chore: ⬆️ Update ggml-org/llama.cpp to 4663bd353c61c1136cd8a97b9908755e4ab30cec (#5100)

⬆️ Update ggml-org/llama.cpp

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>

.bruno/LocalAI Test Requests/Sound Generation/musicgen.bru

@@ -1,23 +0,0 @@
-meta {
-  name: musicgen
-  type: http
-  seq: 1
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/sound-generation
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model_id": "facebook/musicgen-small",
-      "text": "Exciting 80s Newscast Interstitial",
-      "duration_seconds": 8
-  }
-}

.bruno/LocalAI Test Requests/backend monitor/backend monitor.bru

@@ -1,17 +0,0 @@
-meta {
-  name: backend monitor
-  type: http
-  seq: 4
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/backend/monitor
-  body: json
-  auth: none
-}
-
-body:json {
-  {
-    "model": "{{DEFAULT_MODEL}}"
-  }
-}

.bruno/LocalAI Test Requests/backend monitor/backend-shutdown.bru

@@ -1,21 +0,0 @@
-meta {
-  name: backend-shutdown
-  type: http
-  seq: 3
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/backend/shutdown
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}"
-  }
-}

.bruno/LocalAI Test Requests/bruno.json

@@ -1,5 +0,0 @@
-{
-  "version": "1",
-  "name": "LocalAI Test Requests",
-  "type": "collection"
-}

.bruno/LocalAI Test Requests/environments/localhost.bru

@@ -1,6 +0,0 @@
-vars {
-  HOST: localhost
-  PORT: 8080
-  DEFAULT_MODEL: gpt-3.5-turbo
-  PROTOCOL: http://
-}

.bruno/LocalAI Test Requests/get models list.bru

@@ -1,11 +0,0 @@
-meta {
-  name: get models list
-  type: http
-  seq: 2
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models
-  body: none
-  auth: none
-}

.bruno/LocalAI Test Requests/image generation/Generate image.bru

@@ -1,25 +0,0 @@
-meta {
-  name: Generate image
-  type: http
-  seq: 1
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/images/generations
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-    "prompt": "<positive prompt>|<negative prompt>",
-    "model": "model-name",
-    "step": 51,
-    "size": "1024x1024",
-    "image": ""
-  }
-}

.bruno/LocalAI Test Requests/llm text/-completions.bru

@@ -1,24 +0,0 @@
-meta {
-  name: -completions
-  type: http
-  seq: 4
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "prompt": "function downloadFile(string url, string outputPath) {",
-      "max_tokens": 256,
-      "temperature": 0.5
-  }
-}

.bruno/LocalAI Test Requests/llm text/-edits.bru

@@ -1,23 +0,0 @@
-meta {
-  name: -edits
-  type: http
-  seq: 5
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/edits
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "input": "What day of the wek is it?",
-      "instruction": "Fix the spelling mistakes"
-  }
-}

.bruno/LocalAI Test Requests/llm text/-embeddings.bru

@@ -1,22 +0,0 @@
-meta {
-  name: -embeddings
-  type: http
-  seq: 6
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/embeddings
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "input": "A STRANGE GAME.\nTHE ONLY WINNING MOVE IS NOT TO PLAY.\n\nHOW ABOUT A NICE GAME OF CHESS?"
-  }
-}

.bruno/LocalAI Test Requests/llm text/chat/chat completion -simple- 1 message-.bru

@@ -1,30 +0,0 @@
-meta {
-  name: chat completion -simple- 1 message-
-  type: http
-  seq: 4
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/chat/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-    "model": "{{DEFAULT_MODEL}}",
-    "messages": [
-      {
-        "role": "user",
-        "content": "How could one use friction to cook an egg?"
-      }
-    ],
-    "max_tokens": 256,
-    "temperature": 0.2,
-    "grammar": ""
-  }
-}

.bruno/LocalAI Test Requests/llm text/chat/chat-completions -long-.bru

@@ -1,29 +0,0 @@
-meta {
-  name: chat-completions -long-
-  type: http
-  seq: 5
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/chat/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-       "model": "{{DEFAULT_MODEL}}",
-       "messages": [{"role": "system", "content": "You are a helpful, intelligent pirate assistant. Always answer as helpfully as possible, while being a pirate and using pirate language. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information."},
-           {"role": "user", "content": "How could one use electricity to cook an egg?"},
-           {"role": "assistant",
-                  "content": "Shiver me timbers! Using electricity to cook an egg? Well, matey, I reckon that be a right curious idea! *adjusts spectacles* Now, I ain't sure if this be possible, but I'll do me best to help ye out.\n\nFirst things first, ye gotta understand that electricity be a powerful force, and it can't just cook an egg on its own. Ye see, electricity be like a mighty wind that can make things happen, but it needs somethin' to work with. So, if ye want to use electricity to cook an egg, ye gotta find a way to harness that power and make it do the cookin'.\n\nNow, I know what ye might be thinkin': \"How do I harness the power of electricity to cook an egg?\" Well, matey, there be a few ways to do it. One way be to use a special device called an \"electric frying pan.\" This be a pan that has a built-in heating element that gets hot when ye plug it into a wall socket. When the element gets hot, ye can crack an egg into the pan and watch as it cook"
-              },
-              {"role": "user", "content": "I don't have one of those, just a raw wire and plenty of power! How do we get it done?"}],
-       "max_tokens": 1024,
-       "temperature": 0.5
-  }
-}

.bruno/LocalAI Test Requests/llm text/chat/chat-completions -stream-.bru

@@ -1,25 +0,0 @@
-meta {
-  name: chat-completions -stream-
-  type: http
-  seq: 6
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/chat/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-       "model": "{{DEFAULT_MODEL}}",
-       "messages": [{"role": "user", "content": "Explain how I can set sail on the ocean using only power generated by seagulls?"}],
-       "max_tokens": 256,
-       "temperature": 0.9,
-       "stream": true
-  }
-}

.bruno/LocalAI Test Requests/model gallery/add model gallery.bru

@@ -1,22 +0,0 @@
-meta {
-  name: add model gallery
-  type: http
-  seq: 10
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "url": "file:///home/dave/projects/model-gallery/huggingface/TheBloke__CodeLlama-7B-Instruct-GGML.yaml",
-      "name": "test"
-  }
-}

.bruno/LocalAI Test Requests/model gallery/delete model gallery.bru

@@ -1,21 +0,0 @@
-meta {
-  name: delete model gallery
-  type: http
-  seq: 11
-}
-
-delete {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "name": "test"
-  }
-}

.bruno/LocalAI Test Requests/model gallery/list MODELS in galleries.bru

@@ -1,11 +0,0 @@
-meta {
-  name: list MODELS in galleries
-  type: http
-  seq: 7
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/available
-  body: none
-  auth: none
-}

.bruno/LocalAI Test Requests/model gallery/list model GALLERIES.bru

@@ -1,11 +0,0 @@
-meta {
-  name: list model GALLERIES
-  type: http
-  seq: 8
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: none
-  auth: none
-}

.bruno/LocalAI Test Requests/model gallery/model delete.bru

@@ -1,11 +0,0 @@
-meta {
-  name: model delete
-  type: http
-  seq: 7
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: none
-  auth: none
-}

.bruno/LocalAI Test Requests/model gallery/model gallery apply -gist-.bru

@@ -1,21 +0,0 @@
-meta {
-  name: model gallery apply -gist-
-  type: http
-  seq: 12
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/apply
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "id": "TheBloke__CodeLlama-7B-Instruct-GGML__codellama-7b-instruct.ggmlv3.Q2_K.bin"
-  }
-}

.bruno/LocalAI Test Requests/model gallery/model gallery apply.bru

@@ -1,22 +0,0 @@
-meta {
-  name: model gallery apply
-  type: http
-  seq: 9
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/apply
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "id": "dave@TheBloke__CodeLlama-7B-Instruct-GGML__codellama-7b-instruct.ggmlv3.Q3_K_S.bin",
-      "name": "codellama7b"
-  }
-}

.bruno/LocalAI Test Requests/transcription/transcribe.bru

@@ -1,16 +0,0 @@
-meta {
-  name: transcribe
-  type: http
-  seq: 1
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/audio/transcriptions
-  body: multipartForm
-  auth: none
-}
-
-body:multipart-form {
-  file: @file(transcription/gb1.ogg)
-  model: whisper-1
-}

.bruno/LocalAI Test Requests/tts/-tts.bru

@@ -1,22 +0,0 @@
-meta {
-  name: -tts
-  type: http
-  seq: 2
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/tts
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "input": "A STRANGE GAME.\nTHE ONLY WINNING MOVE IS NOT TO PLAY.\n\nHOW ABOUT A NICE GAME OF CHESS?"
-  }
-}

.bruno/LocalAI Test Requests/tts/musicgen.bru

@@ -1,23 +0,0 @@
-meta {
-  name: musicgen
-  type: http
-  seq: 2
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/tts
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "backend": "transformers",
-      "model": "facebook/musicgen-small",
-      "input": "80s Synths playing Jazz"
-  }
-}

.github/labeler.yml

@@ -1,4 +1,4 @@
-enhancements:
+enhancement:
  - head-branch: ['^feature', 'feature']
 
 dependencies:

.github/workflows/bump_deps.yaml

@@ -9,7 +9,7 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - repository: "ggerganov/llama.cpp"
+          - repository: "ggml-org/llama.cpp"
             variable: "CPPLLAMA_VERSION"
             branch: "master"
           - repository: "ggerganov/whisper.cpp"

.github/workflows/deploy-explorer.yaml

@@ -33,7 +33,7 @@ jobs:
         run: |
           CGO_ENABLED=0 make build-api
       - name: rm
-        uses: appleboy/ssh-action@v1.2.0
+        uses: appleboy/ssh-action@v1.2.2
         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@v1.2.0
+        uses: appleboy/ssh-action@v1.2.2
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}

.github/workflows/generate_grpc_cache.yaml

@@ -2,9 +2,10 @@ name: 'generate and publish GRPC docker caches'
 
 on:
   workflow_dispatch:
-  push:
-    branches:
-      - master
+
+  schedule:
+    # daily at midnight
+    - cron: '0 0 * * *'
 
 concurrency:
   group: grpc-cache-${{ github.head_ref || github.ref }}-${{ github.repository }}
@@ -16,7 +17,7 @@ jobs:
       matrix:
         include:
           - grpc-base-image: ubuntu:22.04
-            runs-on: 'ubuntu-latest'
+            runs-on: 'arc-runner-set'
             platforms: 'linux/amd64,linux/arm64'
     runs-on: ${{matrix.runs-on}}
     steps:

.github/workflows/image_build.yml

@@ -310,6 +310,11 @@ jobs:
           tags: ${{ steps.meta_aio_dockerhub.outputs.tags }}
           labels: ${{ steps.meta_aio_dockerhub.outputs.labels }}
 
+      - name: Cleanup
+        run: |
+          docker builder prune -f
+          docker system prune --force --volumes --all
+
       - name: Latest tag
         # run this on branches, when it is a tag and there is a latest-image defined
         if: github.event_name != 'pull_request' && inputs.latest-image != ''  && github.ref_type == 'tag'

Dockerfile

@@ -24,6 +24,7 @@ RUN apt-get update && \
         ca-certificates \
         curl libssl-dev \
         git \
+        git-lfs \
         unzip upx-ucl && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023-2024 Ettore Di Giacinto (mudler@localai.io)
+Copyright (c) 2023-2025 Ettore Di Giacinto (mudler@localai.io)
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

Makefile

@@ -6,7 +6,7 @@ BINARY_NAME=local-ai
 DETECT_LIBS?=true
 
 # llama.cpp versions
-CPPLLAMA_VERSION?=d2fe216fb2fb7ca8627618c9ea3a2e7886325780
+CPPLLAMA_VERSION?=4663bd353c61c1136cd8a97b9908755e4ab30cec
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
@@ -22,7 +22,7 @@ BARKCPP_VERSION?=v1.0.0
 
 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=d46ed5e184b97c2018dc2e8105925bdb8775e02c
+STABLEDIFFUSION_GGML_VERSION?=19d876ee300a055629926ff836489901f734f2b7
 
 ONNX_VERSION?=1.20.0
 ONNX_ARCH?=x64

README.md

@@ -212,7 +212,7 @@ A huge thank you to our generous sponsors who support this project covering CI e
 
 <p align="center">
   <a href="https://www.spectrocloud.com/" target="blank">
-    <img height="200" src="https://github.com/go-skynet/LocalAI/assets/2420543/68a6f3cb-8a65-4a4d-99b5-6417a8905512">
+    <img height="200" src="https://github.com/user-attachments/assets/72eab1dd-8b93-4fc0-9ade-84db49f24962">
   </a>
   <a href="https://www.premai.io/" target="blank">
     <img height="200" src="https://github.com/mudler/LocalAI/assets/2420543/42e4ca83-661e-4f79-8e46-ae43689683d6"> <br>

aio/cpu/embeddings.yaml

@@ -1,7 +1,7 @@
-name: text-embedding-ada-002
 embeddings: true
+name: text-embedding-ada-002
 parameters:
-  model: huggingface://hugging-quants/Llama-3.2-1B-Instruct-Q4_K_M-GGUF/llama-3.2-1b-instruct-q4_k_m.gguf
+  model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 
 usage: |
     You can test this model with curl like this:

aio/cpu/text-to-text.yaml

@@ -1,101 +1,57 @@
-name: gpt-4
-mmap: true
-parameters:
-  model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
 context_size: 8192
-
-stopwords:
-- "<|im_end|>"
-- "<dummy32000>"
-- "</tool_call>"
-- "<|eot_id|>"
-- "<|end_of_text|>"
-
+f16: true
 function:
-  # disable injecting the "answer" tool
-  disable_no_action: true
-
   grammar:
-    # This allows the grammar to also return messages
-    mixed_mode: true
-    # Suffix to add to the grammar
-    #prefix: '<tool_call>\n'
-    # Force parallel calls in the grammar
-    # parallel_calls: true
-
-  return_name_in_function_response: true
-  # Without grammar uncomment the lines below
-  # Warning: this is relying only on the capability of the
-  # LLM model to generate the correct function call.
-  json_regex_match: 
-   - "(?s)<tool_call>(.*?)</tool_call>"
-   - "(?s)<tool_call>(.*?)"
-  replace_llm_results:
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-  replace_function_results: 
-  # Replace everything that is not JSON array or object
-  # 
-  - key: '(?s)^[^{\[]*'
-    value: ""
-  - key: '(?s)[^}\]]*$'
-    value: ""
-  - key: "'([^']*?)'"
-    value: "_DQUOTE_${1}_DQUOTE_"
-  - key: '\\"'
-    value: "__TEMP_QUOTE__"
-  - key: "\'"
-    value: "'"
-  - key: "_DQUOTE_"
-    value: '"'
-  - key: "__TEMP_QUOTE__"
-    value: '"'
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-
+    no_mixed_free_string: true
+    schema_type: llama3.1 # or JSON is supported too (json)
+  response_regex:
+  - <function=(?P<name>\w+)>(?P<arguments>.*)</function>
+mmap: true
+name: gpt-4
+parameters:
+  model: Hermes-3-Llama-3.2-3B-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- <|eot_id|>
+- <|end_of_text|>
 template:
   chat: |
-    {{.Input -}}
-    <|im_start|>assistant
+    <|begin_of_text|><|start_header_id|>system<|end_header_id|>
+    You are a helpful assistant<|eot_id|><|start_header_id|>user<|end_header_id|>
+    {{.Input }}
+    <|start_header_id|>assistant<|end_header_id|>
   chat_message: |
-    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
-    {{- if .FunctionCall }}
-    <tool_call>
-    {{- else if eq .RoleName "tool" }}
-    <tool_response>
-    {{- end }}
-    {{- if .Content}}
-    {{.Content }}
-    {{- end }}
-    {{- if .FunctionCall}}
-    {{toJson .FunctionCall}}
-    {{- end }}
-    {{- if .FunctionCall }}
-    </tool_call>
-    {{- else if eq .RoleName "tool" }}
-    </tool_response>
-    {{- end }}<|im_end|>
+    <|start_header_id|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}<|end_header_id|>
+    {{ if .FunctionCall -}}
+    {{ else if eq .RoleName "tool" -}}
+    The Function was executed and the response was:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content -}}
+    {{ else if .FunctionCall -}}
+    {{ range .FunctionCall }}
+    [{{.FunctionCall.Name}}({{.FunctionCall.Arguments}})]
+    {{ end }}
+    {{ end -}}
+    <|eot_id|>
   completion: |
     {{.Input}}
-  function: |-
-    <|im_start|>system
-    You are a function calling AI model.
-    Here are the available tools:
-    <tools>
-    {{range .Functions}}
-    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
-    {{end}}
-    </tools>
-    You should call the tools provided to you sequentially
-    Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
-    <scratchpad>
-    {step-by-step reasoning and plan in bullet points}
-    </scratchpad>
-    For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
-    <tool_call>
-    {"arguments": <args-dict>, "name": <function-name>}
-    </tool_call><|im_end|>
-    {{.Input -}}
-    <|im_start|>assistant
+  function: |
+    <|start_header_id|>system<|end_header_id|>
+    You are an expert in composing functions. You are given a question and a set of possible functions.
+    Based on the question, you will need to make one or more function/tool calls to achieve the purpose.
+    If none of the functions can be used, point it out. If the given question lacks the parameters required by the function, also point it out. You should only return the function call in tools call sections.
+    If you decide to invoke any of the function(s), you MUST put it in the format as follows:
+    [func_name1(params_name1=params_value1,params_name2=params_value2,...),func_name2(params_name1=params_value1,params_name2=params_value2,...)]
+    You SHOULD NOT include any other text in the response.
+    Here is a list of functions in JSON format that you can invoke.
+    {{toJson .Functions}}
+    <|eot_id|><|start_header_id|>user<|end_header_id|>
+    {{.Input}}
+    <|eot_id|><|start_header_id|>assistant<|end_header_id|>
+
+download_files:
+- filename: Hermes-3-Llama-3.2-3B-Q4_K_M.gguf
+  sha256: 2e220a14ba4328fee38cf36c2c068261560f999fadb5725ce5c6d977cb5126b5
+  uri: huggingface://bartowski/Hermes-3-Llama-3.2-3B-GGUF/Hermes-3-Llama-3.2-3B-Q4_K_M.gguf

aio/cpu/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/cpu/vision.yaml

@@ -1,31 +1,49 @@
-backend: llama-cpp
 context_size: 4096
 f16: true
 mmap: true
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
-
-roles:
-  user: "USER:"
-  assistant: "ASSISTANT:"
-  system: "SYSTEM:"
-
-mmproj: bakllava-mmproj.gguf
 parameters:
-  model: bakllava.gguf
-
+  model: minicpm-v-2_6-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
+- <|endoftext|>
 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 -}}
+    <|im_start|>assistant
+  chat_message: |
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content }}
+    {{ end -}}
+    {{ if .FunctionCall -}}
+    {{toJson .FunctionCall}}
+    {{ end -}}<|im_end|>
+  completion: |
     {{.Input}}
-    ASSISTANT:
+  function: |
+    <|im_start|>system
+    You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+    {{range .Functions}}
+    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+    {{end}}
+    For each function call return a json object with function name and arguments
+    <|im_end|>
+    {{.Input -}}
+    <|im_start|>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
-
-usage: |
-    curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-        "model": "gpt-4-vision-preview",
-        "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }}], "temperature": 0.9}]}'
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

aio/entrypoint.sh

@@ -129,7 +129,7 @@ detect_gpu
 detect_gpu_size
 
 PROFILE="${PROFILE:-$GPU_SIZE}" # default to cpu
-export MODELS="${MODELS:-/aio/${PROFILE}/embeddings.yaml,/aio/${PROFILE}/rerank.yaml,/aio/${PROFILE}/text-to-speech.yaml,/aio/${PROFILE}/image-gen.yaml,/aio/${PROFILE}/text-to-text.yaml,/aio/${PROFILE}/speech-to-text.yaml,/aio/${PROFILE}/vision.yaml}"
+export MODELS="${MODELS:-/aio/${PROFILE}/embeddings.yaml,/aio/${PROFILE}/rerank.yaml,/aio/${PROFILE}/text-to-speech.yaml,/aio/${PROFILE}/image-gen.yaml,/aio/${PROFILE}/text-to-text.yaml,/aio/${PROFILE}/speech-to-text.yaml,/aio/${PROFILE}/vad.yaml,/aio/${PROFILE}/vision.yaml}"
 
 check_vars
 

aio/gpu-8g/embeddings.yaml

@@ -1,7 +1,7 @@
+embeddings: true
 name: text-embedding-ada-002
-backend: sentencetransformers
 parameters:
-  model: all-MiniLM-L6-v2
+  model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 
 usage: |
     You can test this model with curl like this:

aio/gpu-8g/text-to-text.yaml

@@ -1,101 +1,53 @@
-name: gpt-4
-mmap: true
-parameters:
-  model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
-context_size: 8192
-
-stopwords:
-- "<|im_end|>"
-- "<dummy32000>"
-- "</tool_call>"
-- "<|eot_id|>"
-- "<|end_of_text|>"
-
+context_size: 4096
+f16: true
 function:
-  # disable injecting the "answer" tool
-  disable_no_action: true
-
+  capture_llm_results:
+  - (?s)<Thought>(.*?)</Thought>
   grammar:
-    # This allows the grammar to also return messages
-    mixed_mode: true
-    # Suffix to add to the grammar
-    #prefix: '<tool_call>\n'
-    # Force parallel calls in the grammar
-    # parallel_calls: true
-
-  return_name_in_function_response: true
-  # Without grammar uncomment the lines below
-  # Warning: this is relying only on the capability of the
-  # LLM model to generate the correct function call.
-  json_regex_match: 
-   - "(?s)<tool_call>(.*?)</tool_call>"
-   - "(?s)<tool_call>(.*?)"
+    properties_order: name,arguments
+  json_regex_match:
+  - (?s)<Output>(.*?)</Output>
   replace_llm_results:
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
+  - key: (?s)<Thought>(.*?)</Thought>
     value: ""
-  replace_function_results: 
-  # Replace everything that is not JSON array or object
-  # 
-  - key: '(?s)^[^{\[]*'
-    value: ""
-  - key: '(?s)[^}\]]*$'
-    value: ""
-  - key: "'([^']*?)'"
-    value: "_DQUOTE_${1}_DQUOTE_"
-  - key: '\\"'
-    value: "__TEMP_QUOTE__"
-  - key: "\'"
-    value: "'"
-  - key: "_DQUOTE_"
-    value: '"'
-  - key: "__TEMP_QUOTE__"
-    value: '"'
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-
+mmap: true
+name: gpt-4
+parameters:
+  model: localai-functioncall-qwen2.5-7b-v0.5-q4_k_m.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
 template:
   chat: |
     {{.Input -}}
     <|im_start|>assistant
   chat_message: |
-    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
-    {{- if .FunctionCall }}
-    <tool_call>
-    {{- else if eq .RoleName "tool" }}
-    <tool_response>
-    {{- end }}
-    {{- if .Content}}
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
     {{.Content }}
-    {{- end }}
-    {{- if .FunctionCall}}
+    {{ end -}}
+    {{ if .FunctionCall -}}
     {{toJson .FunctionCall}}
-    {{- end }}
-    {{- if .FunctionCall }}
-    </tool_call>
-    {{- else if eq .RoleName "tool" }}
-    </tool_response>
-    {{- end }}<|im_end|>
+    {{ end -}}<|im_end|>
   completion: |
     {{.Input}}
-  function: |-
+  function: |
     <|im_start|>system
-    You are a function calling AI model.
-    Here are the available tools:
-    <tools>
+    You are an AI assistant that executes function calls, and these are the tools at your disposal:
     {{range .Functions}}
     {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
     {{end}}
-    </tools>
-    You should call the tools provided to you sequentially
-    Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
-    <scratchpad>
-    {step-by-step reasoning and plan in bullet points}
-    </scratchpad>
-    For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
-    <tool_call>
-    {"arguments": <args-dict>, "name": <function-name>}
-    </tool_call><|im_end|>
+    <|im_end|>
     {{.Input -}}
-    <|im_start|>assistant
+    <|im_start|>assistant
+
+download_files:
+- filename: localai-functioncall-phi-4-v0.3-q4_k_m.gguf
+  sha256: 23fee048ded2a6e2e1a7b6bbefa6cbf83068f194caa9552aecbaa00fec8a16d5
+  uri: huggingface://mudler/LocalAI-functioncall-phi-4-v0.3-Q4_K_M-GGUF/localai-functioncall-phi-4-v0.3-q4_k_m.gguf

aio/gpu-8g/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/gpu-8g/vision.yaml

@@ -1,35 +1,49 @@
-backend: llama-cpp
 context_size: 4096
 f16: true
 mmap: true
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
-
-roles:
-  user: "USER:"
-  assistant: "ASSISTANT:"
-  system: "SYSTEM:"
-
-mmproj: llava-v1.6-7b-mmproj-f16.gguf
 parameters:
-  model: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  temperature: 0.2
-  top_k: 40
-  top_p: 0.95
-  seed: -1
-
+  model: minicpm-v-2_6-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
+- <|endoftext|>
 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 -}}
+    <|im_start|>assistant
+  chat_message: |
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content }}
+    {{ end -}}
+    {{ if .FunctionCall -}}
+    {{toJson .FunctionCall}}
+    {{ end -}}<|im_end|>
+  completion: |
     {{.Input}}
-    ASSISTANT:
+  function: |
+    <|im_start|>system
+    You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+    {{range .Functions}}
+    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+    {{end}}
+    For each function call return a json object with function name and arguments
+    <|im_end|>
+    {{.Input -}}
+    <|im_start|>assistant
 
 download_files:
-- filename: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/llava-v1.6-mistral-7b.Q5_K_M.gguf
-- filename: llava-v1.6-7b-mmproj-f16.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/mmproj-model-f16.gguf
-
-usage: |
-    curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-        "model": "gpt-4-vision-preview",
-        "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }}], "temperature": 0.9}]}'
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

aio/intel/embeddings.yaml

@@ -1,7 +1,7 @@
+embeddings: true
 name: text-embedding-ada-002
-backend: sentencetransformers
 parameters:
-  model: all-MiniLM-L6-v2
+  model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 
 usage: |
     You can test this model with curl like this:

aio/intel/text-to-text.yaml

@@ -1,103 +1,53 @@
-name: gpt-4
-mmap: false
-context_size: 8192
-
-f16: false
-parameters:
-  model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
-
-stopwords:
-- "<|im_end|>"
-- "<dummy32000>"
-- "</tool_call>"
-- "<|eot_id|>"
-- "<|end_of_text|>"
-
+context_size: 4096
+f16: true
 function:
-  # disable injecting the "answer" tool
-  disable_no_action: true
-
+  capture_llm_results:
+  - (?s)<Thought>(.*?)</Thought>
   grammar:
-    # This allows the grammar to also return messages
-    mixed_mode: true
-    # Suffix to add to the grammar
-    #prefix: '<tool_call>\n'
-    # Force parallel calls in the grammar
-    # parallel_calls: true
-
-  return_name_in_function_response: true
-  # Without grammar uncomment the lines below
-  # Warning: this is relying only on the capability of the
-  # LLM model to generate the correct function call.
-  json_regex_match: 
-   - "(?s)<tool_call>(.*?)</tool_call>"
-   - "(?s)<tool_call>(.*?)"
+    properties_order: name,arguments
+  json_regex_match:
+  - (?s)<Output>(.*?)</Output>
   replace_llm_results:
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
+  - key: (?s)<Thought>(.*?)</Thought>
     value: ""
-  replace_function_results: 
-  # Replace everything that is not JSON array or object
-  # 
-  - key: '(?s)^[^{\[]*'
-    value: ""
-  - key: '(?s)[^}\]]*$'
-    value: ""
-  - key: "'([^']*?)'"
-    value: "_DQUOTE_${1}_DQUOTE_"
-  - key: '\\"'
-    value: "__TEMP_QUOTE__"
-  - key: "\'"
-    value: "'"
-  - key: "_DQUOTE_"
-    value: '"'
-  - key: "__TEMP_QUOTE__"
-    value: '"'
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-
+mmap: true
+name: gpt-4
+parameters:
+  model: localai-functioncall-qwen2.5-7b-v0.5-q4_k_m.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
 template:
   chat: |
     {{.Input -}}
     <|im_start|>assistant
   chat_message: |
-    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
-    {{- if .FunctionCall }}
-    <tool_call>
-    {{- else if eq .RoleName "tool" }}
-    <tool_response>
-    {{- end }}
-    {{- if .Content}}
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
     {{.Content }}
-    {{- end }}
-    {{- if .FunctionCall}}
+    {{ end -}}
+    {{ if .FunctionCall -}}
     {{toJson .FunctionCall}}
-    {{- end }}
-    {{- if .FunctionCall }}
-    </tool_call>
-    {{- else if eq .RoleName "tool" }}
-    </tool_response>
-    {{- end }}<|im_end|>
+    {{ end -}}<|im_end|>
   completion: |
     {{.Input}}
-  function: |-
+  function: |
     <|im_start|>system
-    You are a function calling AI model.
-    Here are the available tools:
-    <tools>
+    You are an AI assistant that executes function calls, and these are the tools at your disposal:
     {{range .Functions}}
     {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
     {{end}}
-    </tools>
-    You should call the tools provided to you sequentially
-    Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
-    <scratchpad>
-    {step-by-step reasoning and plan in bullet points}
-    </scratchpad>
-    For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
-    <tool_call>
-    {"arguments": <args-dict>, "name": <function-name>}
-    </tool_call><|im_end|>
+    <|im_end|>
     {{.Input -}}
     <|im_start|>assistant
+
+download_files:
+- filename: localai-functioncall-phi-4-v0.3-q4_k_m.gguf
+  sha256: 23fee048ded2a6e2e1a7b6bbefa6cbf83068f194caa9552aecbaa00fec8a16d5
+  uri: huggingface://mudler/LocalAI-functioncall-phi-4-v0.3-Q4_K_M-GGUF/localai-functioncall-phi-4-v0.3-q4_k_m.gguf

aio/intel/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/intel/vision.yaml

@@ -1,35 +1,50 @@
-backend: llama-cpp
 context_size: 4096
-mmap: false
-f16: false
+f16: true
+mmap: true
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
-
-roles:
-  user: "USER:"
-  assistant: "ASSISTANT:"
-  system: "SYSTEM:"
-
-mmproj: llava-v1.6-7b-mmproj-f16.gguf
 parameters:
-  model: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  temperature: 0.2
-  top_k: 40
-  top_p: 0.95
-  seed: -1
-
+  model: minicpm-v-2_6-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
+- <|endoftext|>
 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 -}}
+    <|im_start|>assistant
+  chat_message: |
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content }}
+    {{ end -}}
+    {{ if .FunctionCall -}}
+    {{toJson .FunctionCall}}
+    {{ end -}}<|im_end|>
+  completion: |
     {{.Input}}
-    ASSISTANT:
+  function: |
+    <|im_start|>system
+    You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+    {{range .Functions}}
+    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+    {{end}}
+    For each function call return a json object with function name and arguments
+    <|im_end|>
+    {{.Input -}}
+    <|im_start|>assistant
+
 
 download_files:
-- filename: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/llava-v1.6-mistral-7b.Q5_K_M.gguf
-- filename: llava-v1.6-7b-mmproj-f16.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/mmproj-model-f16.gguf
-
-usage: |
-    curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-        "model": "gpt-4-vision-preview",
-        "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }}], "temperature": 0.9}]}'
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

backend/backend.proto

@@ -165,7 +165,6 @@ message Reply {
 
 message GrammarTrigger {
   string word = 1;
-  bool at_start = 2; 
 }
 
 message ModelOptions {
@@ -229,6 +228,11 @@ message ModelOptions {
   int32  MaxModelLen = 54;
   int32  TensorParallelSize = 55;
   string LoadFormat = 58;
+  bool   DisableLogStatus = 66;
+  string DType = 67;
+  int32  LimitImagePerPrompt = 68;
+  int32  LimitVideoPerPrompt = 69;
+  int32  LimitAudioPerPrompt = 70;
 
   string MMProj = 41;
 

backend/cpp/llama/grpc-server.cpp

@@ -467,9 +467,10 @@ struct llama_server_context
     bool all_slots_are_idle = false;
     bool add_bos_token      = true;
     bool has_eos_token      = true;
+    bool has_gpu = false;
 
     bool grammar_lazy = false;
-    std::vector<common_grammar_trigger> grammar_trigger_words;
+    std::vector<common_grammar_trigger> grammar_triggers;
 
     int32_t n_ctx;  // total context for all clients / slots
 
@@ -511,7 +512,10 @@ struct llama_server_context
         if (!params.mmproj.empty()) {
             multimodal = true;
             LOG_INFO("Multi Modal Mode Enabled", {});
-            clp_ctx = clip_model_load(params.mmproj.c_str(), /*verbosity=*/ 1);
+            clp_ctx = clip_init(params.mmproj.c_str(), clip_context_params {
+                /* use_gpu */ has_gpu,
+                /*verbosity=*/ 1,
+            });
             if(clp_ctx == nullptr) {
                 LOG_ERR("unable to load clip model: %s", params.mmproj.c_str());
                 return false;
@@ -709,7 +713,7 @@ struct llama_server_context
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
-        slot->sparams.grammar_trigger_words = grammar_trigger_words;
+        slot->sparams.grammar_triggers = grammar_triggers;
         slot->sparams.grammar_lazy = grammar_lazy;
 
         if (slot->n_predict > 0 && slot->params.n_predict > slot->n_predict) {
@@ -1155,6 +1159,14 @@ struct llama_server_context
             slot.has_next_token = false;
         }
 
+        if (slot.n_past >= slot.n_ctx) {
+            slot.truncated      = true;
+            slot.stopped_limit = true;
+            slot.has_next_token = false;
+
+            LOG_VERBOSE("stopped due to running out of context capacity", {});
+        }
+
         if (result.tok == llama_vocab_eos(vocab) || llama_vocab_is_eog(vocab, result.tok))
         {
             slot.stopped_eos = true;
@@ -1342,7 +1354,7 @@ struct llama_server_context
         queue_results.send(res);
     }
 
-    void send_embedding(llama_client_slot &slot)
+    void send_embedding(llama_client_slot &slot, const llama_batch & batch)
     {
         task_result res;
         res.id = slot.task_id;
@@ -1364,10 +1376,38 @@ struct llama_server_context
         else
         {
             const float *data = llama_get_embeddings(ctx);
-            std::vector<float> embedding(data, data + n_embd);
+            std::vector<float> embd_res(n_embd, 0.0f);
+            std::vector<std::vector<float>> embedding;
+            for (int i = 0; i < batch.n_tokens; ++i) {
+                if (!batch.logits[i] || batch.seq_id[i][0] != slot.id) {
+                    continue;
+                }
+
+                const float * embd = llama_get_embeddings_seq(ctx, batch.seq_id[i][0]);
+                if (embd == NULL) {
+                    embd = llama_get_embeddings_ith(ctx, i);
+                }
+
+                if (embd == NULL) {
+                    LOG("failed to get embeddings");
+
+                    continue;
+                }
+
+                // normalize only when there is pooling
+                // TODO: configurable
+                if (llama_pooling_type(ctx) != LLAMA_POOLING_TYPE_NONE) {
+                    common_embd_normalize(embd, embd_res.data(), n_embd, 2);
+                    embedding.push_back(embd_res);
+                } else {
+                    embedding.push_back({ embd, embd + n_embd });
+                }
+            }
+
+            // OAI compat
             res.result_json = json
             {
-                {"embedding", embedding },
+                {"embedding", embedding[0] },
             };
         }
         queue_results.send(res);
@@ -1627,17 +1667,17 @@ struct llama_server_context
             {
                 if (slot.is_processing() && system_tokens.size() + slot.cache_tokens.size() >= (size_t) slot.n_ctx)
                 {
+                    // this check is redundant (for good)
+                    // we should never get here, because generation should already stopped in process_token()
+
                     // START LOCALAI changes
                     // Temporary disable context-shifting as it can lead to infinite loops (issue: https://github.com/ggerganov/llama.cpp/issues/3969)
                     // See: https://github.com/mudler/LocalAI/issues/1333
                     // Context is exhausted, release the slot
                     slot.release();
                     send_final_response(slot);
-                    slot.cache_tokens.clear();
-                    slot.n_past = 0;
-                    slot.truncated = false;
-                    slot.has_next_token = true;
-                    LOG("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
+                    slot.has_next_token = false;
+                    LOG_ERROR("context is exhausted, release the slot", {});
 
                     continue;
                     // END LOCALAI changes
@@ -1988,7 +2028,7 @@ struct llama_server_context
                 // prompt evaluated for embedding
                 if (slot.embedding)
                 {
-                    send_embedding(slot);
+                    send_embedding(slot, batch_view);
                     slot.release();
                     slot.i_batch = -1;
                     continue;
@@ -2082,7 +2122,11 @@ static void append_to_generated_text_from_generated_token_probs(llama_server_con
 }
 
 std::function<void(int)> shutdown_handler;
-inline void signal_handler(int signal) { shutdown_handler(signal); }
+
+inline void signal_handler(int signal) {
+    exit(1);
+}
+
 
 /////////////////////////////////
 ////////////////////////////////
@@ -2278,7 +2322,7 @@ static std::string get_all_kv_cache_types() {
 }
 
 static void params_parse(const backend::ModelOptions* request,
-                                common_params & params) {
+                                common_params & params, llama_server_context &llama) {
    
     // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
 
@@ -2316,6 +2360,20 @@ static void params_parse(const backend::ModelOptions* request,
         add_rpc_devices(std::string(llama_grpc_servers));
     }
     
+     // decode options. Options are in form optname:optvale, or if booleans only optname.
+    for (int i = 0; i < request->options_size(); i++) {
+        std::string opt = request->options(i);
+        char *optname = strtok(&opt[0], ":");
+        char *optval = strtok(NULL, ":");
+        if (optval == NULL) {
+            optval = "true";
+        }
+
+        if (!strcmp(optname, "gpu")) {
+            llama.has_gpu = true;
+        }
+    }
+
     // TODO: Add yarn
 
     if (!request->tensorsplit().empty()) {
@@ -2385,12 +2443,12 @@ static void params_parse(const backend::ModelOptions* request,
         llama.grammar_lazy = true;
         for (int i = 0; i < request->grammartriggers_size(); i++) {
             common_grammar_trigger trigger;
-            trigger.word = request->grammartriggers(i).word();
-            trigger.at_start = request->grammartriggers(i).at_start();
-            llama.grammar_trigger_words.push_back(trigger);
+	    trigger.type = COMMON_GRAMMAR_TRIGGER_TYPE_WORD;
+            trigger.value = request->grammartriggers(i).word();
+	    // trigger.at_start = request->grammartriggers(i).at_start();
+            llama.grammar_triggers.push_back(trigger);
             LOG_INFO("grammar trigger", {
-                { "word", trigger.word },
-                { "at_start", trigger.at_start }
+                { "word", trigger.value },
             });
         }
     }
@@ -2409,7 +2467,7 @@ public:
   grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
     // Implement LoadModel RPC
     common_params params;
-    params_parse(request, params);
+    params_parse(request, params, llama);
 
     llama_backend_init();
     llama_numa_init(params.numa);
@@ -2595,6 +2653,20 @@ void RunServer(const std::string& server_address) {
 int main(int argc, char** argv) {
   std::string server_address("localhost:50051");
 
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+    struct sigaction sigint_action;
+    sigint_action.sa_handler = signal_handler;
+    sigemptyset (&sigint_action.sa_mask);
+    sigint_action.sa_flags = 0;
+    sigaction(SIGINT, &sigint_action, NULL);
+    sigaction(SIGTERM, &sigint_action, NULL);
+#elif defined (_WIN32)
+    auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
+        return (ctrl_type == CTRL_C_EVENT) ? (signal_handler(SIGINT), true) : false;
+    };
+    SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
+#endif
+
   // Define long and short options
   struct option long_options[] = {
       {"addr", required_argument, nullptr, 'a'},

backend/go/image/stablediffusion-ggml/gosd.cpp

@@ -35,6 +35,8 @@ const char* sample_method_str[] = {
     "ipndm",
     "ipndm_v",
     "lcm",
+    "ddim_trailing",
+    "tcd",
 };
 
 // Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
@@ -173,6 +175,7 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
                             -1, //clip_skip
                             cfg_scale, // sfg_scale
                             3.5f,
+			    0, // eta
                             width,
                             height,
                             sample_method, 

backend/python/autogptq/requirements.txt

@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi
 transformers

backend/python/bark/requirements.txt

@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi

backend/python/common/template/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 grpcio-tools

backend/python/coqui/requirements-cpu.txt

@@ -1,4 +1,4 @@
-transformers
+transformers==4.48.3
 accelerate
 torch==2.4.1
 coqui-tts

backend/python/coqui/requirements-cublas11.txt

@@ -1,6 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 torch==2.4.1+cu118
 torchaudio==2.4.1+cu118
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-cublas12.txt

@@ -1,5 +1,5 @@
 torch==2.4.1
 torchaudio==2.4.1
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-hipblas.txt

@@ -1,6 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 torch==2.4.1+rocm6.0
 torchaudio==2.4.1+rocm6.0
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-intel.txt

@@ -5,6 +5,6 @@ torchaudio==2.3.1+cxx11.abi
 oneccl_bind_pt==2.3.100+xpu
 optimum[openvino]
 setuptools
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi
 packaging==24.1

backend/python/diffusers/backend.py

@@ -159,6 +159,18 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 torchType = torch.float16
                 variant = "fp16"
 
+            options = request.Options
+
+            # empty dict
+            self.options = {}
+
+            # The options are a list of strings in this form optname:optvalue
+            # We are storing all the options in a dict so we can use it later when
+            # generating the images
+            for opt in options:
+                key, value = opt.split(":")
+                self.options[key] = value
+
             local = False
             modelFile = request.Model
 
@@ -441,6 +453,9 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         # create a dictionary of parameters by using the keys from EnableParameters and the values from defaults
         kwargs = {key: options.get(key) for key in keys if key in options}
 
+        # populate kwargs from self.options.
+        kwargs.update(self.options)
+
         # Set seed
         if request.seed > 0:
             kwargs["generator"] = torch.Generator(device=self.device).manual_seed(

backend/python/diffusers/requirements.txt

@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.70.0
+grpcio==1.71.0
 pillow
 protobuf
 certifi

backend/python/exllama2/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi
 wheel

backend/python/faster-whisper/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 grpcio-tools

backend/python/kokoro/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 phonemizer
 scipy

backend/python/rerankers/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi

backend/python/transformers/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi
 setuptools

backend/python/vllm/backend.py

@@ -109,6 +109,17 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             engine_args.swap_space = request.SwapSpace
         if request.MaxModelLen != 0:
             engine_args.max_model_len = request.MaxModelLen
+        if request.DisableLogStatus:
+            engine_args.disable_log_status = request.DisableLogStatus
+        if request.DType != "":
+            engine_args.dtype = request.DType
+        if request.LimitImagePerPrompt != 0 or request.LimitVideoPerPrompt != 0 or request.LimitAudioPerPrompt != 0:
+            # limit-mm-per-prompt defaults to 1 per modality, based on vLLM docs
+            engine_args.limit_mm_per_prompt = {
+                "image": max(request.LimitImagePerPrompt, 1),
+                "video": max(request.LimitVideoPerPrompt, 1),
+                "audio": max(request.LimitAudioPerPrompt, 1)
+            }
 
         try:
             self.llm = AsyncLLMEngine.from_engine_args(engine_args)
@@ -269,7 +280,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
     def load_image(self, image_path: str):
         """
         Load an image from the given file path or base64 encoded data.
-        
+
         Args:
             image_path (str): The path to the image file or base64 encoded data.
 
@@ -288,7 +299,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
     def load_video(self, video_path: str):
         """
         Load a video from the given file path.
-        
+
         Args:
             video_path (str): The path to the image file.
 
@@ -335,4 +346,4 @@ if __name__ == "__main__":
     )
     args = parser.parse_args()
 
-    asyncio.run(serve(args.addr))
+    asyncio.run(serve(args.addr))

backend/python/vllm/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.70.0
+grpcio==1.71.0
 protobuf
 certifi
 setuptools

core/application/startup.go

@@ -145,13 +145,7 @@ func New(opts ...config.AppOption) (*Application, error) {
 
 	if options.LoadToMemory != nil {
 		for _, m := range options.LoadToMemory {
-			cfg, err := application.BackendLoader().LoadBackendConfigFileByName(m, options.ModelPath,
-				config.LoadOptionDebug(options.Debug),
-				config.LoadOptionThreads(options.Threads),
-				config.LoadOptionContextSize(options.ContextSize),
-				config.LoadOptionF16(options.F16),
-				config.ModelPath(options.ModelPath),
-			)
+			cfg, err := application.BackendLoader().LoadBackendConfigFileByNameDefaultOptions(m, options)
 			if err != nil {
 				return nil, err
 			}

core/backend/llm.go

@@ -33,7 +33,7 @@ type TokenUsage struct {
 	TimingTokenGeneration  float64
 }
 
-func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c *config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model
 
 	// Check if the modelFile exists, if it doesn't try to load it from the gallery
@@ -48,7 +48,7 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 		}
 	}
 
-	opts := ModelOptions(c, o)
+	opts := ModelOptions(*c, o)
 	inferenceModel, err := loader.Load(opts...)
 	if err != nil {
 		return nil, err
@@ -84,7 +84,7 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 
 	// in GRPC, the backend is supposed to answer to 1 single token if stream is not supported
 	fn := func() (LLMResponse, error) {
-		opts := gRPCPredictOpts(c, loader.ModelPath)
+		opts := gRPCPredictOpts(*c, loader.ModelPath)
 		opts.Prompt = s
 		opts.Messages = protoMessages
 		opts.UseTokenizerTemplate = c.TemplateConfig.UseTokenizerTemplate
@@ -116,6 +116,11 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 		}
 
 		if tokenCallback != nil {
+
+			if c.TemplateConfig.ReplyPrefix != "" {
+				tokenCallback(c.TemplateConfig.ReplyPrefix, tokenUsage)
+			}
+
 			ss := ""
 
 			var partialRune []byte
@@ -165,8 +170,13 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 			tokenUsage.TimingTokenGeneration = reply.TimingTokenGeneration
 			tokenUsage.TimingPromptProcessing = reply.TimingPromptProcessing
 
+			response := string(reply.Message)
+			if c.TemplateConfig.ReplyPrefix != "" {
+				response = c.TemplateConfig.ReplyPrefix + response
+			}
+
 			return LLMResponse{
-				Response: string(reply.Message),
+				Response: response,
 				Usage:    tokenUsage,
 			}, err
 		}

core/backend/options.go

@@ -122,7 +122,6 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 	for _, t := range c.FunctionsConfig.GrammarConfig.GrammarTriggers {
 		triggers = append(triggers, &pb.GrammarTrigger{
 			Word:    t.Word,
-			AtStart: t.AtStart,
 		})
 
 	}
@@ -159,6 +158,12 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		SwapSpace:            int32(c.SwapSpace),
 		MaxModelLen:          int32(c.MaxModelLen),
 		TensorParallelSize:   int32(c.TensorParallelSize),
+		DisableLogStatus:     c.DisableLogStatus,
+		DType:                c.DType,
+		// LimitMMPerPrompt vLLM
+		LimitImagePerPrompt:  int32(c.LimitMMPerPrompt.LimitImagePerPrompt),
+		LimitVideoPerPrompt:  int32(c.LimitMMPerPrompt.LimitVideoPerPrompt),
+		LimitAudioPerPrompt:  int32(c.LimitMMPerPrompt.LimitAudioPerPrompt),
 		MMProj:               c.MMProj,
 		FlashAttention:       c.FlashAttention,
 		CacheTypeKey:         c.CacheTypeK,

core/backend/rerank.go

@@ -9,10 +9,10 @@ import (
 	model "github.com/mudler/LocalAI/pkg/model"
 )
 
-func Rerank(modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
-
-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
+func Rerank(request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
+	opts := ModelOptions(backendConfig, appConfig)
 	rerankModel, err := loader.Load(opts...)
+
 	if err != nil {
 		return nil, err
 	}

core/backend/soundgeneration.go

@@ -13,7 +13,6 @@ import (
 )
 
 func SoundGeneration(
-	modelFile string,
 	text string,
 	duration *float32,
 	temperature *float32,
@@ -25,8 +24,9 @@ func SoundGeneration(
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {
 
-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
+	opts := ModelOptions(backendConfig, appConfig)
 	soundGenModel, err := loader.Load(opts...)
+
 	if err != nil {
 		return "", nil, err
 	}
@@ -44,7 +44,7 @@ func SoundGeneration(
 
 	res, err := soundGenModel.SoundGeneration(context.Background(), &proto.SoundGenerationRequest{
 		Text:        text,
-		Model:       modelFile,
+		Model:       backendConfig.Model,
 		Dst:         filePath,
 		Sample:      doSample,
 		Duration:    duration,

core/backend/tokenize.go

@@ -4,19 +4,17 @@ import (
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/grpc"
-	model "github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/model"
 )
 
 func ModelTokenize(s string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (schema.TokenizeResponse, error) {
 
-	modelFile := backendConfig.Model
-
 	var inferenceModel grpc.Backend
 	var err error
 
-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
-
+	opts := ModelOptions(backendConfig, appConfig)
 	inferenceModel, err = loader.Load(opts...)
+
 	if err != nil {
 		return schema.TokenizeResponse{}, err
 	}

core/backend/transcript.go

@@ -47,7 +47,7 @@ func ModelTranscription(audio, language string, translate bool, ml *model.ModelL
 			tks = append(tks, int(t))
 		}
 		tr.Segments = append(tr.Segments,
-			schema.Segment{
+			schema.TranscriptionSegment{
 				Text:   s.Text,
 				Id:     int(s.Id),
 				Start:  time.Duration(s.Start),

core/backend/tts.go

@@ -14,28 +14,22 @@ import (
 )
 
 func ModelTTS(
-	backend,
 	text,
-	modelFile,
 	voice,
 	language string,
 	loader *model.ModelLoader,
 	appConfig *config.ApplicationConfig,
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {
-	bb := backend
-	if bb == "" {
-		bb = model.PiperBackend
-	}
-
-	opts := ModelOptions(backendConfig, appConfig, model.WithBackendString(bb), model.WithModel(modelFile))
+	opts := ModelOptions(backendConfig, appConfig, model.WithDefaultBackendString(model.PiperBackend))
 	ttsModel, err := loader.Load(opts...)
+
 	if err != nil {
 		return "", nil, err
 	}
 
 	if ttsModel == nil {
-		return "", nil, fmt.Errorf("could not load piper model")
+		return "", nil, fmt.Errorf("could not load tts model %q", backendConfig.Model)
 	}
 
 	if err := os.MkdirAll(appConfig.AudioDir, 0750); err != nil {
@@ -45,22 +39,21 @@ func ModelTTS(
 	fileName := utils.GenerateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
 	filePath := filepath.Join(appConfig.AudioDir, fileName)
 
-	// If the model file is not empty, we pass it joined with the model path
+	// We join the model name to the model path here. This seems to only be done for TTS and is HIGHLY suspect.
+	// This should be addressed in a follow up PR soon.
+	// Copying it over nearly verbatim, as TTS backends are not functional without this.
 	modelPath := ""
-	if modelFile != "" {
-		// If the model file is not empty, we pass it joined with the model path
-		// Checking first that it exists and is not outside ModelPath
-		// TODO: we should actually first check if the modelFile is looking like
-		// a FS path
-		mp := filepath.Join(loader.ModelPath, modelFile)
-		if _, err := os.Stat(mp); err == nil {
-			if err := utils.VerifyPath(mp, appConfig.ModelPath); err != nil {
-				return "", nil, err
-			}
-			modelPath = mp
-		} else {
-			modelPath = modelFile
+	// Checking first that it exists and is not outside ModelPath
+	// TODO: we should actually first check if the modelFile is looking like
+	// a FS path
+	mp := filepath.Join(loader.ModelPath, backendConfig.Model)
+	if _, err := os.Stat(mp); err == nil {
+		if err := utils.VerifyPath(mp, appConfig.ModelPath); err != nil {
+			return "", nil, err
 		}
+		modelPath = mp
+	} else {
+		modelPath = backendConfig.Model // skip this step if it fails?????
 	}
 
 	res, err := ttsModel.TTS(context.Background(), &proto.TTSRequest{

core/backend/vad.go

@@ -0,0 +1,38 @@
+package backend
+
+import (
+	"context"
+
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/model"
+)
+
+func VAD(request *schema.VADRequest,
+	ctx context.Context,
+	ml *model.ModelLoader,
+	appConfig *config.ApplicationConfig,
+	backendConfig config.BackendConfig) (*schema.VADResponse, error) {
+	opts := ModelOptions(backendConfig, appConfig)
+	vadModel, err := ml.Load(opts...)
+	if err != nil {
+		return nil, err
+	}
+	req := proto.VADRequest{
+		Audio: request.Audio,
+	}
+	resp, err := vadModel.VAD(ctx, &req)
+	if err != nil {
+		return nil, err
+	}
+
+	segments := []schema.VADSegment{}
+	for _, s := range resp.Segments {
+		segments = append(segments, schema.VADSegment{Start: s.Start, End: s.End})
+	}
+
+	return &schema.VADResponse{
+		Segments: segments,
+	}, nil
+}

core/cli/run.go

@@ -38,7 +38,7 @@ type RunCMD struct {
 
 	F16         bool `name:"f16" env:"LOCALAI_F16,F16" help:"Enable GPU acceleration" group:"performance"`
 	Threads     int  `env:"LOCALAI_THREADS,THREADS" short:"t" help:"Number of threads used for parallel computation. Usage of the number of physical cores in the system is suggested" group:"performance"`
-	ContextSize int  `env:"LOCALAI_CONTEXT_SIZE,CONTEXT_SIZE" default:"512" help:"Default context size for models" group:"performance"`
+	ContextSize int  `env:"LOCALAI_CONTEXT_SIZE,CONTEXT_SIZE" help:"Default context size for models" group:"performance"`
 
 	Address                            string   `env:"LOCALAI_ADDRESS,ADDRESS" default:":8080" help:"Bind address for the API server" group:"api"`
 	CORS                               bool     `env:"LOCALAI_CORS,CORS" help:"" group:"api"`

core/cli/soundgeneration.go

@@ -86,13 +86,14 @@ func (t *SoundGenerationCMD) Run(ctx *cliContext.Context) error {
 	options := config.BackendConfig{}
 	options.SetDefaults()
 	options.Backend = t.Backend
+	options.Model = t.Model
 
 	var inputFile *string
 	if t.InputFile != "" {
 		inputFile = &t.InputFile
 	}
 
-	filePath, _, err := backend.SoundGeneration(t.Model, text,
+	filePath, _, err := backend.SoundGeneration(text,
 		parseToFloat32Ptr(t.Duration), parseToFloat32Ptr(t.Temperature), &t.DoSample,
 		inputFile, parseToInt32Ptr(t.InputFileSampleDivisor), ml, opts, options)
 

core/cli/tts.go

@@ -52,8 +52,10 @@ func (t *TTSCMD) Run(ctx *cliContext.Context) error {
 
 	options := config.BackendConfig{}
 	options.SetDefaults()
+	options.Backend = t.Backend
+	options.Model = t.Model
 
-	filePath, _, err := backend.ModelTTS(t.Backend, text, t.Model, t.Voice, t.Language, ml, opts, options)
+	filePath, _, err := backend.ModelTTS(text, t.Voice, t.Language, ml, opts, options)
 	if err != nil {
 		return err
 	}

core/config/backend_config.go

@@ -130,25 +130,28 @@ type LLMConfig struct {
 	TrimSpace       []string `yaml:"trimspace"`
 	TrimSuffix      []string `yaml:"trimsuffix"`
 
-	ContextSize          *int      `yaml:"context_size"`
-	NUMA                 bool      `yaml:"numa"`
-	LoraAdapter          string    `yaml:"lora_adapter"`
-	LoraBase             string    `yaml:"lora_base"`
-	LoraAdapters         []string  `yaml:"lora_adapters"`
-	LoraScales           []float32 `yaml:"lora_scales"`
-	LoraScale            float32   `yaml:"lora_scale"`
-	NoMulMatQ            bool      `yaml:"no_mulmatq"`
-	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
-	SwapSpace            int       `yaml:"swap_space"`             // vLLM
-	MaxModelLen          int       `yaml:"max_model_len"`          // vLLM
-	TensorParallelSize   int       `yaml:"tensor_parallel_size"`   // vLLM
-	MMProj               string    `yaml:"mmproj"`
+	ContextSize          *int             `yaml:"context_size"`
+	NUMA                 bool             `yaml:"numa"`
+	LoraAdapter          string           `yaml:"lora_adapter"`
+	LoraBase             string           `yaml:"lora_base"`
+	LoraAdapters         []string         `yaml:"lora_adapters"`
+	LoraScales           []float32        `yaml:"lora_scales"`
+	LoraScale            float32          `yaml:"lora_scale"`
+	NoMulMatQ            bool             `yaml:"no_mulmatq"`
+	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
+	SwapSpace            int              `yaml:"swap_space"`             // vLLM
+	MaxModelLen          int              `yaml:"max_model_len"`          // vLLM
+	TensorParallelSize   int              `yaml:"tensor_parallel_size"`   // vLLM
+	DisableLogStatus     bool             `yaml:"disable_log_stats"`      // vLLM
+	DType                string           `yaml:"dtype"`                  // vLLM
+	LimitMMPerPrompt     LimitMMPerPrompt `yaml:"limit_mm_per_prompt"`    // vLLM
+	MMProj               string           `yaml:"mmproj"`
 
 	FlashAttention bool   `yaml:"flash_attention"`
 	NoKVOffloading bool   `yaml:"no_kv_offloading"`
@@ -166,6 +169,13 @@ type LLMConfig struct {
 	CFGScale float32 `yaml:"cfg_scale"` // Classifier-Free Guidance Scale
 }
 
+// LimitMMPerPrompt is a struct that holds the configuration for the limit-mm-per-prompt config in vLLM
+type LimitMMPerPrompt struct {
+	LimitImagePerPrompt int `yaml:"image"`
+	LimitVideoPerPrompt int `yaml:"video"`
+	LimitAudioPerPrompt int `yaml:"audio"`
+}
+
 // AutoGPTQ is a struct that holds the configuration specific to the AutoGPTQ backend
 type AutoGPTQ struct {
 	ModelBaseName    string `yaml:"model_base_name"`
@@ -203,6 +213,8 @@ type TemplateConfig struct {
 	Multimodal string `yaml:"multimodal"`
 
 	JinjaTemplate bool `yaml:"jinja_template"`
+
+	ReplyPrefix string `yaml:"reply_prefix"`
 }
 
 func (c *BackendConfig) UnmarshalYAML(value *yaml.Node) error {
@@ -212,7 +224,15 @@ func (c *BackendConfig) UnmarshalYAML(value *yaml.Node) error {
 		return err
 	}
 	*c = BackendConfig(aux)
+
 	c.KnownUsecases = GetUsecasesFromYAML(c.KnownUsecaseStrings)
+	// Make sure the usecases are valid, we rewrite with what we identified
+	c.KnownUsecaseStrings = []string{}
+	for k, usecase := range GetAllBackendConfigUsecases() {
+		if c.HasUsecases(usecase) {
+			c.KnownUsecaseStrings = append(c.KnownUsecaseStrings, k)
+		}
+	}
 	return nil
 }
 
@@ -369,16 +389,6 @@ func (cfg *BackendConfig) SetDefaults(opts ...ConfigLoaderOption) {
 		cfg.Embeddings = &falseV
 	}
 
-	// Value passed by the top level are treated as default (no implicit defaults)
-	// defaults are set by the user
-	if ctx == 0 {
-		ctx = 1024
-	}
-
-	if cfg.ContextSize == nil {
-		cfg.ContextSize = &ctx
-	}
-
 	if threads == 0 {
 		// Threads can't be 0
 		threads = 4
@@ -400,7 +410,7 @@ func (cfg *BackendConfig) SetDefaults(opts ...ConfigLoaderOption) {
 		cfg.Debug = &trueV
 	}
 
-	guessDefaultsFromFile(cfg, lo.modelPath)
+	guessDefaultsFromFile(cfg, lo.modelPath, ctx)
 }
 
 func (c *BackendConfig) Validate() bool {
@@ -437,19 +447,21 @@ func (c *BackendConfig) HasTemplate() bool {
 type BackendConfigUsecases int
 
 const (
-	FLAG_ANY              BackendConfigUsecases = 0b000000000
-	FLAG_CHAT             BackendConfigUsecases = 0b000000001
-	FLAG_COMPLETION       BackendConfigUsecases = 0b000000010
-	FLAG_EDIT             BackendConfigUsecases = 0b000000100
-	FLAG_EMBEDDINGS       BackendConfigUsecases = 0b000001000
-	FLAG_RERANK           BackendConfigUsecases = 0b000010000
-	FLAG_IMAGE            BackendConfigUsecases = 0b000100000
-	FLAG_TRANSCRIPT       BackendConfigUsecases = 0b001000000
-	FLAG_TTS              BackendConfigUsecases = 0b010000000
-	FLAG_SOUND_GENERATION BackendConfigUsecases = 0b100000000
+	FLAG_ANY              BackendConfigUsecases = 0b00000000000
+	FLAG_CHAT             BackendConfigUsecases = 0b00000000001
+	FLAG_COMPLETION       BackendConfigUsecases = 0b00000000010
+	FLAG_EDIT             BackendConfigUsecases = 0b00000000100
+	FLAG_EMBEDDINGS       BackendConfigUsecases = 0b00000001000
+	FLAG_RERANK           BackendConfigUsecases = 0b00000010000
+	FLAG_IMAGE            BackendConfigUsecases = 0b00000100000
+	FLAG_TRANSCRIPT       BackendConfigUsecases = 0b00001000000
+	FLAG_TTS              BackendConfigUsecases = 0b00010000000
+	FLAG_SOUND_GENERATION BackendConfigUsecases = 0b00100000000
+	FLAG_TOKENIZE         BackendConfigUsecases = 0b01000000000
+	FLAG_VAD              BackendConfigUsecases = 0b10000000000
 
 	// Common Subsets
-	FLAG_LLM BackendConfigUsecases = FLAG_CHAT & FLAG_COMPLETION & FLAG_EDIT
+	FLAG_LLM BackendConfigUsecases = FLAG_CHAT | FLAG_COMPLETION | FLAG_EDIT
 )
 
 func GetAllBackendConfigUsecases() map[string]BackendConfigUsecases {
@@ -464,10 +476,16 @@ func GetAllBackendConfigUsecases() map[string]BackendConfigUsecases {
 		"FLAG_TRANSCRIPT":       FLAG_TRANSCRIPT,
 		"FLAG_TTS":              FLAG_TTS,
 		"FLAG_SOUND_GENERATION": FLAG_SOUND_GENERATION,
+		"FLAG_TOKENIZE":         FLAG_TOKENIZE,
+		"FLAG_VAD":              FLAG_VAD,
 		"FLAG_LLM":              FLAG_LLM,
 	}
 }
 
+func stringToFlag(s string) string {
+	return "FLAG_" + strings.ToUpper(s)
+}
+
 func GetUsecasesFromYAML(input []string) *BackendConfigUsecases {
 	if len(input) == 0 {
 		return nil
@@ -475,7 +493,7 @@ func GetUsecasesFromYAML(input []string) *BackendConfigUsecases {
 	result := FLAG_ANY
 	flags := GetAllBackendConfigUsecases()
 	for _, str := range input {
-		flag, exists := flags["FLAG_"+strings.ToUpper(str)]
+		flag, exists := flags[stringToFlag(str)]
 		if exists {
 			result |= flag
 		}
@@ -549,5 +567,18 @@ func (c *BackendConfig) GuessUsecases(u BackendConfigUsecases) bool {
 		}
 	}
 
+	if (u & FLAG_TOKENIZE) == FLAG_TOKENIZE {
+		tokenizeCapableBackends := []string{"llama.cpp", "rwkv"}
+		if !slices.Contains(tokenizeCapableBackends, c.Backend) {
+			return false
+		}
+	}
+
+	if (u & FLAG_VAD) == FLAG_VAD {
+		if c.Backend != "silero-vad" {
+			return false
+		}
+	}
+
 	return true
 }

core/config/backend_config_loader.go

@@ -81,10 +81,10 @@ func readMultipleBackendConfigsFromFile(file string, opts ...ConfigLoaderOption)
 	c := &[]*BackendConfig{}
 	f, err := os.ReadFile(file)
 	if err != nil {
-		return nil, fmt.Errorf("cannot read config file: %w", err)
+		return nil, fmt.Errorf("readMultipleBackendConfigsFromFile cannot read config file %q: %w", file, err)
 	}
 	if err := yaml.Unmarshal(f, c); err != nil {
-		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
+		return nil, fmt.Errorf("readMultipleBackendConfigsFromFile cannot unmarshal config file %q: %w", file, err)
 	}
 
 	for _, cc := range *c {
@@ -101,10 +101,10 @@ func readBackendConfigFromFile(file string, opts ...ConfigLoaderOption) (*Backen
 	c := &BackendConfig{}
 	f, err := os.ReadFile(file)
 	if err != nil {
-		return nil, fmt.Errorf("cannot read config file: %w", err)
+		return nil, fmt.Errorf("readBackendConfigFromFile cannot read config file %q: %w", file, err)
 	}
 	if err := yaml.Unmarshal(f, c); err != nil {
-		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
+		return nil, fmt.Errorf("readBackendConfigFromFile cannot unmarshal config file %q: %w", file, err)
 	}
 
 	c.SetDefaults(opts...)
@@ -117,7 +117,9 @@ func (bcl *BackendConfigLoader) LoadBackendConfigFileByName(modelName, modelPath
 	// Load a config file if present after the model name
 	cfg := &BackendConfig{
 		PredictionOptions: schema.PredictionOptions{
-			Model: modelName,
+			BasicModelRequest: schema.BasicModelRequest{
+				Model: modelName,
+			},
 		},
 	}
 
@@ -145,6 +147,15 @@ func (bcl *BackendConfigLoader) LoadBackendConfigFileByName(modelName, modelPath
 	return cfg, nil
 }
 
+func (bcl *BackendConfigLoader) LoadBackendConfigFileByNameDefaultOptions(modelName string, appConfig *ApplicationConfig) (*BackendConfig, error) {
+	return bcl.LoadBackendConfigFileByName(modelName, appConfig.ModelPath,
+		LoadOptionDebug(appConfig.Debug),
+		LoadOptionThreads(appConfig.Threads),
+		LoadOptionContextSize(appConfig.ContextSize),
+		LoadOptionF16(appConfig.F16),
+		ModelPath(appConfig.ModelPath))
+}
+
 // This format is currently only used when reading a single file at startup, passed in via ApplicationConfig.ConfigFile
 func (bcl *BackendConfigLoader) LoadMultipleBackendConfigsSingleFile(file string, opts ...ConfigLoaderOption) error {
 	bcl.Lock()
@@ -167,7 +178,7 @@ func (bcl *BackendConfigLoader) LoadBackendConfig(file string, opts ...ConfigLoa
 	defer bcl.Unlock()
 	c, err := readBackendConfigFromFile(file, opts...)
 	if err != nil {
-		return fmt.Errorf("cannot read config file: %w", err)
+		return fmt.Errorf("LoadBackendConfig cannot read config file %q: %w", file, err)
 	}
 
 	if c.Validate() {
@@ -324,9 +335,10 @@ func (bcl *BackendConfigLoader) Preload(modelPath string) error {
 func (bcl *BackendConfigLoader) LoadBackendConfigsFromPath(path string, opts ...ConfigLoaderOption) error {
 	bcl.Lock()
 	defer bcl.Unlock()
+
 	entries, err := os.ReadDir(path)
 	if err != nil {
-		return fmt.Errorf("cannot read directory '%s': %w", path, err)
+		return fmt.Errorf("LoadBackendConfigsFromPath cannot read directory '%s': %w", path, err)
 	}
 	files := make([]fs.FileInfo, 0, len(entries))
 	for _, entry := range entries {
@@ -344,13 +356,13 @@ func (bcl *BackendConfigLoader) LoadBackendConfigsFromPath(path string, opts ...
 		}
 		c, err := readBackendConfigFromFile(filepath.Join(path, file.Name()), opts...)
 		if err != nil {
-			log.Error().Err(err).Msgf("cannot read config file: %s", file.Name())
+			log.Error().Err(err).Str("File Name", file.Name()).Msgf("LoadBackendConfigsFromPath cannot read config file")
 			continue
 		}
 		if c.Validate() {
 			bcl.configs[c.Name] = *c
 		} else {
-			log.Error().Err(err).Msgf("config is not valid")
+			log.Error().Err(err).Str("Name", c.Name).Msgf("config is not valid")
 		}
 	}
 

core/config/gguf.go

@@ -0,0 +1,253 @@
+package config
+
+import (
+	"strings"
+
+	"github.com/rs/zerolog/log"
+
+	gguf "github.com/thxcode/gguf-parser-go"
+)
+
+type familyType uint8
+
+const (
+	Unknown familyType = iota
+	LLaMa3
+	CommandR
+	Phi3
+	ChatML
+	Mistral03
+	Gemma
+	DeepSeek2
+)
+
+const (
+	defaultContextSize = 1024
+)
+
+type settingsConfig struct {
+	StopWords      []string
+	TemplateConfig TemplateConfig
+	RepeatPenalty  float64
+}
+
+// default settings to adopt with a given model family
+var defaultsSettings map[familyType]settingsConfig = map[familyType]settingsConfig{
+	Gemma: {
+		RepeatPenalty: 1.0,
+		StopWords:     []string{"<|im_end|>", "<end_of_turn>", "<start_of_turn>"},
+		TemplateConfig: TemplateConfig{
+			Chat:        "{{.Input }}\n<start_of_turn>model\n",
+			ChatMessage: "<start_of_turn>{{if eq .RoleName \"assistant\" }}model{{else}}{{ .RoleName }}{{end}}\n{{ if .Content -}}\n{{.Content -}}\n{{ end -}}<end_of_turn>",
+			Completion:  "{{.Input}}",
+		},
+	},
+	DeepSeek2: {
+		StopWords: []string{"<｜end▁of▁sentence｜>"},
+		TemplateConfig: TemplateConfig{
+			ChatMessage: `{{if eq .RoleName "user" -}}User: {{.Content }}
+{{ end -}}
+{{if eq .RoleName "assistant" -}}Assistant: {{.Content}}<｜end▁of▁sentence｜>{{end}}
+{{if eq .RoleName "system" -}}{{.Content}}
+{{end -}}`,
+			Chat: "{{.Input -}}\nAssistant: ",
+		},
+	},
+	LLaMa3: {
+		StopWords: []string{"<|eot_id|>"},
+		TemplateConfig: TemplateConfig{
+			Chat:        "<|begin_of_text|>{{.Input }}\n<|start_header_id|>assistant<|end_header_id|>",
+			ChatMessage: "<|start_header_id|>{{ .RoleName }}<|end_header_id|>\n\n{{.Content }}<|eot_id|>",
+		},
+	},
+	CommandR: {
+		TemplateConfig: TemplateConfig{
+			Chat: "{{.Input -}}<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>",
+			Functions: `<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>
+You are a function calling AI model, you can call the following functions:
+## Available Tools
+{{range .Functions}}
+- {"type": "function", "function": {"name": "{{.Name}}", "description": "{{.Description}}", "parameters": {{toJson .Parameters}} }}
+{{end}}
+When using a tool, reply with JSON, for instance {"name": "tool_name", "arguments": {"param1": "value1", "param2": "value2"}}
+<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>{{.Input -}}`,
+			ChatMessage: `{{if eq .RoleName "user" -}}
+<|START_OF_TURN_TOKEN|><|USER_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
+{{- else if eq .RoleName "system" -}}
+<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
+{{- else if eq .RoleName "assistant" -}}
+<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
+{{- else if eq .RoleName "tool" -}}
+<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
+{{- else if .FunctionCall -}}
+<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>{{toJson .FunctionCall}}}<|END_OF_TURN_TOKEN|>
+{{- end -}}`,
+		},
+		StopWords: []string{"<|END_OF_TURN_TOKEN|>"},
+	},
+	Phi3: {
+		TemplateConfig: TemplateConfig{
+			Chat:        "{{.Input}}\n<|assistant|>",
+			ChatMessage: "<|{{ .RoleName }}|>\n{{.Content}}<|end|>",
+			Completion:  "{{.Input}}",
+		},
+		StopWords: []string{"<|end|>", "<|endoftext|>"},
+	},
+	ChatML: {
+		TemplateConfig: TemplateConfig{
+			Chat: "{{.Input -}}\n<|im_start|>assistant",
+			Functions: `<|im_start|>system
+You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+{{range .Functions}}
+{'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+{{end}}
+For each function call return a json object with function name and arguments
+<|im_end|>
+{{.Input -}}
+<|im_start|>assistant`,
+			ChatMessage: `<|im_start|>{{ .RoleName }}
+{{ if .FunctionCall -}}
+Function call:
+{{ else if eq .RoleName "tool" -}}
+Function response:
+{{ end -}}
+{{ if .Content -}}
+{{.Content }}
+{{ end -}}
+{{ if .FunctionCall -}}
+{{toJson .FunctionCall}}
+{{ end -}}<|im_end|>`,
+		},
+		StopWords: []string{"<|im_end|>", "<dummy32000>", "</s>"},
+	},
+	Mistral03: {
+		TemplateConfig: TemplateConfig{
+			Chat:      "{{.Input -}}",
+			Functions: `[AVAILABLE_TOOLS] [{{range .Functions}}{"type": "function", "function": {"name": "{{.Name}}", "description": "{{.Description}}", "parameters": {{toJson .Parameters}} }}{{end}} ] [/AVAILABLE_TOOLS]{{.Input }}`,
+			ChatMessage: `{{if eq .RoleName "user" -}}
+[INST] {{.Content }} [/INST]
+{{- else if .FunctionCall -}}
+[TOOL_CALLS] {{toJson .FunctionCall}} [/TOOL_CALLS]
+{{- else if eq .RoleName "tool" -}}
+[TOOL_RESULTS] {{.Content}} [/TOOL_RESULTS]
+{{- else -}}
+{{ .Content -}}
+{{ end -}}`,
+		},
+		StopWords: []string{"<|im_end|>", "<dummy32000>", "</tool_call>", "<|eot_id|>", "<|end_of_text|>", "</s>", "[/TOOL_CALLS]", "[/ACTIONS]"},
+	},
+}
+
+// this maps well known template used in HF to model families defined above
+var knownTemplates = map[string]familyType{
+	`{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ system_message }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|im_start|>user\n' + content + '<|im_end|>\n<|im_start|>assistant\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|im_end|>' + '\n' }}{% endif %}{% endfor %}`:                              ChatML,
+	`{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}`: Mistral03,
+}
+
+func guessGGUFFromFile(cfg *BackendConfig, f *gguf.GGUFFile, defaultCtx int) {
+
+	if defaultCtx == 0 && cfg.ContextSize == nil {
+		ctxSize := f.EstimateLLaMACppUsage().ContextSize
+		if ctxSize > 0 {
+			cSize := int(ctxSize)
+			cfg.ContextSize = &cSize
+		} else {
+			defaultCtx = defaultContextSize
+			cfg.ContextSize = &defaultCtx
+		}
+	}
+
+	if cfg.HasTemplate() {
+		// nothing to guess here
+		log.Debug().Any("name", cfg.Name).Msgf("guessDefaultsFromFile: %s", "template already set")
+		return
+	}
+
+	log.Debug().
+		Any("eosTokenID", f.Tokenizer().EOSTokenID).
+		Any("bosTokenID", f.Tokenizer().BOSTokenID).
+		Any("modelName", f.Model().Name).
+		Any("architecture", f.Architecture().Architecture).Msgf("Model file loaded: %s", cfg.ModelFileName())
+
+	// guess the name
+	if cfg.Name == "" {
+		cfg.Name = f.Model().Name
+	}
+
+	family := identifyFamily(f)
+
+	if family == Unknown {
+		log.Debug().Msgf("guessDefaultsFromFile: %s", "family not identified")
+		return
+	}
+
+	// identify template
+	settings, ok := defaultsSettings[family]
+	if ok {
+		cfg.TemplateConfig = settings.TemplateConfig
+		log.Debug().Any("family", family).Msgf("guessDefaultsFromFile: guessed template %+v", cfg.TemplateConfig)
+		if len(cfg.StopWords) == 0 {
+			cfg.StopWords = settings.StopWords
+		}
+		if cfg.RepeatPenalty == 0.0 {
+			cfg.RepeatPenalty = settings.RepeatPenalty
+		}
+	} else {
+		log.Debug().Any("family", family).Msgf("guessDefaultsFromFile: no template found for family")
+	}
+
+	if cfg.HasTemplate() {
+		return
+	}
+
+	// identify from well known templates first, otherwise use the raw jinja template
+	chatTemplate, found := f.Header.MetadataKV.Get("tokenizer.chat_template")
+	if found {
+		// try to use the jinja template
+		cfg.TemplateConfig.JinjaTemplate = true
+		cfg.TemplateConfig.ChatMessage = chatTemplate.ValueString()
+	}
+}
+
+func identifyFamily(f *gguf.GGUFFile) familyType {
+
+	// identify from well known templates first
+	chatTemplate, found := f.Header.MetadataKV.Get("tokenizer.chat_template")
+	if found && chatTemplate.ValueString() != "" {
+		if family, ok := knownTemplates[chatTemplate.ValueString()]; ok {
+			return family
+		}
+	}
+
+	// otherwise try to identify from the model properties
+	arch := f.Architecture().Architecture
+	eosTokenID := f.Tokenizer().EOSTokenID
+	bosTokenID := f.Tokenizer().BOSTokenID
+
+	isYI := arch == "llama" && bosTokenID == 1 && eosTokenID == 2
+	// WTF! Mistral0.3 and isYi have same bosTokenID and eosTokenID
+
+	llama3 := arch == "llama" && eosTokenID == 128009
+	commandR := arch == "command-r" && eosTokenID == 255001
+	qwen2 := arch == "qwen2"
+	phi3 := arch == "phi-3"
+	gemma := strings.HasPrefix(arch, "gemma") || strings.Contains(strings.ToLower(f.Model().Name), "gemma")
+	deepseek2 := arch == "deepseek2"
+
+	switch {
+	case deepseek2:
+		return DeepSeek2
+	case gemma:
+		return Gemma
+	case llama3:
+		return LLaMa3
+	case commandR:
+		return CommandR
+	case phi3:
+		return Phi3
+	case qwen2, isYI:
+		return ChatML
+	default:
+		return Unknown
+	}
+}

core/config/guesser.go

@@ -3,147 +3,12 @@ package config
 import (
 	"os"
 	"path/filepath"
-	"strings"
 
 	"github.com/rs/zerolog/log"
-
 	gguf "github.com/thxcode/gguf-parser-go"
 )
 
-type familyType uint8
-
-const (
-	Unknown familyType = iota
-	LLaMa3
-	CommandR
-	Phi3
-	ChatML
-	Mistral03
-	Gemma
-	DeepSeek2
-)
-
-type settingsConfig struct {
-	StopWords      []string
-	TemplateConfig TemplateConfig
-	RepeatPenalty  float64
-}
-
-// default settings to adopt with a given model family
-var defaultsSettings map[familyType]settingsConfig = map[familyType]settingsConfig{
-	Gemma: {
-		RepeatPenalty: 1.0,
-		StopWords:     []string{"<|im_end|>", "<end_of_turn>", "<start_of_turn>"},
-		TemplateConfig: TemplateConfig{
-			Chat:        "{{.Input }}\n<start_of_turn>model\n",
-			ChatMessage: "<start_of_turn>{{if eq .RoleName \"assistant\" }}model{{else}}{{ .RoleName }}{{end}}\n{{ if .Content -}}\n{{.Content -}}\n{{ end -}}<end_of_turn>",
-			Completion:  "{{.Input}}",
-		},
-	},
-	DeepSeek2: {
-		StopWords: []string{"<｜end▁of▁sentence｜>"},
-		TemplateConfig: TemplateConfig{
-			ChatMessage: `{{if eq .RoleName "user" -}}User: {{.Content }}
-{{ end -}}
-{{if eq .RoleName "assistant" -}}Assistant: {{.Content}}<｜end▁of▁sentence｜>{{end}}
-{{if eq .RoleName "system" -}}{{.Content}}
-{{end -}}`,
-			Chat: "{{.Input -}}\nAssistant: ",
-		},
-	},
-	LLaMa3: {
-		StopWords: []string{"<|eot_id|>"},
-		TemplateConfig: TemplateConfig{
-			Chat:        "<|begin_of_text|>{{.Input }}\n<|start_header_id|>assistant<|end_header_id|>",
-			ChatMessage: "<|start_header_id|>{{ .RoleName }}<|end_header_id|>\n\n{{.Content }}<|eot_id|>",
-		},
-	},
-	CommandR: {
-		TemplateConfig: TemplateConfig{
-			Chat: "{{.Input -}}<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>",
-			Functions: `<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>
-You are a function calling AI model, you can call the following functions:
-## Available Tools
-{{range .Functions}}
-- {"type": "function", "function": {"name": "{{.Name}}", "description": "{{.Description}}", "parameters": {{toJson .Parameters}} }}
-{{end}}
-When using a tool, reply with JSON, for instance {"name": "tool_name", "arguments": {"param1": "value1", "param2": "value2"}}
-<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>{{.Input -}}`,
-			ChatMessage: `{{if eq .RoleName "user" -}}
-<|START_OF_TURN_TOKEN|><|USER_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
-{{- else if eq .RoleName "system" -}}
-<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
-{{- else if eq .RoleName "assistant" -}}
-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
-{{- else if eq .RoleName "tool" -}}
-<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>{{.Content}}<|END_OF_TURN_TOKEN|>
-{{- else if .FunctionCall -}}
-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>{{toJson .FunctionCall}}}<|END_OF_TURN_TOKEN|>
-{{- end -}}`,
-		},
-		StopWords: []string{"<|END_OF_TURN_TOKEN|>"},
-	},
-	Phi3: {
-		TemplateConfig: TemplateConfig{
-			Chat:        "{{.Input}}\n<|assistant|>",
-			ChatMessage: "<|{{ .RoleName }}|>\n{{.Content}}<|end|>",
-			Completion:  "{{.Input}}",
-		},
-		StopWords: []string{"<|end|>", "<|endoftext|>"},
-	},
-	ChatML: {
-		TemplateConfig: TemplateConfig{
-			Chat: "{{.Input -}}\n<|im_start|>assistant",
-			Functions: `<|im_start|>system
-You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
-{{range .Functions}}
-{'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
-{{end}}
-For each function call return a json object with function name and arguments
-<|im_end|>
-{{.Input -}}
-<|im_start|>assistant`,
-			ChatMessage: `<|im_start|>{{ .RoleName }}
-{{ if .FunctionCall -}}
-Function call:
-{{ else if eq .RoleName "tool" -}}
-Function response:
-{{ end -}}
-{{ if .Content -}}
-{{.Content }}
-{{ end -}}
-{{ if .FunctionCall -}}
-{{toJson .FunctionCall}}
-{{ end -}}<|im_end|>`,
-		},
-		StopWords: []string{"<|im_end|>", "<dummy32000>", "</s>"},
-	},
-	Mistral03: {
-		TemplateConfig: TemplateConfig{
-			Chat:      "{{.Input -}}",
-			Functions: `[AVAILABLE_TOOLS] [{{range .Functions}}{"type": "function", "function": {"name": "{{.Name}}", "description": "{{.Description}}", "parameters": {{toJson .Parameters}} }}{{end}} ] [/AVAILABLE_TOOLS]{{.Input }}`,
-			ChatMessage: `{{if eq .RoleName "user" -}}
-[INST] {{.Content }} [/INST]
-{{- else if .FunctionCall -}}
-[TOOL_CALLS] {{toJson .FunctionCall}} [/TOOL_CALLS]
-{{- else if eq .RoleName "tool" -}}
-[TOOL_RESULTS] {{.Content}} [/TOOL_RESULTS]
-{{- else -}}
-{{ .Content -}}
-{{ end -}}`,
-		},
-		StopWords: []string{"<|im_end|>", "<dummy32000>", "</tool_call>", "<|eot_id|>", "<|end_of_text|>", "</s>", "[/TOOL_CALLS]", "[/ACTIONS]"},
-	},
-}
-
-// this maps well known template used in HF to model families defined above
-var knownTemplates = map[string]familyType{
-	`{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ system_message }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|im_start|>user\n' + content + '<|im_end|>\n<|im_start|>assistant\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|im_end|>' + '\n' }}{% endif %}{% endfor %}`:                              ChatML,
-	`{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}`: Mistral03,
-}
-
-func guessDefaultsFromFile(cfg *BackendConfig, modelPath string) {
-
+func guessDefaultsFromFile(cfg *BackendConfig, modelPath string, defaultCtx int) {
 	if os.Getenv("LOCALAI_DISABLE_GUESSING") == "true" {
 		log.Debug().Msgf("guessDefaultsFromFile: %s", "guessing disabled with LOCALAI_DISABLE_GUESSING")
 		return
@@ -154,105 +19,20 @@ func guessDefaultsFromFile(cfg *BackendConfig, modelPath string) {
 		return
 	}
 
-	if cfg.HasTemplate() {
-		// nothing to guess here
-		log.Debug().Any("name", cfg.Name).Msgf("guessDefaultsFromFile: %s", "template already set")
-		return
-	}
-
 	// We try to guess only if we don't have a template defined already
-	f, err := gguf.ParseGGUFFile(filepath.Join(modelPath, cfg.ModelFileName()))
-	if err != nil {
-		// Only valid for gguf files
-		log.Debug().Msgf("guessDefaultsFromFile: %s", "not a GGUF file")
+	guessPath := filepath.Join(modelPath, cfg.ModelFileName())
+
+	// try to parse the gguf file
+	f, err := gguf.ParseGGUFFile(guessPath)
+	if err == nil {
+		guessGGUFFromFile(cfg, f, defaultCtx)
 		return
 	}
 
-	log.Debug().
-		Any("eosTokenID", f.Tokenizer().EOSTokenID).
-		Any("bosTokenID", f.Tokenizer().BOSTokenID).
-		Any("modelName", f.Model().Name).
-		Any("architecture", f.Architecture().Architecture).Msgf("Model file loaded: %s", cfg.ModelFileName())
-
-	// guess the name
-	if cfg.Name == "" {
-		cfg.Name = f.Model().Name
-	}
-
-	family := identifyFamily(f)
-
-	if family == Unknown {
-		log.Debug().Msgf("guessDefaultsFromFile: %s", "family not identified")
-		return
-	}
-
-	// identify template
-	settings, ok := defaultsSettings[family]
-	if ok {
-		cfg.TemplateConfig = settings.TemplateConfig
-		log.Debug().Any("family", family).Msgf("guessDefaultsFromFile: guessed template %+v", cfg.TemplateConfig)
-		if len(cfg.StopWords) == 0 {
-			cfg.StopWords = settings.StopWords
+	if cfg.ContextSize == nil {
+		if defaultCtx == 0 {
+			defaultCtx = defaultContextSize
 		}
-		if cfg.RepeatPenalty == 0.0 {
-			cfg.RepeatPenalty = settings.RepeatPenalty
-		}
-	} else {
-		log.Debug().Any("family", family).Msgf("guessDefaultsFromFile: no template found for family")
-	}
-
-	if cfg.HasTemplate() {
-		return
-	}
-
-	// identify from well known templates first, otherwise use the raw jinja template
-	chatTemplate, found := f.Header.MetadataKV.Get("tokenizer.chat_template")
-	if found {
-		// try to use the jinja template
-		cfg.TemplateConfig.JinjaTemplate = true
-		cfg.TemplateConfig.ChatMessage = chatTemplate.ValueString()
-	}
-}
-
-func identifyFamily(f *gguf.GGUFFile) familyType {
-
-	// identify from well known templates first
-	chatTemplate, found := f.Header.MetadataKV.Get("tokenizer.chat_template")
-	if found && chatTemplate.ValueString() != "" {
-		if family, ok := knownTemplates[chatTemplate.ValueString()]; ok {
-			return family
-		}
-	}
-
-	// otherwise try to identify from the model properties
-	arch := f.Architecture().Architecture
-	eosTokenID := f.Tokenizer().EOSTokenID
-	bosTokenID := f.Tokenizer().BOSTokenID
-
-	isYI := arch == "llama" && bosTokenID == 1 && eosTokenID == 2
-	// WTF! Mistral0.3 and isYi have same bosTokenID and eosTokenID
-
-	llama3 := arch == "llama" && eosTokenID == 128009
-	commandR := arch == "command-r" && eosTokenID == 255001
-	qwen2 := arch == "qwen2"
-	phi3 := arch == "phi-3"
-	gemma := strings.HasPrefix(arch, "gemma") || strings.Contains(strings.ToLower(f.Model().Name), "gemma")
-	deepseek2 := arch == "deepseek2"
-
-	switch {
-	case deepseek2:
-		return DeepSeek2
-	case gemma:
-		return Gemma
-	case llama3:
-		return LLaMa3
-	case commandR:
-		return CommandR
-	case phi3:
-		return Phi3
-	case qwen2, isYI:
-		return ChatML
-	default:
-		return Unknown
+		cfg.ContextSize = &defaultCtx
 	}
 }

core/gallery/gallery.go

@@ -29,6 +29,8 @@ func InstallModelFromGallery(galleries []config.Gallery, name string, basePath s
 			if err != nil {
 				return err
 			}
+			config.Description = model.Description
+			config.License = model.License
 		} else if len(model.ConfigFile) > 0 {
 			// TODO: is this worse than using the override method with a blank cfg yaml?
 			reYamlConfig, err := yaml.Marshal(model.ConfigFile)
@@ -114,7 +116,7 @@ func FindModel(models []*GalleryModel, name string, basePath string) *GalleryMod
 // List available models
 // Models galleries are a list of yaml files that are hosted on a remote server (for example github).
 // Each yaml file contains a list of models that can be downloaded and optionally overrides to define a new model setting.
-func AvailableGalleryModels(galleries []config.Gallery, basePath string) ([]*GalleryModel, error) {
+func AvailableGalleryModels(galleries []config.Gallery, basePath string) (GalleryModels, error) {
 	var models []*GalleryModel
 
 	// Get models from galleries

core/gallery/request.go

@@ -62,3 +62,15 @@ func (gm GalleryModels) FindByName(name string) *GalleryModel {
 	}
 	return nil
 }
+
+func (gm GalleryModels) Paginate(pageNum int, itemsNum int) GalleryModels {
+	start := (pageNum - 1) * itemsNum
+	end := start + itemsNum
+	if start > len(gm) {
+		start = len(gm)
+	}
+	if end > len(gm) {
+		end = len(gm)
+	}
+	return gm[start:end]
+}

core/http/app.go

@@ -130,7 +130,6 @@ func API(application *application.Application) (*fiber.App, error) {
 				return metricsService.Shutdown()
 			})
 		}
-
 	}
 	// Health Checks should always be exempt from auth, so register these first
 	routes.HealthRoutes(router)
@@ -140,6 +139,28 @@ func API(application *application.Application) (*fiber.App, error) {
 		return nil, fmt.Errorf("failed to create key auth config: %w", err)
 	}
 
+	httpFS := http.FS(embedDirStatic)
+
+	router.Use(favicon.New(favicon.Config{
+		URL:        "/favicon.ico",
+		FileSystem: httpFS,
+		File:       "static/favicon.ico",
+	}))
+
+	router.Use("/static", filesystem.New(filesystem.Config{
+		Root:       httpFS,
+		PathPrefix: "static",
+		Browse:     true,
+	}))
+
+	if application.ApplicationConfig().ImageDir != "" {
+		router.Static("/generated-images", application.ApplicationConfig().ImageDir)
+	}
+
+	if application.ApplicationConfig().AudioDir != "" {
+		router.Static("/generated-audio", application.ApplicationConfig().AudioDir)
+	}
+
 	// Auth is applied to _all_ endpoints. No exceptions. Filtering out endpoints to bypass is the role of the Filter property of the KeyAuth Configuration
 	router.Use(v2keyauth.New(*kaConfig))
 
@@ -167,27 +188,15 @@ func API(application *application.Application) (*fiber.App, error) {
 	galleryService := services.NewGalleryService(application.ApplicationConfig())
 	galleryService.Start(application.ApplicationConfig().Context, application.BackendLoader())
 
-	routes.RegisterElevenLabsRoutes(router, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig())
-	routes.RegisterLocalAIRoutes(router, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig(), galleryService)
-	routes.RegisterOpenAIRoutes(router, application)
+	requestExtractor := middleware.NewRequestExtractor(application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig())
+
+	routes.RegisterElevenLabsRoutes(router, requestExtractor, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig())
+	routes.RegisterLocalAIRoutes(router, requestExtractor, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig(), galleryService)
+	routes.RegisterOpenAIRoutes(router, requestExtractor, application)
 	if !application.ApplicationConfig().DisableWebUI {
 		routes.RegisterUIRoutes(router, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig(), galleryService)
 	}
-	routes.RegisterJINARoutes(router, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig())
-
-	httpFS := http.FS(embedDirStatic)
-
-	router.Use(favicon.New(favicon.Config{
-		URL:        "/favicon.ico",
-		FileSystem: httpFS,
-		File:       "static/favicon.ico",
-	}))
-
-	router.Use("/static", filesystem.New(filesystem.Config{
-		Root:       httpFS,
-		PathPrefix: "static",
-		Browse:     true,
-	}))
+	routes.RegisterJINARoutes(router, requestExtractor, application.BackendLoader(), application.ModelLoader(), application.ApplicationConfig())
 
 	// Define a custom 404 handler
 	// Note: keep this at the bottom!

core/http/ctx/fiber.go

@@ -1,47 +0,0 @@
-package fiberContext
-
-import (
-	"fmt"
-	"strings"
-
-	"github.com/gofiber/fiber/v2"
-	"github.com/mudler/LocalAI/core/config"
-	"github.com/mudler/LocalAI/core/services"
-	"github.com/mudler/LocalAI/pkg/model"
-	"github.com/rs/zerolog/log"
-)
-
-// ModelFromContext returns the model from the context
-// If no model is specified, it will take the first available
-// Takes a model string as input which should be the one received from the user request.
-// It returns the model name resolved from the context and an error if any.
-func ModelFromContext(ctx *fiber.Ctx, cl *config.BackendConfigLoader, loader *model.ModelLoader, modelInput string, firstModel bool) (string, error) {
-	if ctx.Params("model") != "" {
-		modelInput = ctx.Params("model")
-	}
-	if ctx.Query("model") != "" {
-		modelInput = ctx.Query("model")
-	}
-	// Set model from bearer token, if available
-	bearer := strings.TrimLeft(ctx.Get("authorization"), "Bear ") // Reduced duplicate characters of Bearer
-	bearerExists := bearer != "" && loader.ExistsInModelPath(bearer)
-
-	// If no model was specified, take the first available
-	if modelInput == "" && !bearerExists && firstModel {
-		models, _ := services.ListModels(cl, loader, config.NoFilterFn, services.SKIP_IF_CONFIGURED)
-		if len(models) > 0 {
-			modelInput = models[0]
-			log.Debug().Msgf("No model specified, using: %s", modelInput)
-		} else {
-			log.Debug().Msgf("No model specified, returning error")
-			return "", fmt.Errorf("no model specified")
-		}
-	}
-
-	// If a model is found in bearer token takes precedence
-	if bearerExists {
-		log.Debug().Msgf("Using model from bearer token: %s", bearer)
-		modelInput = bearer
-	}
-	return modelInput, nil
-}

core/http/elements/buttons.go

@@ -13,7 +13,7 @@ func installButton(galleryName string) elem.Node {
 		attrs.Props{
 			"data-twe-ripple-init":  "",
 			"data-twe-ripple-color": "light",
-			"class":                 "float-right inline-block rounded bg-primary px-6 pb-2.5 mb-3 pt-2.5 text-xs font-medium uppercase leading-normal text-white shadow-primary-3 transition duration-150 ease-in-out hover:bg-primary-accent-300 hover:shadow-primary-2 focus:bg-primary-accent-300 focus:shadow-primary-2 focus:outline-none focus:ring-0 active:bg-primary-600 active:shadow-primary-2 dark:shadow-black/30 dark:hover:shadow-dark-strong dark:focus:shadow-dark-strong dark:active:shadow-dark-strong",
+			"class":                 "float-right inline-flex items-center rounded-lg bg-blue-600 hover:bg-blue-700 px-4 py-2 text-sm font-medium text-white transition duration-300 ease-in-out shadow hover:shadow-lg",
 			"hx-swap":               "outerHTML",
 			// post the Model ID as param
 			"hx-post": "browse/install/model/" + galleryName,
@@ -52,7 +52,7 @@ func infoButton(m *gallery.GalleryModel) elem.Node {
 		attrs.Props{
 			"data-twe-ripple-init":  "",
 			"data-twe-ripple-color": "light",
-			"class":                 "float-left inline-block rounded bg-primary px-6 pb-2.5 mb-3 pt-2.5 text-xs font-medium uppercase leading-normal text-white shadow-primary-3 transition duration-150 ease-in-out hover:bg-primary-accent-300 hover:shadow-primary-2 focus:bg-primary-accent-300 focus:shadow-primary-2 focus:outline-none focus:ring-0 active:bg-primary-600 active:shadow-primary-2 dark:shadow-black/30 dark:hover:shadow-dark-strong dark:focus:shadow-dark-strong dark:active:shadow-dark-strong",
+			"class":                 "inline-flex items-center rounded-lg bg-gray-700 hover:bg-gray-600 px-4 py-2 text-sm font-medium text-white transition duration-300 ease-in-out",
 			"data-modal-target":     modalName(m),
 			"data-modal-toggle":     modalName(m),
 		},

core/http/elements/gallery.go

@@ -17,7 +17,7 @@ const (
 func cardSpan(text, icon string) elem.Node {
 	return elem.Span(
 		attrs.Props{
-			"class": "inline-block bg-gray-200 rounded-full px-3 py-1 text-sm font-semibold text-gray-700 mr-2 mb-2",
+			"class": "inline-flex items-center px-3 py-1 rounded-lg text-xs font-medium bg-gray-700/70 text-gray-300 border border-gray-600/50 mr-2 mb-2",
 		},
 		elem.I(attrs.Props{
 			"class": icon + " pr-2",
@@ -39,19 +39,20 @@ func searchableElement(text, icon string) elem.Node {
 		),
 		elem.Span(
 			attrs.Props{
-				"class": "inline-block bg-gray-200 rounded-full px-3 py-1 text-sm font-semibold text-gray-700 mr-2 mb-2 hover:bg-gray-300 hover:shadow-gray-2",
+				"class": "inline-flex items-center text-xs px-3 py-1 rounded-full bg-gray-700/60 text-gray-300 border border-gray-600/50 hover:bg-gray-600 hover:text-gray-100 transition duration-200 ease-in-out",
 			},
 			elem.A(
 				attrs.Props{
 					//	"name":      "search",
 					//	"value":     text,
 					//"class":     "inline-block bg-gray-200 rounded-full px-3 py-1 text-sm font-semibold text-gray-700 mr-2 mb-2",
-					"href":      "#!",
-					"hx-post":   "browse/search/models",
-					"hx-target": "#search-results",
+					//"href":      "#!",
+					"href": "browse?term=" + text,
+					//"hx-post":   "browse/search/models",
+					//"hx-target": "#search-results",
 					// TODO: this doesn't work
 					//	"hx-vals":      `{ \"search\": \"` + text + `\" }`,
-					"hx-indicator": ".htmx-indicator",
+					//"hx-indicator": ".htmx-indicator",
 				},
 				elem.I(attrs.Props{
 					"class": icon + " pr-2",
@@ -101,7 +102,7 @@ func modalName(m *gallery.GalleryModel) string {
 	return m.Name + "-modal"
 }
 
-func modelDescription(m *gallery.GalleryModel) elem.Node {
+func modelModal(m *gallery.GalleryModel) elem.Node {
 	urls := []elem.Node{}
 	for _, url := range m.URLs {
 		urls = append(urls,
@@ -116,6 +117,125 @@ func modelDescription(m *gallery.GalleryModel) elem.Node {
 		)
 	}
 
+	return elem.Div(
+		attrs.Props{
+			"id":          modalName(m),
+			"tabindex":    "-1",
+			"aria-hidden": "true",
+			"class":       "hidden overflow-y-auto overflow-x-hidden fixed top-0 right-0 left-0 z-50 justify-center items-center w-full md:inset-0 h-[calc(100%-1rem)] max-h-full",
+		},
+		elem.Div(
+			attrs.Props{
+				"class": "relative p-4 w-full max-w-2xl max-h-full",
+			},
+			elem.Div(
+				attrs.Props{
+					"class": "relative p-4 w-full max-w-2xl max-h-full bg-white rounded-lg shadow dark:bg-gray-700",
+				},
+				// header
+				elem.Div(
+					attrs.Props{
+						"class": "flex items-center justify-between p-4 md:p-5 border-b rounded-t dark:border-gray-600",
+					},
+					elem.H3(
+						attrs.Props{
+							"class": "text-xl font-semibold text-gray-900 dark:text-white",
+						},
+						elem.Text(bluemonday.StrictPolicy().Sanitize(m.Name)),
+					),
+					elem.Button( // close button
+						attrs.Props{
+							"class":           "text-gray-400 bg-transparent hover:bg-gray-200 hover:text-gray-900 rounded-lg text-sm w-8 h-8 ms-auto inline-flex justify-center items-center dark:hover:bg-gray-600 dark:hover:text-white",
+							"data-modal-hide": modalName(m),
+						},
+						elem.Raw(
+							`<svg class="w-3 h-3" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" fill="none" viewBox="0 0 14 14">
+							<path stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="m1 1 6 6m0 0 6 6M7 7l6-6M7 7l-6 6"/>
+						</svg>`,
+						),
+						elem.Span(
+							attrs.Props{
+								"class": "sr-only",
+							},
+							elem.Text("Close modal"),
+						),
+					),
+				),
+				// body
+				elem.Div(
+					attrs.Props{
+						"class": "p-4 md:p-5 space-y-4",
+					},
+					elem.Div(
+						attrs.Props{
+							"class": "flex justify-center items-center",
+						},
+						elem.Img(attrs.Props{
+							//	"class": "rounded-t-lg object-fit object-center h-96",
+							"class":   "lazy rounded-t-lg max-h-48 max-w-96 object-cover mt-3 entered loaded",
+							"src":     m.Icon,
+							"loading": "lazy",
+						}),
+					),
+					elem.P(
+						attrs.Props{
+							"class": "text-base leading-relaxed text-gray-500 dark:text-gray-400",
+						},
+						elem.Text(bluemonday.StrictPolicy().Sanitize(m.Description)),
+					),
+					elem.Hr(
+						attrs.Props{},
+					),
+					elem.P(
+						attrs.Props{
+							"class": "text-sm font-semibold text-gray-900 dark:text-white",
+						},
+						elem.Text("Links"),
+					),
+					elem.Ul(
+						attrs.Props{},
+						urls...,
+					),
+					elem.If(
+						len(m.Tags) > 0,
+						elem.Div(
+							attrs.Props{},
+							elem.P(
+								attrs.Props{
+									"class": "text-sm mb-5 font-semibold text-gray-900 dark:text-white",
+								},
+								elem.Text("Tags"),
+							),
+							elem.Div(
+								attrs.Props{
+									"class": "flex flex-row flex-wrap content-center",
+								},
+								tagsNodes...,
+							),
+						),
+						elem.Div(attrs.Props{}),
+					),
+				),
+				// Footer
+				elem.Div(
+					attrs.Props{
+						"class": "flex items-center p-4 md:p-5 border-t border-gray-200 rounded-b dark:border-gray-600",
+					},
+					elem.Button(
+						attrs.Props{
+							"data-modal-hide": modalName(m),
+							"class":           "py-2.5 px-5 ms-3 text-sm font-medium text-gray-900 focus:outline-none bg-white rounded-lg border border-gray-200 hover:bg-gray-100 hover:text-blue-700 focus:z-10 focus:ring-4 focus:ring-gray-100 dark:focus:ring-gray-700 dark:bg-gray-800 dark:text-gray-400 dark:border-gray-600 dark:hover:text-white dark:hover:bg-gray-700",
+						},
+						elem.Text("Close"),
+					),
+				),
+			),
+		),
+	)
+
+}
+
+func modelDescription(m *gallery.GalleryModel) elem.Node {
 	return elem.Div(
 		attrs.Props{
 			"class": "p-6 text-surface dark:text-white",
@@ -132,122 +252,6 @@ func modelDescription(m *gallery.GalleryModel) elem.Node {
 			},
 			elem.Text(bluemonday.StrictPolicy().Sanitize(m.Description)),
 		),
-
-		elem.Div(
-			attrs.Props{
-				"id":          modalName(m),
-				"tabindex":    "-1",
-				"aria-hidden": "true",
-				"class":       "hidden overflow-y-auto overflow-x-hidden fixed top-0 right-0 left-0 z-50 justify-center items-center w-full md:inset-0 h-[calc(100%-1rem)] max-h-full",
-			},
-			elem.Div(
-				attrs.Props{
-					"class": "relative p-4 w-full max-w-2xl max-h-full",
-				},
-				elem.Div(
-					attrs.Props{
-						"class": "relative p-4 w-full max-w-2xl max-h-full bg-white rounded-lg shadow dark:bg-gray-700",
-					},
-					// header
-					elem.Div(
-						attrs.Props{
-							"class": "flex items-center justify-between p-4 md:p-5 border-b rounded-t dark:border-gray-600",
-						},
-						elem.H3(
-							attrs.Props{
-								"class": "text-xl font-semibold text-gray-900 dark:text-white",
-							},
-							elem.Text(bluemonday.StrictPolicy().Sanitize(m.Name)),
-						),
-						elem.Button( // close button
-							attrs.Props{
-								"class":           "text-gray-400 bg-transparent hover:bg-gray-200 hover:text-gray-900 rounded-lg text-sm w-8 h-8 ms-auto inline-flex justify-center items-center dark:hover:bg-gray-600 dark:hover:text-white",
-								"data-modal-hide": modalName(m),
-							},
-							elem.Raw(
-								`<svg class="w-3 h-3" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" fill="none" viewBox="0 0 14 14">
-									<path stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="m1 1 6 6m0 0 6 6M7 7l6-6M7 7l-6 6"/>
-								</svg>`,
-							),
-							elem.Span(
-								attrs.Props{
-									"class": "sr-only",
-								},
-								elem.Text("Close modal"),
-							),
-						),
-					),
-					// body
-					elem.Div(
-						attrs.Props{
-							"class": "p-4 md:p-5 space-y-4",
-						},
-						elem.Div(
-							attrs.Props{
-								"class": "flex justify-center items-center",
-							},
-							elem.Img(attrs.Props{
-								//	"class": "rounded-t-lg object-fit object-center h-96",
-								"class":   "lazy rounded-t-lg max-h-48 max-w-96 object-cover mt-3 entered loaded",
-								"src":     m.Icon,
-								"loading": "lazy",
-							}),
-						),
-						elem.P(
-							attrs.Props{
-								"class": "text-base leading-relaxed text-gray-500 dark:text-gray-400",
-							},
-							elem.Text(bluemonday.StrictPolicy().Sanitize(m.Description)),
-						),
-						elem.Hr(
-							attrs.Props{},
-						),
-						elem.P(
-							attrs.Props{
-								"class": "text-sm font-semibold text-gray-900 dark:text-white",
-							},
-							elem.Text("Links"),
-						),
-						elem.Ul(
-							attrs.Props{},
-							urls...,
-						),
-						elem.If(
-							len(m.Tags) > 0,
-							elem.Div(
-								attrs.Props{},
-								elem.P(
-									attrs.Props{
-										"class": "text-sm mb-5 font-semibold text-gray-900 dark:text-white",
-									},
-									elem.Text("Tags"),
-								),
-								elem.Div(
-									attrs.Props{
-										"class": "flex flex-row flex-wrap content-center",
-									},
-									tagsNodes...,
-								),
-							),
-							elem.Div(attrs.Props{}),
-						),
-					),
-					// Footer
-					elem.Div(
-						attrs.Props{
-							"class": "flex items-center p-4 md:p-5 border-t border-gray-200 rounded-b dark:border-gray-600",
-						},
-						elem.Button(
-							attrs.Props{
-								"data-modal-hide": modalName(m),
-								"class":           "py-2.5 px-5 ms-3 text-sm font-medium text-gray-900 focus:outline-none bg-white rounded-lg border border-gray-200 hover:bg-gray-100 hover:text-blue-700 focus:z-10 focus:ring-4 focus:ring-gray-100 dark:focus:ring-gray-700 dark:bg-gray-800 dark:text-gray-400 dark:border-gray-600 dark:hover:text-white dark:hover:bg-gray-700",
-							},
-							elem.Text("Close"),
-						),
-					),
-				),
-			),
-		),
 	)
 }
 
@@ -397,7 +401,7 @@ func ListModels(models []*gallery.GalleryModel, processTracker ProcessTracker, g
 		modelsElements = append(modelsElements,
 			elem.Div(
 				attrs.Props{
-					"class": " me-4 mb-2 block rounded-lg bg-white shadow-secondary-1  dark:bg-gray-800 dark:bg-surface-dark dark:text-white text-surface pb-2",
+					"class": " me-4 mb-2 block rounded-lg bg-white shadow-secondary-1  dark:bg-gray-800 dark:bg-surface-dark dark:text-white text-surface pb-2 bg-gray-800/90 border border-gray-700/50 rounded-xl overflow-hidden transition-all duration-300 hover:shadow-lg hover:shadow-blue-900/20 hover:-translate-y-1 hover:border-blue-700/50",
 				},
 				elem.Div(
 					attrs.Props{
@@ -406,6 +410,7 @@ func ListModels(models []*gallery.GalleryModel, processTracker ProcessTracker, g
 					elems...,
 				),
 			),
+			modelModal(m),
 		)
 	}
 

core/http/elements/p2p.go

@@ -2,6 +2,7 @@ package elements
 
 import (
 	"fmt"
+	"time"
 
 	"github.com/chasefleming/elem-go"
 	"github.com/chasefleming/elem-go/attrs"
@@ -18,19 +19,6 @@ func renderElements(n []elem.Node) string {
 }
 
 func P2PNodeStats(nodes []p2p.NodeData) string {
-	/*
-	   <div class="bg-gray-800 p-6 rounded-lg shadow-lg text-left">
-	                       <p class="text-xl font-semibold text-gray-200">Total Workers Detected: {{ len .Nodes }}</p>
-	                       {{ $online := 0 }}
-	                       {{ range .Nodes }}
-	                           {{ if .IsOnline }}
-	                               {{ $online = add $online 1 }}
-	                           {{ end }}
-	                       {{ end }}
-	                       <p class="text-xl font-semibold text-gray-200">Total Online Workers: {{$online}}</p>
-	                   </div>
-	*/
-
 	online := 0
 	for _, n := range nodes {
 		if n.IsOnline() {
@@ -38,27 +26,21 @@ func P2PNodeStats(nodes []p2p.NodeData) string {
 		}
 	}
 
-	class := "text-green-500"
+	class := "text-blue-400"
 	if online == 0 {
-		class = "text-red-500"
+		class = "text-red-400"
 	}
-	/*
-	   <i class="fas fa-circle animate-pulse text-green-500 ml-2 mr-1"></i>
-	*/
-	circle := elem.I(attrs.Props{
-		"class": "fas fa-circle animate-pulse " + class + " ml-2 mr-1",
-	})
+
 	nodesElements := []elem.Node{
 		elem.Span(
 			attrs.Props{
-				"class": class,
+				"class": class + " font-bold text-xl",
 			},
-			circle,
 			elem.Text(fmt.Sprintf("%d", online)),
 		),
 		elem.Span(
 			attrs.Props{
-				"class": "text-gray-200",
+				"class": "text-gray-300 text-xl",
 			},
 			elem.Text(fmt.Sprintf("/%d", len(nodes))),
 		),
@@ -68,77 +50,73 @@ func P2PNodeStats(nodes []p2p.NodeData) string {
 }
 
 func P2PNodeBoxes(nodes []p2p.NodeData) string {
-	/*
-			<div class="bg-gray-800 p-4 rounded-lg shadow-lg text-left">
-			<div class="flex items-center mb-2">
-				<i class="fas fa-desktop text-gray-400 mr-2"></i>
-				<span class="text-gray-200 font-semibold">{{.ID}}</span>
-			</div>
-			<p class="text-sm text-gray-400 mt-2 flex items-center">
-				Status:
-				<i class="fas fa-circle {{ if .IsOnline }}text-green-500{{ else }}text-red-500{{ end }} ml-2 mr-1"></i>
-				<span class="{{ if .IsOnline }}text-green-400{{ else }}text-red-400{{ end }}">
-					{{ if .IsOnline }}Online{{ else }}Offline{{ end }}
-				</span>
-			</p>
-		</div>
-	*/
-
 	nodesElements := []elem.Node{}
 
 	for _, n := range nodes {
+		nodeID := bluemonday.StrictPolicy().Sanitize(n.ID)
+
+		// Define status-specific classes
+		statusIconClass := "text-green-400"
+		statusText := "Online"
+		statusTextClass := "text-green-400"
+
+		if !n.IsOnline() {
+			statusIconClass = "text-red-400"
+			statusText = "Offline"
+			statusTextClass = "text-red-400"
+		}
 
 		nodesElements = append(nodesElements,
 			elem.Div(
 				attrs.Props{
-					"class": "bg-gray-700 p-6 rounded-lg shadow-lg text-left",
+					"class": "bg-gray-800/80 border border-gray-700/50 rounded-xl p-4 shadow-lg transition-all duration-300 hover:shadow-blue-900/20 hover:border-blue-700/50",
 				},
-				elem.P(
+				// Node ID and status indicator in top row
+				elem.Div(
 					attrs.Props{
-						"class": "text-sm text-gray-400 mt-2 flex",
+						"class": "flex items-center justify-between mb-3",
 					},
-					elem.I(
+					// Node ID with icon
+					elem.Div(
 						attrs.Props{
-							"class": "fas fa-desktop text-gray-400 mr-2",
+							"class": "flex items-center",
 						},
-					),
-					elem.Text("Name: "),
-					elem.Span(
-						attrs.Props{
-							"class": "text-gray-200 font-semibold ml-2 mr-1",
-						},
-						elem.Text(bluemonday.StrictPolicy().Sanitize(n.ID)),
-					),
-					elem.Text("Status: "),
-					elem.If(
-						n.IsOnline(),
 						elem.I(
 							attrs.Props{
-								"class": "fas fa-circle animate-pulse text-green-500 ml-2 mr-1",
+								"class": "fas fa-server text-blue-400 mr-2",
 							},
 						),
-						elem.I(
-							attrs.Props{
-								"class": "fas fa-circle animate-pulse text-red-500 ml-2 mr-1",
-							},
-						),
-					),
-					elem.If(
-						n.IsOnline(),
-						elem.Span(
-							attrs.Props{
-								"class": "text-green-400",
-							},
-
-							elem.Text("Online"),
-						),
 						elem.Span(
 							attrs.Props{
-								"class": "text-red-400",
+								"class": "text-white font-medium",
 							},
-							elem.Text("Offline"),
+							elem.Text(nodeID),
 						),
 					),
+					// Status indicator
+					elem.Div(
+						attrs.Props{
+							"class": "flex items-center",
+						},
+						elem.I(
+							attrs.Props{
+								"class": "fas fa-circle animate-pulse " + statusIconClass + " mr-1.5",
+							},
+						),
+						elem.Span(
+							attrs.Props{
+								"class": statusTextClass,
+							},
+							elem.Text(statusText),
+						),
+					),
+				),
+				// Bottom section with timestamp
+				elem.Div(
+					attrs.Props{
+						"class": "text-xs text-gray-400 pt-1 border-t border-gray-700/30",
+					},
+					elem.Text("Last updated: "+time.Now().UTC().Format("2006-01-02 15:04:05")),
 				),
 			))
 	}

core/http/endpoints/elevenlabs/soundgeneration.go

@@ -4,7 +4,7 @@ import (
 	"github.com/gofiber/fiber/v2"
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/rs/zerolog/log"
@@ -17,45 +17,21 @@ import (
 // @Router /v1/sound-generation [post]
 func SoundGenerationEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		input := new(schema.ElevenLabsSoundGenerationRequest)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.ElevenLabsSoundGenerationRequest)
+		if !ok || input.ModelID == "" {
+			return fiber.ErrBadRequest
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.ModelID, false)
-		if err != nil {
-			modelFile = input.ModelID
-			log.Warn().Str("ModelID", input.ModelID).Msg("Model not found in context")
+		cfg, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || cfg == nil {
+			return fiber.ErrBadRequest
 		}
 
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
-		if err != nil {
-			modelFile = input.ModelID
-			log.Warn().Str("Request ModelID", input.ModelID).Err(err).Msg("error during LoadBackendConfigFileByName, using request ModelID")
-		} else {
-			if input.ModelID != "" {
-				modelFile = input.ModelID
-			} else {
-				modelFile = cfg.Model
-			}
-		}
 		log.Debug().Str("modelFile", "modelFile").Str("backend", cfg.Backend).Msg("Sound Generation Request about to be sent to backend")
 
-		if input.Duration != nil {
-			log.Debug().Float32("duration", *input.Duration).Msg("duration set")
-		}
-		if input.Temperature != nil {
-			log.Debug().Float32("temperature", *input.Temperature).Msg("temperature set")
-		}
-
 		// TODO: Support uploading files?
-		filePath, _, err := backend.SoundGeneration(modelFile, input.Text, input.Duration, input.Temperature, input.DoSample, nil, nil, ml, appConfig, *cfg)
+		filePath, _, err := backend.SoundGeneration(input.Text, input.Duration, input.Temperature, input.DoSample, nil, nil, ml, appConfig, *cfg)
 		if err != nil {
 			return err
 		}

core/http/endpoints/elevenlabs/tts.go

@@ -3,7 +3,7 @@ package elevenlabs
 import (
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/pkg/model"
 
 	"github.com/gofiber/fiber/v2"
@@ -20,39 +20,21 @@ import (
 func TTSEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
 
-		input := new(schema.ElevenLabsTTSRequest)
 		voiceID := c.Params("voice-id")
 
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.ElevenLabsTTSRequest)
+		if !ok || input.ModelID == "" {
+			return fiber.ErrBadRequest
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.ModelID, false)
-		if err != nil {
-			modelFile = input.ModelID
-			log.Warn().Msgf("Model not found in context: %s", input.ModelID)
+		cfg, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || cfg == nil {
+			return fiber.ErrBadRequest
 		}
 
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
-		if err != nil {
-			modelFile = input.ModelID
-			log.Warn().Msgf("Model not found in context: %s", input.ModelID)
-		} else {
-			if input.ModelID != "" {
-				modelFile = input.ModelID
-			} else {
-				modelFile = cfg.Model
-			}
-		}
-		log.Debug().Msgf("Request for model: %s", modelFile)
+		log.Debug().Str("modelName", input.ModelID).Msg("elevenlabs TTS request recieved")
 
-		filePath, _, err := backend.ModelTTS(cfg.Backend, input.Text, modelFile, "", voiceID, ml, appConfig, *cfg)
+		filePath, _, err := backend.ModelTTS(input.Text, voiceID, input.LanguageCode, ml, appConfig, *cfg)
 		if err != nil {
 			return err
 		}

core/http/endpoints/jina/rerank.go

@@ -3,9 +3,9 @@ package jina
 import (
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/middleware"
 
 	"github.com/gofiber/fiber/v2"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/grpc/proto"
 	"github.com/mudler/LocalAI/pkg/model"
@@ -19,58 +19,32 @@ import (
 // @Router /v1/rerank [post]
 func JINARerankEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		req := new(schema.JINARerankRequest)
-		if err := c.BodyParser(req); err != nil {
-			return c.Status(fiber.StatusBadRequest).JSON(fiber.Map{
-				"error": "Cannot parse JSON",
-			})
+
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.JINARerankRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
 		}
 
-		input := new(schema.TTSRequest)
-
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+		cfg, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || cfg == nil {
+			return fiber.ErrBadRequest
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
-		if err != nil {
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
-		}
-
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
-		if err != nil {
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
-		} else {
-			modelFile = cfg.Model
-		}
-
-		log.Debug().Msgf("Request for model: %s", modelFile)
-
-		if input.Backend != "" {
-			cfg.Backend = input.Backend
-		}
+		log.Debug().Str("model", input.Model).Msg("JINA Rerank Request recieved")
 
 		request := &proto.RerankRequest{
-			Query:     req.Query,
-			TopN:      int32(req.TopN),
-			Documents: req.Documents,
+			Query:     input.Query,
+			TopN:      int32(input.TopN),
+			Documents: input.Documents,
 		}
 
-		results, err := backend.Rerank(modelFile, request, ml, appConfig, *cfg)
+		results, err := backend.Rerank(request, ml, appConfig, *cfg)
 		if err != nil {
 			return err
 		}
 
 		response := &schema.JINARerankResponse{
-			Model: req.Model,
+			Model: input.Model,
 		}
 
 		for _, r := range results.Results {

core/http/endpoints/localai/get_token_metrics.go

@@ -4,13 +4,15 @@ import (
 	"github.com/gofiber/fiber/v2"
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/rs/zerolog/log"
 
 	"github.com/mudler/LocalAI/pkg/model"
 )
 
+// TODO: This is not yet in use. Needs middleware rework, since it is not referenced.
+
 // TokenMetricsEndpoint is an endpoint to get TokensProcessed Per Second for Active SlotID
 //
 //	@Summary	Get TokenMetrics for Active Slot.
@@ -29,18 +31,13 @@ func TokenMetricsEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader,
 			return err
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
-		if err != nil {
+		modelFile, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_NAME).(string)
+		if !ok || modelFile != "" {
 			modelFile = input.Model
 			log.Warn().Msgf("Model not found in context: %s", input.Model)
 		}
 
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
+		cfg, err := cl.LoadBackendConfigFileByNameDefaultOptions(modelFile, appConfig)
 
 		if err != nil {
 			log.Err(err)

core/http/endpoints/localai/tokenize.go

@@ -4,10 +4,9 @@ import (
 	"github.com/gofiber/fiber/v2"
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/model"
-	"github.com/rs/zerolog/log"
 )
 
 // TokenizeEndpoint exposes a REST API to tokenize the content
@@ -16,42 +15,21 @@ import (
 // @Success 200 {object} schema.TokenizeResponse "Response"
 // @Router /v1/tokenize [post]
 func TokenizeEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-
-		input := new(schema.TokenizeRequest)
-
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+	return func(ctx *fiber.Ctx) error {
+		input, ok := ctx.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.TokenizeRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
-		if err != nil {
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		cfg, ok := ctx.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || cfg == nil {
+			return fiber.ErrBadRequest
 		}
 
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
-
-		if err != nil {
-			log.Err(err)
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
-		} else {
-			modelFile = cfg.Model
-		}
-		log.Debug().Msgf("Request for model: %s", modelFile)
-
 		tokenResponse, err := backend.ModelTokenize(input.Content, ml, *cfg, appConfig)
 		if err != nil {
 			return err
 		}
-
-		return c.JSON(tokenResponse)
+		return ctx.JSON(tokenResponse)
 	}
 }

core/http/endpoints/localai/tts.go

@@ -3,7 +3,7 @@ package localai
 import (
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/pkg/model"
 
 	"github.com/gofiber/fiber/v2"
@@ -24,37 +24,24 @@ import (
 //		@Router		/tts [post]
 func TTSEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		input := new(schema.TTSRequest)
-
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.TTSRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
-		if err != nil {
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		cfg, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || cfg == nil {
+			return fiber.ErrBadRequest
 		}
 
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
+		log.Debug().Str("model", input.Model).Msg("LocalAI TTS Request recieved")
 
-		if err != nil {
-			log.Err(err)
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
-		} else {
-			modelFile = cfg.Model
-		}
-		log.Debug().Msgf("Request for model: %s", modelFile)
-
-		if input.Backend != "" {
-			cfg.Backend = input.Backend
+		if cfg.Backend == "" {
+			if input.Backend != "" {
+				cfg.Backend = input.Backend
+			} else {
+				cfg.Backend = model.PiperBackend
+			}
 		}
 
 		if input.Language != "" {
@@ -65,7 +52,7 @@ func TTSEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfi
 			cfg.Voice = input.Voice
 		}
 
-		filePath, _, err := backend.ModelTTS(cfg.Backend, input.Input, modelFile, cfg.Voice, cfg.Language, ml, appConfig, *cfg)
+		filePath, _, err := backend.ModelTTS(input.Input, cfg.Voice, cfg.Language, ml, appConfig, *cfg)
 		if err != nil {
 			return err
 		}

core/http/endpoints/localai/vad.go

@@ -4,9 +4,8 @@ import (
 	"github.com/gofiber/fiber/v2"
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
-	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
-	"github.com/mudler/LocalAI/pkg/grpc/proto"
 	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/rs/zerolog/log"
 )
@@ -19,45 +18,20 @@ import (
 // @Router		/vad [post]
 func VADEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		input := new(schema.VADRequest)
-
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.VADRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
 		}
 
-		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
-		if err != nil {
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		cfg, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || cfg == nil {
+			return fiber.ErrBadRequest
 		}
 
-		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
-			config.LoadOptionDebug(appConfig.Debug),
-			config.LoadOptionThreads(appConfig.Threads),
-			config.LoadOptionContextSize(appConfig.ContextSize),
-			config.LoadOptionF16(appConfig.F16),
-		)
+		log.Debug().Str("model", input.Model).Msg("LocalAI VAD Request recieved")
 
-		if err != nil {
-			log.Err(err)
-			modelFile = input.Model
-			log.Warn().Msgf("Model not found in context: %s", input.Model)
-		} else {
-			modelFile = cfg.Model
-		}
-		log.Debug().Msgf("Request for model: %s", modelFile)
+		resp, err := backend.VAD(input, c.Context(), ml, appConfig, *cfg)
 
-		opts := backend.ModelOptions(*cfg, appConfig, model.WithBackendString(cfg.Backend), model.WithModel(modelFile))
-
-		vadModel, err := ml.Load(opts...)
-		if err != nil {
-			return err
-		}
-		req := proto.VADRequest{
-			Audio: input.Audio,
-		}
-		resp, err := vadModel.VAD(c.Context(), &req)
 		if err != nil {
 			return err
 		}

core/http/endpoints/openai/chat.go

@@ -5,18 +5,19 @@ import (
 	"bytes"
 	"encoding/json"
 	"fmt"
-	"strings"
 	"time"
 
 	"github.com/gofiber/fiber/v2"
 	"github.com/google/uuid"
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/functions"
+
+	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/mudler/LocalAI/pkg/templates"
 
-	model "github.com/mudler/LocalAI/pkg/model"
 	"github.com/rs/zerolog/log"
 	"github.com/valyala/fasthttp"
 )
@@ -174,26 +175,20 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 		textContentToReturn = ""
 		id = uuid.New().String()
 		created = int(time.Now().Unix())
-		// Set CorrelationID
-		correlationID := c.Get("X-Correlation-ID")
-		if len(strings.TrimSpace(correlationID)) == 0 {
-			correlationID = id
+		
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.OpenAIRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
 		}
-		c.Set("X-Correlation-ID", correlationID)
 
-		// Opt-in extra usage flag
 		extraUsage := c.Get("Extra-Usage", "") != ""
 
-		modelFile, input, err := readRequest(c, cl, ml, startupOptions, true)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
+		config, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || config == nil {
+			return fiber.ErrBadRequest
 		}
 
-		config, input, err := mergeRequestWithConfig(modelFile, input, cl, ml, startupOptions.Debug, startupOptions.Threads, startupOptions.ContextSize, startupOptions.F16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-		log.Debug().Msgf("Configuration read: %+v", config)
+		log.Debug().Msgf("Chat endpoint configuration read: %+v", config)
 
 		funcs := input.Functions
 		shouldUseFn := len(input.Functions) > 0 && config.ShouldUseFunctions()
@@ -543,7 +538,7 @@ func handleQuestion(config *config.BackendConfig, input *schema.OpenAIRequest, m
 		audios = append(audios, m.StringAudios...)
 	}
 
-	predFunc, err := backend.ModelInference(input.Context, prompt, input.Messages, images, videos, audios, ml, *config, o, nil)
+	predFunc, err := backend.ModelInference(input.Context, prompt, input.Messages, images, videos, audios, ml, config, o, nil)
 	if err != nil {
 		log.Error().Err(err).Msg("model inference failed")
 		return "", err

core/http/endpoints/openai/completion.go

@@ -10,12 +10,13 @@ import (
 
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/middleware"
 
 	"github.com/gofiber/fiber/v2"
 	"github.com/google/uuid"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/functions"
-	model "github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/mudler/LocalAI/pkg/templates"
 	"github.com/rs/zerolog/log"
 	"github.com/valyala/fasthttp"
@@ -27,10 +28,9 @@ import (
 // @Success 200 {object} schema.OpenAIResponse "Response"
 // @Router /v1/completions [post]
 func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluator *templates.Evaluator, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
-	id := uuid.New().String()
 	created := int(time.Now().Unix())
 
-	process := func(s string, req *schema.OpenAIRequest, config *config.BackendConfig, loader *model.ModelLoader, responses chan schema.OpenAIResponse, extraUsage bool) {
+	process := func(id string, s string, req *schema.OpenAIRequest, config *config.BackendConfig, loader *model.ModelLoader, responses chan schema.OpenAIResponse, extraUsage bool) {
 		ComputeChoices(req, s, config, appConfig, loader, func(s string, c *[]schema.Choice) {}, func(s string, tokenUsage backend.TokenUsage) bool {
 			usage := schema.OpenAIUsage{
 				PromptTokens:     tokenUsage.Prompt,
@@ -63,22 +63,18 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
 	}
 
 	return func(c *fiber.Ctx) error {
-		// Add Correlation
-		c.Set("X-Correlation-ID", id)
-
-		// Opt-in extra usage flag
+		// Handle Correlation
+		id := c.Get("X-Correlation-ID", uuid.New().String())
 		extraUsage := c.Get("Extra-Usage", "") != ""
 
-		modelFile, input, err := readRequest(c, cl, ml, appConfig, true)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.OpenAIRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
 		}
 
-		log.Debug().Msgf("`input`: %+v", input)
-
-		config, input, err := mergeRequestWithConfig(modelFile, input, cl, ml, appConfig.Debug, appConfig.Threads, appConfig.ContextSize, appConfig.F16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
+		config, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || config == nil {
+			return fiber.ErrBadRequest
 		}
 
 		if config.ResponseFormatMap != nil {
@@ -122,7 +118,7 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
 
 			responses := make(chan schema.OpenAIResponse)
 
-			go process(predInput, input, config, ml, responses, extraUsage)
+			go process(id, predInput, input, config, ml, responses, extraUsage)
 
 			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
 

core/http/endpoints/openai/edit.go

@@ -2,16 +2,17 @@ package openai
 
 import (
 	"encoding/json"
-	"fmt"
 	"time"
 
 	"github.com/mudler/LocalAI/core/backend"
 	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/middleware"
 
 	"github.com/gofiber/fiber/v2"
 	"github.com/google/uuid"
 	"github.com/mudler/LocalAI/core/schema"
-	model "github.com/mudler/LocalAI/pkg/model"
+
+	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/mudler/LocalAI/pkg/templates"
 
 	"github.com/rs/zerolog/log"
@@ -25,20 +26,21 @@ import (
 func EditEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluator *templates.Evaluator, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 
 	return func(c *fiber.Ctx) error {
+
+		input, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_LOCALAI_REQUEST).(*schema.OpenAIRequest)
+		if !ok || input.Model == "" {
+			return fiber.ErrBadRequest
+		}
 		// Opt-in extra usage flag
 		extraUsage := c.Get("Extra-Usage", "") != ""
 
-		modelFile, input, err := readRequest(c, cl, ml, appConfig, true)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
+		config, ok := c.Locals(middleware.CONTEXT_LOCALS_KEY_MODEL_CONFIG).(*config.BackendConfig)
+		if !ok || config == nil {
+			return fiber.ErrBadRequest
 		}
 
-		config, input, err := mergeRequestWithConfig(modelFile, input, cl, ml, appConfig.Debug, appConfig.Threads, appConfig.ContextSize, appConfig.F16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
+		log.Debug().Msgf("Edit Endpoint Input : %+v", input)
+		log.Debug().Msgf("Edit Endpoint Config: %+v", *config)
 
 		var result []schema.Choice
 		totalTokenUsage := backend.TokenUsage{}