feat: embedded model configurations, add popular model examples, refactoring (#1532)

* move downloader out * separate startup functions for preloading configuration files * docs: add popular model examples Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * shorteners * Add llava * Add mistral-openorca * Better link to build section * docs: update * fixup * Drop code dups * Minor fixups * Apply suggestions from code review Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com> * ci: try to cache gRPC build during tests Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * ci: do not build all images for tests, just necessary * ci: cache gRPC also in release pipeline * fixes * Update model_preload_test.go Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
2024-12-24 06:46:39 +00:00 · 2024-01-05 17:16:33 -05:00 · 2024-01-05 17:16:33 -05:00 · 09e5d9007b
commit 09e5d9007b
parent db926896bd
26 changed files with 586 additions and 150 deletions
--- a/.github/workflows/image-pr.yml
+++ b/.github/workflows/image-pr.yml
@ -0,0 +1,86 @@
+---
+name: 'build container images tests'
+
+on:
+  pull_request:
+
+concurrency:
+  group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  extras-image-build:
+    uses: ./.github/workflows/image_build.yml
+    with:
+      tag-latest: ${{ matrix.tag-latest }}
+      tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
+      build-type: ${{ matrix.build-type }}
+      cuda-major-version: ${{ matrix.cuda-major-version }}
+      cuda-minor-version: ${{ matrix.cuda-minor-version }}
+      platforms: ${{ matrix.platforms }}
+      runs-on: ${{ matrix.runs-on }}
+    secrets:
+      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+    strategy:
+      # Pushing with all jobs in parallel
+      # eats the bandwidth of all the nodes
+      max-parallel: ${{ github.event_name != 'pull_request' && 2 || 4 }}
+      matrix:
+        include:
+          - build-type: ''
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+  core-image-build:
+    uses: ./.github/workflows/image_build.yml
+    with:
+      tag-latest: ${{ matrix.tag-latest }}
+      tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
+      build-type: ${{ matrix.build-type }}
+      cuda-major-version: ${{ matrix.cuda-major-version }}
+      cuda-minor-version: ${{ matrix.cuda-minor-version }}
+      platforms: ${{ matrix.platforms }}
+      runs-on: ${{ matrix.runs-on }}
+    secrets:
+      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+    strategy:
+      matrix:
+        include:
+          - build-type: ''
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
--- a/.github/workflows/image.yml
+++ b/.github/workflows/image.yml
@ -2,7 +2,6 @@
 name: 'build container images'

 on:
-  pull_request:
  push:
    branches:
      - master
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@ -34,10 +34,22 @@ jobs:
          sudo apt-get update
          sudo apt-get install build-essential ffmpeg

+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v3
+        with:
+          path: grpc
+          key: ${{ runner.os }}-grpc
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
          git clone --recurse-submodules -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
          cd grpc && mkdir -p cmake/build && cd cmake/build && cmake -DgRPC_INSTALL=ON \
            -DgRPC_BUILD_TESTS=OFF \
-                ../.. && sudo make -j12 install
+            ../.. && sudo make -j12
+      - name: Install gRPC
+        run: |
+          cd grpc && cd cmake/build && sudo make -j12 install

      - name: Build
        id: build
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@ -86,11 +86,22 @@ jobs:
          sudo cp -rfv sources/go-piper/piper-phonemize/pi/lib/. /usr/lib/ && \
          # Pre-build stable diffusion before we install a newer version of abseil (not compatible with stablediffusion-ncn)
          GO_TAGS="stablediffusion tts" GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
-
+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v3
+        with:
+          path: grpc
+          key: ${{ runner.os }}-grpc
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
          git clone --recurse-submodules -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
          cd grpc && mkdir -p cmake/build && cd cmake/build && cmake -DgRPC_INSTALL=ON \
            -DgRPC_BUILD_TESTS=OFF \
-                ../.. && sudo make -j12 install
+            ../.. && sudo make -j12
+      - name: Install gRPC
+        run: |
+          cd grpc && cd cmake/build && sudo make -j12 install
      - name: Test
        run: |
          GO_TAGS="stablediffusion tts" make test
--- a/api/api.go
+++ b/api/api.go
@ -5,7 +5,6 @@ import (
 	"errors"
 	"fmt"
 	"os"
-	"path/filepath"
 	"strings"

 	config "github.com/go-skynet/LocalAI/api/config"
@ -17,7 +16,7 @@ import (
 	"github.com/go-skynet/LocalAI/metrics"
 	"github.com/go-skynet/LocalAI/pkg/assets"
 	"github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/go-skynet/LocalAI/pkg/startup"

 	"github.com/gofiber/fiber/v2"
 	"github.com/gofiber/fiber/v2/middleware/cors"
@ -38,25 +37,7 @@ func Startup(opts ...options.AppOption) (*options.Option, *config.ConfigLoader,
 	log.Info().Msgf("Starting LocalAI using %d threads, with models path: %s", options.Threads, options.Loader.ModelPath)
 	log.Info().Msgf("LocalAI version: %s", internal.PrintableVersion())

-	modelPath := options.Loader.ModelPath
-	if len(options.ModelsURL) > 0 {
-		for _, url := range options.ModelsURL {
-			if utils.LooksLikeURL(url) {
-				// md5 of model name
-				md5Name := utils.MD5(url)
-
-				// check if file exists
-				if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
-					err := utils.DownloadFile(url, filepath.Join(modelPath, md5Name)+".yaml", "", func(fileName, current, total string, percent float64) {
-						utils.DisplayDownloadFunction(fileName, current, total, percent)
-					})
-					if err != nil {
-						log.Error().Msgf("error loading model: %s", err.Error())
-					}
-				}
-			}
-		}
-	}
+	startup.PreloadModelsConfigurations(options.Loader.ModelPath, options.ModelsURL...)

 	cl := config.NewConfigLoader()
 	if err := cl.LoadConfigs(options.Loader.ModelPath); err != nil {
--- a/api/api_test.go
+++ b/api/api_test.go
@ -16,9 +16,9 @@ import (
 	. "github.com/go-skynet/LocalAI/api"
 	"github.com/go-skynet/LocalAI/api/options"
 	"github.com/go-skynet/LocalAI/metrics"
+	"github.com/go-skynet/LocalAI/pkg/downloader"
 	"github.com/go-skynet/LocalAI/pkg/gallery"
 	"github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/utils"
 	"github.com/gofiber/fiber/v2"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
@ -61,7 +61,7 @@ func getModelStatus(url string) (response map[string]interface{}) {
 }

 func getModels(url string) (response []gallery.GalleryModel) {
-	utils.GetURI(url, func(url string, i []byte) error {
+	downloader.GetURI(url, func(url string, i []byte) error {
 		// Unmarshal YAML data into a struct
 		return json.Unmarshal(i, &response)
 	})
--- a/api/config/config.go
+++ b/api/config/config.go
@ -9,6 +9,7 @@ import (
 	"strings"
 	"sync"

+	"github.com/go-skynet/LocalAI/pkg/downloader"
 	"github.com/go-skynet/LocalAI/pkg/utils"
 	"github.com/rs/zerolog/log"
 	"gopkg.in/yaml.v3"
@ -300,21 +301,21 @@ func (cm *ConfigLoader) Preload(modelPath string) error {
 			// Create file path
 			filePath := filepath.Join(modelPath, file.Filename)

-			if err := utils.DownloadFile(file.URI, filePath, file.SHA256, status); err != nil {
+			if err := downloader.DownloadFile(file.URI, filePath, file.SHA256, status); err != nil {
 				return err
 			}
 		}

 		modelURL := config.PredictionOptions.Model
-		modelURL = utils.ConvertURL(modelURL)
+		modelURL = downloader.ConvertURL(modelURL)

-		if utils.LooksLikeURL(modelURL) {
+		if downloader.LooksLikeURL(modelURL) {
 			// md5 of model name
 			md5Name := utils.MD5(modelURL)

 			// check if file exists
 			if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
-				err := utils.DownloadFile(modelURL, filepath.Join(modelPath, md5Name), "", status)
+				err := downloader.DownloadFile(modelURL, filepath.Join(modelPath, md5Name), "", status)
 				if err != nil {
 					return err
 				}
--- a/docs/content/advanced/_index.en.md
+++ b/docs/content/advanced/_index.en.md
@ -9,7 +9,7 @@ weight = 6

 In order to define default prompts, model parameters (such as custom default `top_p` or `top_k`), LocalAI can be configured to serve user-defined models with a set of default parameters and templates.

-You can create multiple `yaml` files in the models path or either specify a single YAML configuration file. 
+In order to configure a model, you can create multiple `yaml` files in the models path or either specify a single YAML configuration file. 
 Consider the following `models` folder in the `example/chatbot-ui`:

 ```
@ -96,6 +96,12 @@ Specifying a `config-file` via CLI allows to declare models in a single file as

 See also [chatbot-ui](https://github.com/go-skynet/LocalAI/tree/master/examples/chatbot-ui) as an example on how to use config files.

+It is possible to specify a full URL or a short-hand URL to a YAML model configuration file and use it on start with local-ai, for example to use phi-2:
+
+```
+local-ai github://mudler/LocalAI/examples/configurations/phi-2.yaml@master
+```
+
 ### Full config model file reference

 ```yaml
--- a/docs/content/build/_index.en.md
+++ b/docs/content/build/_index.en.md
@ -235,6 +235,14 @@ make GRPC_BACKENDS=backend-assets/grpc/llama-cpp build

 By default, all the backends are built.

+### Specific llama.cpp version
+
+To build with a specific version of llama.cpp, set `CPPLLAMA_VERSION` to the tag or wanted sha:
+
+```
+CPPLLAMA_VERSION=<sha> make build
+```
+
 ### Windows compatibility

 Make sure to give enough resources to the running container. See https://github.com/go-skynet/LocalAI/issues/2
--- a/docs/content/features/GPU-acceleration.md
+++ b/docs/content/features/GPU-acceleration.md
@ -15,11 +15,19 @@ This section contains instruction on how to use LocalAI with GPU acceleration.
 For accelleration for AMD or Metal HW there are no specific container images, see the [build]({{%relref "build/#acceleration" %}})
 {{% /notice %}}

-### CUDA
+### CUDA(NVIDIA) acceleration

 Requirement: nvidia-container-toolkit (installation instructions [1](https://www.server-world.info/en/note?os=Ubuntu_22.04&p=nvidia&f=2) [2](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html))

-To use CUDA, use the images with the `cublas` tag.
+To check what CUDA version do you need, you can either run `nvidia-smi` or `nvcc --version`. 
+
+Alternatively, you can also check nvidia-smi with docker:
+
+```
+docker run --runtime=nvidia --rm nvidia/cuda nvidia-smi
+```
+
+To use CUDA, use the images with the `cublas` tag, for example.

 The image list is on [quay](https://quay.io/repository/go-skynet/local-ai?tab=tags):

--- a/docs/content/getting_started/_index.en.md
+++ b/docs/content/getting_started/_index.en.md
@ -14,6 +14,8 @@ See also our [How to]({{%relref "howtos" %}}) section for end-to-end guided exam

 The easiest way to run LocalAI is by using [`docker compose`](https://docs.docker.com/compose/install/) or with [Docker](https://docs.docker.com/engine/install/) (to build locally, see the [build section]({{%relref "build" %}})).

+LocalAI needs at least a model file to work, or a configuration YAML file, or both. You can customize further model defaults and specific settings with a configuration file (see [advanced]({{%relref "advanced" %}})).
+
 {{% notice note %}}
 To run with GPU Accelleration, see [GPU acceleration]({{%relref "features/gpu-acceleration" %}}).
 {{% /notice %}}
@ -113,8 +115,79 @@ helm install local-ai go-skynet/local-ai -f values.yaml

 {{% /tab %}}

+{{% tab name="From source" %}}
+
+See the [build section]({{%relref "build" %}}).
+  
+{{% /tab %}}
+
 {{< /tabs >}}

+### Running Popular models (one-click!)
+
+{{% notice note %}}
+
+Note: this feature currently is available only on master builds.
+
+{{% /notice %}}
+
+You can run `local-ai` directly with a model name, and it will download the model and start the API with the model loaded.
+
+#### CPU-only
+
+> You can use these images which are lighter and do not have Nvidia dependencies
+
+| Model | Docker command |
+| --- | --- |
+| phi2 | ```docker run -p 8080:8080 -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-ffmpeg-core phi-2``` |
+| llava | ```docker run -p 8080:8080 -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-ffmpeg-core llava``` |
+| mistral-openorca | ```docker run -p 8080:8080 -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-ffmpeg-core mistral-openorca``` |
+
+#### GPU (CUDA 11)
+
+For accellerated images with Nvidia and CUDA11, use the following images.
+
+> If you do not know which version of CUDA do you have available, you can check with `nvidia-smi` or `nvcc --version`
+
+| Model | Docker command |
+| --- | --- |
+| phi-2 | ```docker run -p 8080:8080 --gpus all -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-cublas-cuda11-core phi-2``` |
+| llava | ```docker run -p 8080:8080 -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-cublas-cuda11-core llava``` |
+| mistral-openorca | ```docker run -p 8080:8080 --gpus all -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-cublas-cuda11-core mistral-openorca``` |
+
+#### GPU (CUDA 12)
+
+> If you do not know which version of CUDA do you have available, you can check with `nvidia-smi` or `nvcc --version`
+
+| Model | Docker command |
+| --- | --- |
+| phi-2 | ```docker run -p 8080:8080 -ti --gpus all --rm quay.io/go-skynet/local-ai:{{< version >}}-cublas-cuda12-core phi-2``` |
+| llava | ```docker run -p 8080:8080 -ti --gpus all --rm quay.io/go-skynet/local-ai:{{< version >}}-cublas-cuda12-core llava``` |
+| mistral-openorca | ```docker run -p 8080:8080 --gpus all -ti --rm quay.io/go-skynet/local-ai:{{< version >}}-cublas-cuda12-core mistral-openorca``` |
+
+{{% notice note %}}
+
+LocalAI can be started (either the container image or the binary) with a list of model config files URLs or our short-handed format (e.g. `huggingface://`. `github://`). It works by passing the urls as arguments or environment variable, for example:
+
+```
+local-ai github://owner/repo/file.yaml@branch
+
+# Env
+MODELS="github://owner/repo/file.yaml@branch,github://owner/repo/file.yaml@branch" local-ai
+
+# Args
+local-ai --models github://owner/repo/file.yaml@branch --models github://owner/repo/file.yaml@branch
+```
+
+For example, to start localai with phi-2, it's possible for instance to also use a full config file from gists:
+
+```bash
+./local-ai https://gist.githubusercontent.com/mudler/ad601a0488b497b69ec549150d9edd18/raw/a8a8869ef1bb7e3830bf5c0bae29a0cce991ff8d/phi-2.yaml
+```
+
+The file should be a valid YAML configuration file, for the full syntax see [advanced]({{%relref "advanced" %}}).
+{{% /notice %}}
+
 ### Container images

 LocalAI has a set of images to support CUDA, ffmpeg and 'vanilla' (CPU-only). The image list is on [quay](https://quay.io/repository/go-skynet/local-ai?tab=tags):
@ -131,6 +204,11 @@ Core Images - Smaller images without predownload python dependencies
 {{% /tab %}}

 {{% tab name="GPU Images CUDA 11" %}}
+
+Images with Nvidia accelleration support
+
+> If you do not know which version of CUDA do you have available, you can check with `nvidia-smi` or `nvcc --version`
+
 - `master-cublas-cuda11`
 - `master-cublas-cuda11-core`
 - `{{< version >}}-cublas-cuda11`
@ -142,6 +220,11 @@ Core Images - Smaller images without predownload python dependencies
 {{% /tab %}}

 {{% tab name="GPU Images CUDA 12" %}}
+
+Images with Nvidia accelleration support
+
+> If you do not know which version of CUDA do you have available, you can check with `nvidia-smi` or `nvcc --version`
+
 - `master-cublas-cuda12`
 - `master-cublas-cuda12-core`
 - `{{< version >}}-cublas-cuda12`
@ -357,10 +440,6 @@ affinity: {}
 </details>


-### Build from source
-
-See the [build section]({{%relref "build" %}}).
-
 ### Other examples

 ![Screenshot from 2023-04-26 23-59-55](https://user-images.githubusercontent.com/2420543/234715439-98d12e03-d3ce-4f94-ab54-2b256808e05e.png)
--- a/docs/content/model-compatibility/diffusers.md
+++ b/docs/content/model-compatibility/diffusers.md
@ -167,11 +167,6 @@ curl -H "Content-Type: application/json" -d @-  http://localhost:8080/v1/images/

 ## img2vid

-{{% notice note %}}
-
-Experimental and available only on master builds. See: https://github.com/mudler/LocalAI/pull/1442
-
-{{% /notice %}}

 ```yaml
 name: img2vid
@ -193,12 +188,6 @@ curl -H "Content-Type: application/json" -X POST -d @- http://localhost:8080/v1/

 ## txt2vid

-{{% notice note %}}
-
-Experimental and available only on master builds. See: https://github.com/mudler/LocalAI/pull/1442
-
-{{% /notice %}}
-
 ```yaml
 name: txt2vid
 parameters:
--- a/embedded/embedded.go
+++ b/embedded/embedded.go
@ -0,0 +1,53 @@
+package embedded
+
+import (
+	"embed"
+	"fmt"
+	"slices"
+	"strings"
+
+	"github.com/go-skynet/LocalAI/pkg/assets"
+	"gopkg.in/yaml.v3"
+)
+
+var modelShorteners map[string]string
+
+//go:embed model_library.yaml
+var modelLibrary []byte
+
+//go:embed models/*
+var embeddedModels embed.FS
+
+func ModelShortURL(s string) string {
+	if _, ok := modelShorteners[s]; ok {
+		s = modelShorteners[s]
+	}
+
+	return s
+}
+
+func init() {
+	yaml.Unmarshal(modelLibrary, &modelShorteners)
+}
+
+// ExistsInModelsLibrary checks if a model exists in the embedded models library
+func ExistsInModelsLibrary(s string) bool {
+	f := fmt.Sprintf("%s.yaml", s)
+
+	a := []string{}
+
+	for _, j := range assets.ListFiles(embeddedModels) {
+		a = append(a, strings.TrimPrefix(j, "models/"))
+	}
+
+	return slices.Contains(a, f)
+}
+
+// ResolveContent returns the content in the embedded model library
+func ResolveContent(s string) ([]byte, error) {
+	if ExistsInModelsLibrary(s) {
+		return embeddedModels.ReadFile(fmt.Sprintf("models/%s.yaml", s))
+	}
+
+	return nil, fmt.Errorf("cannot find model %s", s)
+}
--- a/embedded/model_library.yaml
+++ b/embedded/model_library.yaml
@ -0,0 +1,9 @@
+### 
+###
+### This file contains the list of models that are available in the library
+### The URLs are automatically expanded when local-ai is being called with the key as argument
+###
+### For models with an entire YAML file to be embededd, put the file inside the `models`
+### directory, it will be automatically available with the file name as key (without the .yaml extension)
+
+phi-2:  "github://mudler/LocalAI/examples/configurations/phi-2.yaml@master"
--- a/embedded/models/llava.yaml
+++ b/embedded/models/llava.yaml
@ -0,0 +1,31 @@
+backend: llama-cpp
+context_size: 4096
+f16: true
+
+gpu_layers: 90
+mmap: true
+name: llava
+
+roles:
+  user: "USER:"
+  assistant: "ASSISTANT:"
+  system: "SYSTEM:"
+
+mmproj: bakllava-mmproj.gguf
+parameters:
+  model: bakllava.gguf
+  temperature: 0.2
+  top_k: 40
+  top_p: 0.95
+
+template:
+  chat: |
+    A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.
+    {{.Input}}
+    ASSISTANT:
+
+download_files:
+- filename: bakllava.gguf
+  uri: huggingface://mys/ggml_bakllava-1/ggml-model-q4_k.gguf
+- filename: bakllava-mmproj.gguf
+  uri: huggingface://mys/ggml_bakllava-1/mmproj-model-f16.gguf
--- a/embedded/models/mistral-openorca.yaml
+++ b/embedded/models/mistral-openorca.yaml
@ -0,0 +1,23 @@
+name: mistral-openorca
+mmap: true
+parameters:
+  model: huggingface://TheBloke/Mistral-7B-OpenOrca-GGUF/mistral-7b-openorca.Q6_K.gguf
+  temperature: 0.2
+  top_k: 40
+  top_p: 0.95
+template:
+  chat_message: |
+    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "user"}}user{{end}}
+    {{if .Content}}{{.Content}}{{end}}
+    <|im_end|>
+    
+  chat: |
+    {{.Input}}
+    <|im_start|>assistant
+    
+  completion: |
+    {{.Input}}
+context_size: 4096
+f16: true
+stopwords:
+- <|im_end|>
--- a/pkg/assets/list.go
+++ b/pkg/assets/list.go
@ -0,0 +1,22 @@
+package assets
+
+import (
+	"embed"
+	"io/fs"
+)
+
+func ListFiles(content embed.FS) (files []string) {
+	fs.WalkDir(content, ".", func(path string, d fs.DirEntry, err error) error {
+		if err != nil {
+			return err
+		}
+
+		if d.IsDir() {
+			return nil
+		}
+
+		files = append(files, path)
+		return nil
+	})
+	return
+}
--- a/pkg/downloader/progress.go
+++ b/pkg/downloader/progress.go
@ -0,0 +1,26 @@
+package downloader
+
+import "hash"
+
+type progressWriter struct {
+	fileName       string
+	total          int64
+	written        int64
+	downloadStatus func(string, string, string, float64)
+	hash           hash.Hash
+}
+
+func (pw *progressWriter) Write(p []byte) (n int, err error) {
+	n, err = pw.hash.Write(p)
+	pw.written += int64(n)
+
+	if pw.total > 0 {
+		percentage := float64(pw.written) / float64(pw.total) * 100
+		//log.Debug().Msgf("Downloading %s: %s/%s (%.2f%%)", pw.fileName, formatBytes(pw.written), formatBytes(pw.total), percentage)
+		pw.downloadStatus(pw.fileName, formatBytes(pw.written), formatBytes(pw.total), percentage)
+	} else {
+		pw.downloadStatus(pw.fileName, formatBytes(pw.written), "", 0)
+	}
+
+	return
+}
--- a/pkg/downloader/uri.go
+++ b/pkg/downloader/uri.go
@ -1,10 +1,9 @@
-package utils
+package downloader

 import (
-	"crypto/md5"
 	"crypto/sha256"
+	"encoding/base64"
 	"fmt"
-	"hash"
 	"io"
 	"net/http"
 	"os"
@ -12,9 +11,18 @@ import (
 	"strconv"
 	"strings"

+	"github.com/go-skynet/LocalAI/pkg/utils"
 	"github.com/rs/zerolog/log"
 )

+const (
+	HuggingFacePrefix = "huggingface://"
+	HTTPPrefix        = "http://"
+	HTTPSPrefix       = "https://"
+	GithubURI         = "github:"
+	GithubURI2        = "github://"
+)
+
 func GetURI(url string, f func(url string, i []byte) error) error {
 	url = ConvertURL(url)

@ -52,14 +60,6 @@ func GetURI(url string, f func(url string, i []byte) error) error {
 	return f(url, body)
 }

-const (
-	HuggingFacePrefix = "huggingface://"
-	HTTPPrefix        = "http://"
-	HTTPSPrefix       = "https://"
-	GithubURI         = "github:"
-	GithubURI2        = "github://"
-)
-
 func LooksLikeURL(s string) bool {
 	return strings.HasPrefix(s, HTTPPrefix) ||
 		strings.HasPrefix(s, HTTPSPrefix) ||
@ -229,10 +229,10 @@ func DownloadFile(url string, filePath, sha string, downloadStatus func(string,
 	}

 	log.Info().Msgf("File %q downloaded and verified", filePath)
-	if IsArchive(filePath) {
+	if utils.IsArchive(filePath) {
 		basePath := filepath.Dir(filePath)
 		log.Info().Msgf("File %q is an archive, uncompressing to %s", filePath, basePath)
-		if err := ExtractArchive(filePath, basePath); err != nil {
+		if err := utils.ExtractArchive(filePath, basePath); err != nil {
 			log.Debug().Msgf("Failed decompressing %q: %s", filePath, err.Error())
 			return err
 		}
@ -241,32 +241,35 @@ func DownloadFile(url string, filePath, sha string, downloadStatus func(string,
 	return nil
 }

-type progressWriter struct {
-	fileName       string
-	total          int64
-	written        int64
-	downloadStatus func(string, string, string, float64)
-	hash           hash.Hash
-}
+// this function check if the string is an URL, if it's an URL downloads the image in memory
+// encodes it in base64 and returns the base64 string
+func GetBase64Image(s string) (string, error) {
+	if strings.HasPrefix(s, "http") {
+		// download the image
+		resp, err := http.Get(s)
+		if err != nil {
+			return "", err
+		}
+		defer resp.Body.Close()

-func (pw *progressWriter) Write(p []byte) (n int, err error) {
-	n, err = pw.hash.Write(p)
-	pw.written += int64(n)
-
-	if pw.total > 0 {
-		percentage := float64(pw.written) / float64(pw.total) * 100
-		//log.Debug().Msgf("Downloading %s: %s/%s (%.2f%%)", pw.fileName, formatBytes(pw.written), formatBytes(pw.total), percentage)
-		pw.downloadStatus(pw.fileName, formatBytes(pw.written), formatBytes(pw.total), percentage)
-	} else {
-		pw.downloadStatus(pw.fileName, formatBytes(pw.written), "", 0)
+		// read the image data into memory
+		data, err := io.ReadAll(resp.Body)
+		if err != nil {
+			return "", err
 		}

-	return
-}
+		// encode the image data in base64
+		encoded := base64.StdEncoding.EncodeToString(data)

-// MD5 of a string
-func MD5(s string) string {
-	return fmt.Sprintf("%x", md5.Sum([]byte(s)))
+		// return the base64 string
+		return encoded, nil
+	}
+
+	// if the string instead is prefixed with "data:image/jpeg;base64,", drop it
+	if strings.HasPrefix(s, "data:image/jpeg;base64,") {
+		return strings.ReplaceAll(s, "data:image/jpeg;base64,", ""), nil
+	}
+	return "", fmt.Errorf("not valid string")
 }

 func formatBytes(bytes int64) string {
--- a/pkg/downloader/uri_test.go
+++ b/pkg/downloader/uri_test.go
@ -1,7 +1,7 @@
-package utils_test
+package downloader_test

 import (
-	. "github.com/go-skynet/LocalAI/pkg/utils"
+	. "github.com/go-skynet/LocalAI/pkg/downloader"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 )
--- a/pkg/gallery/gallery.go
+++ b/pkg/gallery/gallery.go
@ -6,7 +6,7 @@ import (
 	"path/filepath"
 	"strings"

-	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/go-skynet/LocalAI/pkg/downloader"
 	"github.com/imdario/mergo"
 	"github.com/rs/zerolog/log"
 	"gopkg.in/yaml.v2"
@ -140,7 +140,7 @@ func AvailableGalleryModels(galleries []Gallery, basePath string) ([]*GalleryMod

 func findGalleryURLFromReferenceURL(url string) (string, error) {
 	var refFile string
-	err := utils.GetURI(url, func(url string, d []byte) error {
+	err := downloader.GetURI(url, func(url string, d []byte) error {
 		refFile = string(d)
 		if len(refFile) == 0 {
 			return fmt.Errorf("invalid reference file at url %s: %s", url, d)
@ -163,7 +163,7 @@ func getGalleryModels(gallery Gallery, basePath string) ([]*GalleryModel, error)
 		}
 	}

-	err := utils.GetURI(gallery.URL, func(url string, d []byte) error {
+	err := downloader.GetURI(gallery.URL, func(url string, d []byte) error {
 		return yaml.Unmarshal(d, &models)
 	})
 	if err != nil {
--- a/pkg/gallery/models.go
+++ b/pkg/gallery/models.go
@ -1,14 +1,11 @@
 package gallery

 import (
-	"crypto/sha256"
 	"fmt"
-	"hash"
-	"io"
 	"os"
 	"path/filepath"
-	"strconv"

+	"github.com/go-skynet/LocalAI/pkg/downloader"
 	"github.com/go-skynet/LocalAI/pkg/utils"
 	"github.com/imdario/mergo"
 	"github.com/rs/zerolog/log"
@ -66,7 +63,7 @@ type PromptTemplate struct {

 func GetGalleryConfigFromURL(url string) (Config, error) {
 	var config Config
-	err := utils.GetURI(url, func(url string, d []byte) error {
+	err := downloader.GetURI(url, func(url string, d []byte) error {
 		return yaml.Unmarshal(d, &config)
 	})
 	if err != nil {
@ -114,7 +111,7 @@ func InstallModel(basePath, nameOverride string, config *Config, configOverrides
 		// Create file path
 		filePath := filepath.Join(basePath, file.Filename)

-		if err := utils.DownloadFile(file.URI, filePath, file.SHA256, downloadStatus); err != nil {
+		if err := downloader.DownloadFile(file.URI, filePath, file.SHA256, downloadStatus); err != nil {
 			return err
 		}
 	}
@ -183,54 +180,3 @@ func InstallModel(basePath, nameOverride string, config *Config, configOverrides

 	return nil
 }
-
-type progressWriter struct {
-	fileName       string
-	total          int64
-	written        int64
-	downloadStatus func(string, string, string, float64)
-	hash           hash.Hash
-}
-
-func (pw *progressWriter) Write(p []byte) (n int, err error) {
-	n, err = pw.hash.Write(p)
-	pw.written += int64(n)
-
-	if pw.total > 0 {
-		percentage := float64(pw.written) / float64(pw.total) * 100
-		//log.Debug().Msgf("Downloading %s: %s/%s (%.2f%%)", pw.fileName, formatBytes(pw.written), formatBytes(pw.total), percentage)
-		pw.downloadStatus(pw.fileName, formatBytes(pw.written), formatBytes(pw.total), percentage)
-	} else {
-		pw.downloadStatus(pw.fileName, formatBytes(pw.written), "", 0)
-	}
-
-	return
-}
-
-func formatBytes(bytes int64) string {
-	const unit = 1024
-	if bytes < unit {
-		return strconv.FormatInt(bytes, 10) + " B"
-	}
-	div, exp := int64(unit), 0
-	for n := bytes / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-	return fmt.Sprintf("%.1f %ciB", float64(bytes)/float64(div), "KMGTPE"[exp])
-}
-
-func calculateSHA(filePath string) (string, error) {
-	file, err := os.Open(filePath)
-	if err != nil {
-		return "", err
-	}
-	defer file.Close()
-
-	hash := sha256.New()
-	if _, err := io.Copy(hash, file); err != nil {
-		return "", err
-	}
-
-	return fmt.Sprintf("%x", hash.Sum(nil)), nil
-}
--- a/pkg/startup/model_preload.go
+++ b/pkg/startup/model_preload.go
@ -0,0 +1,54 @@
+package startup
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+
+	"github.com/go-skynet/LocalAI/embedded"
+	"github.com/go-skynet/LocalAI/pkg/downloader"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/rs/zerolog/log"
+)
+
+// PreloadModelsConfigurations will preload models from the given list of URLs
+// It will download the model if it is not already present in the model path
+// It will also try to resolve if the model is an embedded model YAML configuration
+func PreloadModelsConfigurations(modelPath string, models ...string) {
+	for _, url := range models {
+		url = embedded.ModelShortURL(url)
+
+		switch {
+		case embedded.ExistsInModelsLibrary(url):
+			modelYAML, err := embedded.ResolveContent(url)
+			// If we resolve something, just save it to disk and continue
+			if err != nil {
+				log.Error().Msgf("error loading model: %s", err.Error())
+				continue
+			}
+
+			log.Debug().Msgf("[startup] resolved embedded model: %s", url)
+			md5Name := utils.MD5(url)
+			if err := os.WriteFile(filepath.Join(modelPath, md5Name)+".yaml", modelYAML, os.ModePerm); err != nil {
+				log.Error().Msgf("error loading model: %s", err.Error())
+			}
+		case downloader.LooksLikeURL(url):
+			log.Debug().Msgf("[startup] resolved model to download: %s", url)
+
+			// md5 of model name
+			md5Name := utils.MD5(url)
+
+			// check if file exists
+			if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
+				err := downloader.DownloadFile(url, filepath.Join(modelPath, md5Name)+".yaml", "", func(fileName, current, total string, percent float64) {
+					utils.DisplayDownloadFunction(fileName, current, total, percent)
+				})
+				if err != nil {
+					log.Error().Msgf("error loading model: %s", err.Error())
+				}
+			}
+		default:
+			log.Warn().Msgf("[startup] failed resolving model '%s'", url)
+		}
+	}
+}
--- a/pkg/startup/model_preload_test.go
+++ b/pkg/startup/model_preload_test.go
@ -0,0 +1,66 @@
+package startup_test
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+
+	. "github.com/go-skynet/LocalAI/pkg/startup"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+var _ = Describe("Preload test", func() {
+
+	Context("Preloading from strings", func() {
+		It("loads from embedded full-urls", func() {
+			tmpdir, err := os.MkdirTemp("", "")
+			Expect(err).ToNot(HaveOccurred())
+			url := "https://raw.githubusercontent.com/mudler/LocalAI/master/examples/configurations/phi-2.yaml"
+			fileName := fmt.Sprintf("%s.yaml", utils.MD5(url))
+
+			PreloadModelsConfigurations(tmpdir, url)
+
+			resultFile := filepath.Join(tmpdir, fileName)
+
+			content, err := os.ReadFile(resultFile)
+			Expect(err).ToNot(HaveOccurred())
+
+			Expect(string(content)).To(ContainSubstring("name: phi-2"))
+		})
+		It("loads from embedded short-urls", func() {
+			tmpdir, err := os.MkdirTemp("", "")
+			Expect(err).ToNot(HaveOccurred())
+			url := "phi-2"
+
+			PreloadModelsConfigurations(tmpdir, url)
+
+			entry, err := os.ReadDir(tmpdir)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(entry).To(HaveLen(1))
+			resultFile := entry[0].Name()
+
+			content, err := os.ReadFile(filepath.Join(tmpdir, resultFile))
+			Expect(err).ToNot(HaveOccurred())
+
+			Expect(string(content)).To(ContainSubstring("name: phi-2"))
+		})
+		It("loads from embedded models", func() {
+			tmpdir, err := os.MkdirTemp("", "")
+			Expect(err).ToNot(HaveOccurred())
+			url := "mistral-openorca"
+			fileName := fmt.Sprintf("%s.yaml", utils.MD5(url))
+
+			PreloadModelsConfigurations(tmpdir, url)
+
+			resultFile := filepath.Join(tmpdir, fileName)
+
+			content, err := os.ReadFile(resultFile)
+			Expect(err).ToNot(HaveOccurred())
+
+			Expect(string(content)).To(ContainSubstring("name: mistral-openorca"))
+		})
+	})
+})
--- a/pkg/startup/startup_suite_test.go
+++ b/pkg/startup/startup_suite_test.go
@ -0,0 +1,13 @@
+package startup_test
+
+import (
+	"testing"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func TestStartup(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "LocalAI startup test")
+}
--- a/pkg/utils/hash.go
+++ b/pkg/utils/hash.go
@ -0,0 +1,10 @@
+package utils
+
+import (
+	"crypto/md5"
+	"fmt"
+)
+
+func MD5(s string) string {
+	return fmt.Sprintf("%x", md5.Sum([]byte(s)))
+}