chore(deps): bump llama.cpp to '10f2e81809bbb69ecfe64fc8b4686285f84b0c07'

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

.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'

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

@@ -6,9 +6,7 @@ BINARY_NAME=local-ai
 DETECT_LIBS?=true
 
 # llama.cpp versions
-GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
-GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=5598f475be3e31430fbe17ebb85654ec90dc201e
+CPPLLAMA_VERSION?=10f2e81809bbb69ecfe64fc8b4686285f84b0c07
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
@@ -24,7 +22,7 @@ BARKCPP_VERSION?=v1.0.0
 
 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=5eb15ef4d022bef4a391de4f5f6556e81fbb5024
+STABLEDIFFUSION_GGML_VERSION?=19d876ee300a055629926ff836489901f734f2b7
 
 ONNX_VERSION?=1.20.0
 ONNX_ARCH?=x64
@@ -151,7 +149,6 @@ ifeq ($(BUILD_TYPE),hipblas)
 	LD_LIBRARY_PATH ?= /opt/rocm/lib:/opt/rocm/llvm/lib
 	export CXX=$(ROCM_HOME)/llvm/bin/clang++
 	export CC=$(ROCM_HOME)/llvm/bin/clang
-	# llama-ggml has no hipblas support, so override it here.
 	export STABLE_BUILD_TYPE=
 	export GGML_HIP=1
 	GPU_TARGETS ?= gfx900,gfx906,gfx908,gfx940,gfx941,gfx942,gfx90a,gfx1030,gfx1031,gfx1100,gfx1101
@@ -188,7 +185,6 @@ ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx2
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx512
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-fallback
-ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-grpc
 ALL_GRPC_BACKENDS+=backend-assets/util/llama-cpp-rpc-server
 ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
@@ -222,19 +218,6 @@ endif
 
 all: help
 
-## go-llama.cpp
-sources/go-llama.cpp:
-	mkdir -p sources/go-llama.cpp
-	cd sources/go-llama.cpp && \
-	git init && \
-	git remote add origin $(GOLLAMA_REPO) && \
-	git fetch origin && \
-	git checkout $(GOLLAMA_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/go-llama.cpp/libbinding.a: sources/go-llama.cpp
-	$(MAKE) -C sources/go-llama.cpp BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a
-
 ## bark.cpp
 sources/bark.cpp:
 	git clone --recursive $(BARKCPP_REPO) sources/bark.cpp && \
@@ -310,19 +293,17 @@ sources/whisper.cpp:
 sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
 	cd sources/whisper.cpp && $(MAKE) libwhisper.a libggml.a
 
-get-sources: sources/go-llama.cpp sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp backend/cpp/llama/llama.cpp
+get-sources: sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp backend/cpp/llama/llama.cpp
 
 replace:
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
 	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama.cpp
 
 dropreplace:
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp/bindings/go
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-piper
-	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-llama.cpp
 
 prepare-sources: get-sources replace
 	$(GOCMD) mod download
@@ -330,7 +311,6 @@ prepare-sources: get-sources replace
 ## GENERIC
 rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
-	$(MAKE) -C sources/go-llama.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-piper clean
 	$(MAKE) build
@@ -434,7 +414,7 @@ run: prepare ## run local-ai
 test-models/testmodel.ggml:
 	mkdir test-models
 	mkdir test-dir
-	wget -q https://huggingface.co/TheBloke/orca_mini_3B-GGML/resolve/main/orca-mini-3b.ggmlv3.q4_0.bin -O test-models/testmodel.ggml
+	wget -q https://huggingface.co/RichardErkhov/Qwen_-_Qwen2-1.5B-Instruct-gguf/resolve/main/Qwen2-1.5B-Instruct.Q2_K.gguf -O test-models/testmodel.ggml
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
@@ -449,8 +429,7 @@ test: prepare test-models/testmodel.ggml grpcs
 	export GO_TAGS="tts debug"
 	$(MAKE) prepare-test
 	HUGGINGFACE_GRPC=$(abspath ./)/backend/python/transformers/run.sh TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!llama && !llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
-	$(MAKE) test-llama
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
 	$(MAKE) test-llama-gguf
 	$(MAKE) test-tts
 	$(MAKE) test-stablediffusion
@@ -479,10 +458,6 @@ teardown-e2e:
 	rm -rf $(TEST_DIR) || true
 	docker stop $$(docker ps -q --filter ancestor=localai-tests)
 
-test-llama: prepare-test
-	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
-
 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
 	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
@@ -760,13 +735,6 @@ backend-assets/util/llama-cpp-rpc-server: backend-assets/grpc/llama-cpp-grpc
 	mkdir -p backend-assets/util/
 	cp -rf backend/cpp/llama-grpc/llama.cpp/build/bin/rpc-server backend-assets/util/llama-cpp-rpc-server
 
-backend-assets/grpc/llama-ggml: sources/go-llama.cpp sources/go-llama.cpp/libbinding.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama.cpp LIBRARY_PATH=$(CURDIR)/sources/go-llama.cpp \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-ggml ./backend/go/llm/llama-ggml/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/llama-ggml
-endif
-
 backend-assets/grpc/bark-cpp: backend/go/bark/libbark.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/bark/ LIBRARY_PATH=$(CURDIR)/backend/go/bark/ \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark/

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/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/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/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/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

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

@@ -469,7 +469,7 @@ struct llama_server_context
     bool has_eos_token      = true;
 
     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
 
@@ -709,7 +709,7 @@ 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 +1155,14 @@ 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 +1350,7 @@ 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 +1372,38 @@ 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 +1663,17 @@ 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 +2024,7 @@ 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;
@@ -2385,12 +2421,12 @@ 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 },
             });
         }
     }

backend/cpp/llama/patches/01-llava.patch

@@ -1,13 +1,13 @@
 diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp
-index 3cd0d2fa..6c5e811a 100644
+index 7f892beb..0517e529 100644
 --- a/examples/llava/clip.cpp
 +++ b/examples/llava/clip.cpp
-@@ -2608,7 +2608,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
-                 struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
+@@ -2766,7 +2766,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
+                 int patch_offset = ctx->has_class_embedding ? 1 : 0;
                  int* patches_data = (int*)malloc(ggml_nbytes(patches));
                  for (int i = 0; i < num_patches; i++) {
--                    patches_data[i] = i + 1;
-+                    patches_data[i] = i;
+-                    patches_data[i] = i + patch_offset;
++                    patches_data[i] = i + 1;
                  }
                  ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
                  free(patches_data);

backend/cpp/llama/prepare.sh

@@ -1,5 +1,7 @@
 #!/bin/bash
 
+set -e
+
 ## Patches
 ## Apply patches from the `patches` directory
 for patch in $(ls patches); do

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/go/llm/llama-ggml/llama.go

@@ -1,204 +0,0 @@
-package main
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/go-llama.cpp"
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type LLM struct {
-	base.SingleThread
-
-	llama *llama.LLama
-}
-
-func (llm *LLM) Load(opts *pb.ModelOptions) error {
-	ropeFreqBase := float32(10000)
-	ropeFreqScale := float32(1)
-
-	if opts.RopeFreqBase != 0 {
-		ropeFreqBase = opts.RopeFreqBase
-	}
-	if opts.RopeFreqScale != 0 {
-		ropeFreqScale = opts.RopeFreqScale
-	}
-
-	llamaOpts := []llama.ModelOption{
-		llama.WithRopeFreqBase(ropeFreqBase),
-		llama.WithRopeFreqScale(ropeFreqScale),
-	}
-
-	if opts.NGQA != 0 {
-		llamaOpts = append(llamaOpts, llama.WithGQA(int(opts.NGQA)))
-	}
-
-	if opts.RMSNormEps != 0 {
-		llamaOpts = append(llamaOpts, llama.WithRMSNormEPS(opts.RMSNormEps))
-	}
-
-	if opts.ContextSize != 0 {
-		llamaOpts = append(llamaOpts, llama.SetContext(int(opts.ContextSize)))
-	}
-	if opts.F16Memory {
-		llamaOpts = append(llamaOpts, llama.EnableF16Memory)
-	}
-	if opts.Embeddings {
-		llamaOpts = append(llamaOpts, llama.EnableEmbeddings)
-	}
-	if opts.NGPULayers != 0 {
-		llamaOpts = append(llamaOpts, llama.SetGPULayers(int(opts.NGPULayers)))
-	}
-
-	llamaOpts = append(llamaOpts, llama.SetMMap(opts.MMap))
-	llamaOpts = append(llamaOpts, llama.SetMainGPU(opts.MainGPU))
-	llamaOpts = append(llamaOpts, llama.SetTensorSplit(opts.TensorSplit))
-	if opts.NBatch != 0 {
-		llamaOpts = append(llamaOpts, llama.SetNBatch(int(opts.NBatch)))
-	} else {
-		llamaOpts = append(llamaOpts, llama.SetNBatch(512))
-	}
-
-	if opts.NUMA {
-		llamaOpts = append(llamaOpts, llama.EnableNUMA)
-	}
-
-	if opts.LowVRAM {
-		llamaOpts = append(llamaOpts, llama.EnabelLowVRAM)
-	}
-
-	model, err := llama.New(opts.ModelFile, llamaOpts...)
-	llm.llama = model
-
-	return err
-}
-
-func buildPredictOptions(opts *pb.PredictOptions) []llama.PredictOption {
-	ropeFreqBase := float32(10000)
-	ropeFreqScale := float32(1)
-
-	if opts.RopeFreqBase != 0 {
-		ropeFreqBase = opts.RopeFreqBase
-	}
-	if opts.RopeFreqScale != 0 {
-		ropeFreqScale = opts.RopeFreqScale
-	}
-	predictOptions := []llama.PredictOption{
-		llama.SetTemperature(opts.Temperature),
-		llama.SetTopP(opts.TopP),
-		llama.SetTopK(int(opts.TopK)),
-		llama.SetTokens(int(opts.Tokens)),
-		llama.SetThreads(int(opts.Threads)),
-		llama.WithGrammar(opts.Grammar),
-		llama.SetRopeFreqBase(ropeFreqBase),
-		llama.SetRopeFreqScale(ropeFreqScale),
-		llama.SetNegativePromptScale(opts.NegativePromptScale),
-		llama.SetNegativePrompt(opts.NegativePrompt),
-	}
-
-	if opts.PromptCacheAll {
-		predictOptions = append(predictOptions, llama.EnablePromptCacheAll)
-	}
-
-	if opts.PromptCacheRO {
-		predictOptions = append(predictOptions, llama.EnablePromptCacheRO)
-	}
-
-	// Expected absolute path
-	if opts.PromptCachePath != "" {
-		predictOptions = append(predictOptions, llama.SetPathPromptCache(opts.PromptCachePath))
-	}
-
-	if opts.Mirostat != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostat(int(opts.Mirostat)))
-	}
-
-	if opts.MirostatETA != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostatETA(opts.MirostatETA))
-	}
-
-	if opts.MirostatTAU != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostatTAU(opts.MirostatTAU))
-	}
-
-	if opts.Debug {
-		predictOptions = append(predictOptions, llama.Debug)
-	}
-
-	predictOptions = append(predictOptions, llama.SetStopWords(opts.StopPrompts...))
-
-	if opts.PresencePenalty != 0 {
-		predictOptions = append(predictOptions, llama.SetPenalty(opts.PresencePenalty))
-	}
-
-	if opts.NKeep != 0 {
-		predictOptions = append(predictOptions, llama.SetNKeep(int(opts.NKeep)))
-	}
-
-	if opts.Batch != 0 {
-		predictOptions = append(predictOptions, llama.SetBatch(int(opts.Batch)))
-	}
-
-	if opts.F16KV {
-		predictOptions = append(predictOptions, llama.EnableF16KV)
-	}
-
-	if opts.IgnoreEOS {
-		predictOptions = append(predictOptions, llama.IgnoreEOS)
-	}
-
-	if opts.Seed != 0 {
-		predictOptions = append(predictOptions, llama.SetSeed(int(opts.Seed)))
-	}
-
-	//predictOptions = append(predictOptions, llama.SetLogitBias(c.Seed))
-
-	predictOptions = append(predictOptions, llama.SetFrequencyPenalty(opts.FrequencyPenalty))
-	predictOptions = append(predictOptions, llama.SetMlock(opts.MLock))
-	predictOptions = append(predictOptions, llama.SetMemoryMap(opts.MMap))
-	predictOptions = append(predictOptions, llama.SetPredictionMainGPU(opts.MainGPU))
-	predictOptions = append(predictOptions, llama.SetPredictionTensorSplit(opts.TensorSplit))
-	predictOptions = append(predictOptions, llama.SetTailFreeSamplingZ(opts.TailFreeSamplingZ))
-	predictOptions = append(predictOptions, llama.SetTypicalP(opts.TypicalP))
-	return predictOptions
-}
-
-func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.llama.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	predictOptions := buildPredictOptions(opts)
-
-	predictOptions = append(predictOptions, llama.SetTokenCallback(func(token string) bool {
-		results <- token
-		return true
-	}))
-
-	go func() {
-		_, err := llm.llama.Predict(opts.Prompt, predictOptions...)
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		close(results)
-	}()
-
-	return nil
-}
-
-func (llm *LLM) Embeddings(opts *pb.PredictOptions) ([]float32, error) {
-	predictOptions := buildPredictOptions(opts)
-
-	if len(opts.EmbeddingTokens) > 0 {
-		tokens := []int{}
-		for _, t := range opts.EmbeddingTokens {
-			tokens = append(tokens, int(t))
-		}
-		return llm.llama.TokenEmbeddings(tokens, predictOptions...)
-	}
-
-	return llm.llama.Embeddings(opts.Embeddings, predictOptions...)
-}

backend/go/llm/llama-ggml/main.go

@@ -1,19 +0,0 @@
-package main
-
-import (
-	"flag"
-
-	grpc "github.com/mudler/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
-		panic(err)
-	}
-}

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/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
 }
 
@@ -287,7 +307,8 @@ func (cfg *BackendConfig) SetDefaults(opts ...ConfigLoaderOption) {
 	defaultTopP := 0.95
 	defaultTopK := 40
 	defaultTemp := 0.9
-	defaultMirostat := 2
+	// https://github.com/mudler/LocalAI/issues/2780
+	defaultMirostat := 0
 	defaultMirostatTAU := 5.0
 	defaultMirostatETA := 0.1
 	defaultTypicalP := 1.0
@@ -436,19 +457,21 @@ 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 {
@@ -463,10 +486,16 @@ 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
@@ -474,7 +503,7 @@ 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
 		}
@@ -548,5 +577,18 @@ 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/guesser.go

@@ -161,10 +161,11 @@ func guessDefaultsFromFile(cfg *BackendConfig, modelPath string) {
 	}
 
 	// We try to guess only if we don't have a template defined already
-	f, err := gguf.ParseGGUFFile(filepath.Join(modelPath, cfg.ModelFileName()))
+	guessPath := filepath.Join(modelPath, cfg.ModelFileName())
+	f, err := gguf.ParseGGUFFile(guessPath)
 	if err != nil {
 		// Only valid for gguf files
-		log.Debug().Msgf("guessDefaultsFromFile: %s", "not a GGUF file")
+		log.Debug().Str("filePath", guessPath).Msg("guessDefaultsFromFile: not a GGUF file")
 		return
 	}
 

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/models_test.go

@@ -48,8 +48,10 @@ var _ = Describe("Model test", func() {
 			defer os.RemoveAll(tempdir)
 
 			gallery := []GalleryModel{{
-				Name: "bert",
-				URL:  bertEmbeddingsURL,
+				Metadata: Metadata{
+					Name: "bert",
+					URL:  bertEmbeddingsURL,
+				},
 			}}
 			out, err := yaml.Marshal(gallery)
 			Expect(err).ToNot(HaveOccurred())

core/gallery/request.go

@@ -11,6 +11,14 @@ import (
 // It is used to install the model by resolving the URL and downloading the files.
 // The other fields are used to override the configuration of the model.
 type GalleryModel struct {
+	Metadata `json:",inline" yaml:",inline"`
+	// config_file is read in the situation where URL is blank - and therefore this is a base config.
+	ConfigFile map[string]interface{} `json:"config_file,omitempty" yaml:"config_file,omitempty"`
+	// Overrides are used to override the configuration of the model located at URL
+	Overrides map[string]interface{} `json:"overrides,omitempty" yaml:"overrides,omitempty"`
+}
+
+type Metadata struct {
 	URL         string   `json:"url,omitempty" yaml:"url,omitempty"`
 	Name        string   `json:"name,omitempty" yaml:"name,omitempty"`
 	Description string   `json:"description,omitempty"  yaml:"description,omitempty"`
@@ -18,10 +26,6 @@ type GalleryModel struct {
 	URLs        []string `json:"urls,omitempty" yaml:"urls,omitempty"`
 	Icon        string   `json:"icon,omitempty" yaml:"icon,omitempty"`
 	Tags        []string `json:"tags,omitempty" yaml:"tags,omitempty"`
-	// config_file is read in the situation where URL is blank - and therefore this is a base config.
-	ConfigFile map[string]interface{} `json:"config_file,omitempty" yaml:"config_file,omitempty"`
-	// Overrides are used to override the configuration of the model located at URL
-	Overrides map[string]interface{} `json:"overrides,omitempty" yaml:"overrides,omitempty"`
 	// AdditionalFiles are used to add additional files to the model
 	AdditionalFiles []File `json:"files,omitempty" yaml:"files,omitempty"`
 	// Gallery is a reference to the gallery which contains the model
@@ -58,3 +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/gallery/request_test.go

@@ -9,7 +9,11 @@ import (
 var _ = Describe("Gallery API tests", func() {
 	Context("requests", func() {
 		It("parses github with a branch", func() {
-			req := GalleryModel{URL: "github:go-skynet/model-gallery/gpt4all-j.yaml@main"}
+			req := GalleryModel{
+				Metadata: Metadata{
+					URL: "github:go-skynet/model-gallery/gpt4all-j.yaml@main",
+				},
+			}
 			e, err := GetGalleryConfigFromURL(req.URL, "")
 			Expect(err).ToNot(HaveOccurred())
 			Expect(e.Name).To(Equal("gpt4all-j"))

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/app_test.go

@@ -299,14 +299,18 @@ var _ = Describe("API test", func() {
 
 			g := []gallery.GalleryModel{
 				{
-					Name: "bert",
-					URL:  bertEmbeddingsURL,
+					Metadata: gallery.Metadata{
+						Name: "bert",
+						URL:  bertEmbeddingsURL,
+					},
 				},
 				{
-					Name:            "bert2",
-					URL:             bertEmbeddingsURL,
-					Overrides:       map[string]interface{}{"foo": "bar"},
-					AdditionalFiles: []gallery.File{{Filename: "foo.yaml", URI: bertEmbeddingsURL}},
+					Metadata: gallery.Metadata{
+						Name:            "bert2",
+						URL:             bertEmbeddingsURL,
+						AdditionalFiles: []gallery.File{{Filename: "foo.yaml", URI: bertEmbeddingsURL}},
+					},
+					Overrides: map[string]interface{}{"foo": "bar"},
 				},
 			}
 			out, err := yaml.Marshal(g)
@@ -522,77 +526,6 @@ var _ = Describe("API test", func() {
 				Expect(content["usage"]).To(ContainSubstring("You can test this model with curl like this"))
 			})
 
-			It("runs openllama(llama-ggml backend)", Label("llama"), func() {
-				if runtime.GOOS != "linux" {
-					Skip("test supported only on linux")
-				}
-				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
-					URL:       "github:go-skynet/model-gallery/openllama_3b.yaml",
-					Name:      "openllama_3b",
-					Overrides: map[string]interface{}{"backend": "llama-ggml", "mmap": true, "f16": true, "context_size": 128},
-				})
-
-				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
-
-				uuid := response["uuid"].(string)
-
-				Eventually(func() bool {
-					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
-					return response["processed"].(bool)
-				}, "360s", "10s").Should(Equal(true))
-
-				By("testing completion")
-				resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "openllama_3b", Prompt: "Count up to five: one, two, three, four, "})
-				Expect(err).ToNot(HaveOccurred())
-				Expect(len(resp.Choices)).To(Equal(1))
-				Expect(resp.Choices[0].Text).To(ContainSubstring("five"))
-
-				By("testing functions")
-				resp2, err := client.CreateChatCompletion(
-					context.TODO(),
-					openai.ChatCompletionRequest{
-						Model: "openllama_3b",
-						Messages: []openai.ChatCompletionMessage{
-							{
-								Role:    "user",
-								Content: "What is the weather like in San Francisco (celsius)?",
-							},
-						},
-						Functions: []openai.FunctionDefinition{
-							openai.FunctionDefinition{
-								Name:        "get_current_weather",
-								Description: "Get the current weather",
-								Parameters: jsonschema.Definition{
-									Type: jsonschema.Object,
-									Properties: map[string]jsonschema.Definition{
-										"location": {
-											Type:        jsonschema.String,
-											Description: "The city and state, e.g. San Francisco, CA",
-										},
-										"unit": {
-											Type: jsonschema.String,
-											Enum: []string{"celcius", "fahrenheit"},
-										},
-									},
-									Required: []string{"location"},
-								},
-							},
-						},
-					})
-				Expect(err).ToNot(HaveOccurred())
-				Expect(len(resp2.Choices)).To(Equal(1))
-				Expect(resp2.Choices[0].Message.FunctionCall).ToNot(BeNil())
-				Expect(resp2.Choices[0].Message.FunctionCall.Name).To(Equal("get_current_weather"), resp2.Choices[0].Message.FunctionCall.Name)
-
-				var res map[string]string
-				err = json.Unmarshal([]byte(resp2.Choices[0].Message.FunctionCall.Arguments), &res)
-				Expect(err).ToNot(HaveOccurred())
-				Expect(res["location"]).To(ContainSubstring("San Francisco"), fmt.Sprint(res))
-				Expect(res["unit"]).To(Equal("celcius"), fmt.Sprint(res))
-				Expect(string(resp2.Choices[0].FinishReason)).To(Equal("function_call"), fmt.Sprint(resp2.Choices[0].FinishReason))
-
-			})
-
 			It("runs openllama gguf(llama-cpp)", Label("llama-gguf"), func() {
 				if runtime.GOOS != "linux" {
 					Skip("test supported only on linux")

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/gallery.go

@@ -117,19 +117,25 @@ func (mgs *ModelGalleryEndpointService) DeleteModelGalleryEndpoint() func(c *fib
 // @Router /models/available [get]
 func (mgs *ModelGalleryEndpointService) ListModelFromGalleryEndpoint() func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		log.Debug().Msgf("Listing models from galleries: %+v", mgs.galleries)
 
 		models, err := gallery.AvailableGalleryModels(mgs.galleries, mgs.modelPath)
 		if err != nil {
 			return err
 		}
-		log.Debug().Msgf("Models found from galleries: %+v", models)
-		for _, m := range models {
-			log.Debug().Msgf("Model found from galleries: %+v", m)
+
+		log.Debug().Msgf("Available %d models from %d galleries\n", len(models), len(mgs.galleries))
+
+		m := []gallery.Metadata{}
+
+		for _, mm := range models {
+			m = append(m, mm.Metadata)
 		}
-		dat, err := json.Marshal(models)
+
+		log.Debug().Msgf("Models %#v", m)
+
+		dat, err := json.Marshal(m)
 		if err != nil {
-			return err
+			return fmt.Errorf("could not marshal models: %w", err)
 		}
 		return c.Send(dat)
 	}

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()
@@ -401,6 +396,11 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 				log.Debug().Msgf("Text content to return: %s", textContentToReturn)
 				noActionsToRun := len(results) > 0 && results[0].Name == noActionName || len(results) == 0
 
+				finishReason := "stop"
+				if len(input.Tools) > 0 {
+					finishReason = "tool_calls"
+				}
+
 				switch {
 				case noActionsToRun:
 					result, err := handleQuestion(config, input, ml, startupOptions, results, s, predInput)
@@ -408,19 +408,18 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 						log.Error().Err(err).Msg("error handling question")
 						return
 					}
+
 					*c = append(*c, schema.Choice{
-						Message: &schema.Message{Role: "assistant", Content: &result}})
+						FinishReason: finishReason,
+						Message:      &schema.Message{Role: "assistant", Content: &result}})
 				default:
 					toolChoice := schema.Choice{
+						FinishReason: finishReason,
 						Message: &schema.Message{
 							Role: "assistant",
 						},
 					}
 
-					if len(input.Tools) > 0 {
-						toolChoice.FinishReason = "tool_calls"
-					}
-
 					for _, ss := range results {
 						name, args := ss.Name, ss.Arguments
 						if len(input.Tools) > 0 {
@@ -438,7 +437,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 								},
 							)
 						} else {
-							// otherwise we return more choices directly
+							// otherwise we return more choices directly (deprecated)
 							*c = append(*c, schema.Choice{
 								FinishReason: "function_call",
 								Message: &schema.Message{
@@ -539,7 +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{}

core/http/endpoints/openai/embeddings.go

@@ -2,11 +2,11 @@ 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/mudler/LocalAI/pkg/model"
 
 	"github.com/google/uuid"
@@ -23,14 +23,14 @@ import (
 // @Router /v1/embeddings [post]
 func EmbeddingsEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		model, 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
 		}
 
-		config, input, err := mergeRequestWithConfig(model, 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
 		}
 
 		log.Debug().Msgf("Parameter Config: %+v", config)

core/http/endpoints/openai/image.go

@@ -15,6 +15,7 @@ import (
 
 	"github.com/google/uuid"
 	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/schema"
 
 	"github.com/mudler/LocalAI/core/backend"
@@ -66,25 +67,23 @@ func downloadFile(url string) (string, error) {
 // @Router /v1/images/generations [post]
 func ImageEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		m, input, err := readRequest(c, cl, ml, appConfig, false)
-		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 == "" {
+			log.Error().Msg("Image Endpoint - Invalid Input")
+			return fiber.ErrBadRequest
 		}
-
-		if m == "" {
-			m = "stablediffusion"
-		}
-		log.Debug().Msgf("Loading model: %+v", m)
-
-		config, input, err := mergeRequestWithConfig(m, input, cl, ml, appConfig.Debug, 0, 0, false)
-		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 {
+			log.Error().Msg("Image Endpoint - Invalid Config")
+			return fiber.ErrBadRequest
 		}
 
 		src := ""
 		if input.File != "" {
 
 			fileData := []byte{}
+			var err error
 			// check if input.File is an URL, if so download it and save it
 			// to a temporary file
 			if strings.HasPrefix(input.File, "http://") || strings.HasPrefix(input.File, "https://") {

core/http/endpoints/openai/inference.go

@@ -37,7 +37,7 @@ func ComputeChoices(
 	}
 
 	// get the model function to call for the result
-	predFunc, err := backend.ModelInference(req.Context, predInput, req.Messages, images, videos, audios, loader, *config, o, tokenCallback)
+	predFunc, err := backend.ModelInference(req.Context, predInput, req.Messages, images, videos, audios, loader, config, o, tokenCallback)
 	if err != nil {
 		return result, backend.TokenUsage{}, err
 	}

core/http/endpoints/openai/transcription.go

@@ -1,7 +1,6 @@
 package openai
 
 import (
-	"fmt"
 	"io"
 	"net/http"
 	"os"
@@ -10,6 +9,8 @@ import (
 
 	"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"
 	model "github.com/mudler/LocalAI/pkg/model"
 
 	"github.com/gofiber/fiber/v2"
@@ -25,15 +26,16 @@ import (
 // @Router /v1/audio/transcriptions [post]
 func TranscriptEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		m, input, err := readRequest(c, cl, ml, appConfig, false)
-		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
 		}
 
-		config, input, err := mergeRequestWithConfig(m, 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
 		}
+
 		// retrieve the file data from the request
 		file, err := c.FormFile("file")
 		if err != nil {