 			// If the file doesn't exist then we should just continue to remove it
 			if !errors.Is(err, os.ErrNotExist) {
-				return c.Status(fiber.StatusInternalServerError).SendString(fmt.Sprintf("Unable to delete file: %s, %v", file.Filename, err))
+				return c.Status(fiber.StatusInternalServerError).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to delete file: %s, %v", file.Filename, err)))
 			}
 		}
 
@@ -180,12 +181,12 @@ func GetFilesContentsEndpoint(cm *config.BackendConfigLoader, appConfig *config.
 	return func(c *fiber.Ctx) error {
 		file, err := getFileFromRequest(c)
 		if err != nil {
-			return c.Status(fiber.StatusInternalServerError).SendString(err.Error())
+			return c.Status(fiber.StatusInternalServerError).SendString(bluemonday.StrictPolicy().Sanitize(err.Error()))
 		}
 
 		fileContents, err := os.ReadFile(filepath.Join(appConfig.UploadDir, file.Filename))
 		if err != nil {
-			return c.Status(fiber.StatusInternalServerError).SendString(err.Error())
+			return c.Status(fiber.StatusInternalServerError).SendString(bluemonday.StrictPolicy().Sanitize(err.Error()))
 		}
 
 		return c.Send(fileContents)

core/http/endpoints/openai/request.go

@@ -149,6 +149,10 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 	// Decode each request's message content
 	imgIndex, vidIndex, audioIndex := 0, 0, 0
 	for i, m := range input.Messages {
+		nrOfImgsInMessage := 0
+		nrOfVideosInMessage := 0
+		nrOfAudiosInMessage := 0
+
 		switch content := m.Content.(type) {
 		case string:
 			input.Messages[i].StringContent = content
@@ -156,11 +160,16 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 			dat, _ := json.Marshal(content)
 			c := []schema.Content{}
 			json.Unmarshal(dat, &c)
+
+			textContent := ""
+			// we will template this at the end
+
 		CONTENT:
 			for _, pp := range c {
 				switch pp.Type {
 				case "text":
-					input.Messages[i].StringContent = pp.Text
+					textContent += pp.Text
+					//input.Messages[i].StringContent = pp.Text
 				case "video", "video_url":
 					// Decode content as base64 either if it's an URL or base64 text
 					base64, err := utils.GetContentURIAsBase64(pp.VideoURL.URL)
@@ -169,14 +178,8 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 						continue CONTENT
 					}
 					input.Messages[i].StringVideos = append(input.Messages[i].StringVideos, base64) // TODO: make sure that we only return base64 stuff
-
-					t := "[vid-{{.ID}}]{{.Text}}"
-					if config.TemplateConfig.Video != "" {
-						t = config.TemplateConfig.Video
-					}
-					// set a placeholder for each image
-					input.Messages[i].StringContent, _ = templates.TemplateMultiModal(t, vidIndex, input.Messages[i].StringContent)
 					vidIndex++
+					nrOfVideosInMessage++
 				case "audio_url", "audio":
 					// Decode content as base64 either if it's an URL or base64 text
 					base64, err := utils.GetContentURIAsBase64(pp.AudioURL.URL)
@@ -185,13 +188,8 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 						continue CONTENT
 					}
 					input.Messages[i].StringAudios = append(input.Messages[i].StringAudios, base64) // TODO: make sure that we only return base64 stuff
-					// set a placeholder for each image
-					t := "[audio-{{.ID}}]{{.Text}}"
-					if config.TemplateConfig.Audio != "" {
-						t = config.TemplateConfig.Audio
-					}
-					input.Messages[i].StringContent, _ = templates.TemplateMultiModal(t, audioIndex, input.Messages[i].StringContent)
 					audioIndex++
+					nrOfAudiosInMessage++
 				case "image_url", "image":
 					// Decode content as base64 either if it's an URL or base64 text
 					base64, err := utils.GetContentURIAsBase64(pp.ImageURL.URL)
@@ -200,16 +198,21 @@ func updateRequestConfig(config *config.BackendConfig, input *schema.OpenAIReque
 						continue CONTENT
 					}
 
-					t := "[img-{{.ID}}]{{.Text}}"
-					if config.TemplateConfig.Image != "" {
-						t = config.TemplateConfig.Image
-					}
 					input.Messages[i].StringImages = append(input.Messages[i].StringImages, base64) // TODO: make sure that we only return base64 stuff
-					// set a placeholder for each image
-					input.Messages[i].StringContent, _ = templates.TemplateMultiModal(t, imgIndex, input.Messages[i].StringContent)
+
 					imgIndex++
+					nrOfImgsInMessage++
 				}
 			}
+
+			input.Messages[i].StringContent, _ = templates.TemplateMultiModal(config.TemplateConfig.Multimodal, templates.MultiModalOptions{
+				TotalImages:     imgIndex,
+				TotalVideos:     vidIndex,
+				TotalAudios:     audioIndex,
+				ImagesInMessage: nrOfImgsInMessage,
+				VideosInMessage: nrOfVideosInMessage,
+				AudiosInMessage: nrOfAudiosInMessage,
+			}, textContent)
 		}
 	}
 

core/http/middleware/auth.go

@@ -7,6 +7,7 @@ import (
 	"github.com/dave-gray101/v2keyauth"
 	"github.com/gofiber/fiber/v2"
 	"github.com/gofiber/fiber/v2/middleware/keyauth"
+	"github.com/microcosm-cc/bluemonday"
 	"github.com/mudler/LocalAI/core/config"
 )
 
@@ -38,7 +39,7 @@ func getApiKeyErrorHandler(applicationConfig *config.ApplicationConfig) fiber.Er
 			if applicationConfig.OpaqueErrors {
 				return ctx.SendStatus(403)
 			}
-			return ctx.Status(403).SendString(err.Error())
+			return ctx.Status(403).SendString(bluemonday.StrictPolicy().Sanitize(err.Error()))
 		}
 		if applicationConfig.OpaqueErrors {
 			return ctx.SendStatus(500)