Merge branch 'dev' into embedding-model-fix-and-manual-update

.env.example

@@ -9,4 +9,8 @@ OPENAI_API_KEY=''
 
 # DO NOT TRACK
 SCARF_NO_ANALYTICS=true
-DO_NOT_TRACK=true
\ No newline at end of file
+DO_NOT_TRACK=true
+
+# Use locally bundled version of the LiteLLM cost map json
+# to avoid repetitive startup connections
+LITELLM_LOCAL_MODEL_COST_MAP="True"

.github/workflows/build-release.yml

@@ -57,3 +57,14 @@ jobs:
         path: .
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+    - name: Trigger Docker build workflow
+      uses: actions/github-script@v7
+      with:
+        script: |
+          github.rest.actions.createWorkflowDispatch({
+            owner: context.repo.owner,
+            repo: context.repo.repo,
+            workflow_id: 'docker-build.yaml',
+            ref: 'v${{ steps.get_version.outputs.version }}',
+          })

.github/workflows/docker-build.yaml

-#
-name: Create and publish a Docker image
+name: Create and publish Docker images with specific build args
 
 # Configures this workflow to run every time a change is pushed to the branch called `release`.
 on:
+  workflow_dispatch:
   push:
     branches:
       - main
@@ -23,7 +23,7 @@ jobs:
     permissions:
       contents: read
       packages: write
-      #
+
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
@@ -41,12 +41,11 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: Extract metadata for Docker images
-        id: meta
+      - name: Extract metadata for Docker images (default latest tag)
+        id: meta-latest
         uses: docker/metadata-action@v5
         with:
           images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
-          # This configuration dynamically generates tags based on the branch, tag, commit, and custom suffix for lite version.
           tags: |
             type=ref,event=branch
             type=ref,event=tag
@@ -56,11 +55,29 @@ jobs:
           flavor: |
             latest=${{ github.ref == 'refs/heads/main' }}
 
-      - name: Build and push Docker image
+      - name: Build and push Docker image (latest)
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          push: true
+          platforms: linux/amd64,linux/arm64
+          tags: ${{ steps.meta-latest.outputs.tags }}
+          labels: ${{ steps.meta-latest.outputs.labels }}
+
+      - name: Build and push Docker image with CUDA
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          push: true
+          platforms: linux/amd64,linux/arm64
+          tags: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:cuda
+          build-args: USE_CUDA=true
+
+      - name: Build and push Docker image with Ollama
         uses: docker/build-push-action@v5
         with:
           context: .
           push: true
           platforms: linux/amd64,linux/arm64
-          tags: ${{ steps.meta.outputs.tags }}
-          labels: ${{ steps.meta.outputs.labels }}
+          tags: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:ollama
+          build-args: USE_OLLAMA=true

Dockerfile

 # syntax=docker/dockerfile:1
+# Initialize device type args
+# use build args in the docker build commmand with --build-arg="BUILDARG=true"
+ARG USE_CUDA=false
+ARG USE_OLLAMA=false
+# Tested with cu117 for CUDA 11 and cu121 for CUDA 12 (default)
+ARG USE_CUDA_VER=cu121
+# any sentence transformer model; models to use can be found at https://huggingface.co/models?library=sentence-transformers
+# Leaderboard: https://huggingface.co/spaces/mteb/leaderboard 
+# for better performance and multilangauge support use "intfloat/multilingual-e5-large" (~2.5GB) or "intfloat/multilingual-e5-base" (~1.5GB)
+# IMPORTANT: If you change the default model (all-MiniLM-L6-v2) and vice versa, you aren't able to use RAG Chat with your previous documents loaded in the WebUI! You need to re-embed them.
+ARG USE_EMBEDDING_MODEL=all-MiniLM-L6-v2
 
-FROM node:alpine as build
+######## WebUI frontend ########
+FROM node:21-alpine3.19 as build
 
 WORKDIR /app
 
-# wget embedding model weight from alpine (does not exist from slim-buster)
-RUN wget "https://chroma-onnx-models.s3.amazonaws.com/all-MiniLM-L6-v2/onnx.tar.gz" -O - | \
-    tar -xzf - -C /app
-
 COPY package.json package-lock.json ./
 RUN npm ci
 
 COPY . .
 RUN npm run build
 
-
+######## WebUI backend ########
 FROM python:3.11-slim-bookworm as base
 
-ENV ENV=prod
-ENV PORT ""
-
-ENV OLLAMA_BASE_URL "/ollama"
-
-ENV OPENAI_API_BASE_URL ""
-ENV OPENAI_API_KEY ""
-
-ENV WEBUI_SECRET_KEY ""
-ENV WEBUI_AUTH_TRUSTED_EMAIL_HEADER ""
-
-ENV SCARF_NO_ANALYTICS true
-ENV DO_NOT_TRACK true
-
-######## Preloaded models ########
-# whisper TTS Settings
-ENV WHISPER_MODEL="base"
-ENV WHISPER_MODEL_DIR="/app/backend/data/cache/whisper/models"
-
-# RAG Embedding Model Settings
-# any sentence transformer model; models to use can be found at https://huggingface.co/models?library=sentence-transformers
-# Leaderboard: https://huggingface.co/spaces/mteb/leaderboard 
-# for better persormance and multilangauge support use "intfloat/multilingual-e5-large" (~2.5GB) or "intfloat/multilingual-e5-base" (~1.5GB)
-# IMPORTANT: If you change the default model (all-MiniLM-L6-v2) and vice versa, you aren't able to use RAG Chat with your previous documents loaded in the WebUI! You need to re-embed them.
-ENV RAG_EMBEDDING_MODEL="all-MiniLM-L6-v2"
-# device type for whisper tts and embbeding models - "cpu" (default), "cuda" (nvidia gpu and CUDA required) or "mps" (apple silicon) - choosing this right can lead to better performance
-ENV RAG_EMBEDDING_MODEL_DEVICE_TYPE="cpu"
-ENV RAG_EMBEDDING_MODEL_DIR="/app/backend/data/cache/embedding/models"
-ENV SENTENCE_TRANSFORMERS_HOME $RAG_EMBEDDING_MODEL_DIR
-
-######## Preloaded models ########
+# Use args
+ARG USE_CUDA
+ARG USE_OLLAMA
+ARG USE_CUDA_VER
+ARG USE_EMBEDDING_MODEL
+
+## Basis ##
+ENV ENV=prod \
+    PORT=8080 \
+    # pass build args to the build
+    USE_OLLAMA_DOCKER=${USE_OLLAMA} \
+    USE_CUDA_DOCKER=${USE_CUDA} \
+    USE_CUDA_DOCKER_VER=${USE_CUDA_VER} \
+    USE_EMBEDDING_MODEL_DOCKER=${USE_EMBEDDING_MODEL}
+
+## Basis URL Config ##
+ENV OLLAMA_BASE_URL="/ollama" \
+    OPENAI_API_BASE_URL=""
+
+## API Key and Security Config ##
+ENV OPENAI_API_KEY="" \
+    WEBUI_SECRET_KEY="" \
+    SCARF_NO_ANALYTICS=true \
+    DO_NOT_TRACK=true
+
+# Use locally bundled version of the LiteLLM cost map json
+# to avoid repetitive startup connections
+ENV LITELLM_LOCAL_MODEL_COST_MAP="True"
+
+
+#### Other models #########################################################
+## whisper TTS model settings ##
+ENV WHISPER_MODEL="base" \
+    WHISPER_MODEL_DIR="/app/backend/data/cache/whisper/models"
+
+## RAG Embedding model settings ##
+ENV RAG_EMBEDDING_MODEL="$USE_EMBEDDING_MODEL_DOCKER" \
+    RAG_EMBEDDING_MODEL_DIR="/app/backend/data/cache/embedding/models" \
+    SENTENCE_TRANSFORMERS_HOME="/app/backend/data/cache/embedding/models"
+#### Other models ##########################################################
 
 WORKDIR /app/backend
-
 # install python dependencies
 COPY ./backend/requirements.txt ./requirements.txt
 
-RUN apt-get update && apt-get install ffmpeg libsm6 libxext6  -y
+RUN if [ "$USE_CUDA" = "true" ]; then \
+    # If you use CUDA the whisper and embedding modell will be downloaded on first use
+    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/$USE_CUDA_DOCKER_VER --no-cache-dir && \
+    pip3 install -r requirements.txt --no-cache-dir && \
+    python -c "import os; from faster_whisper import WhisperModel; WhisperModel(os.environ['WHISPER_MODEL'], device='cpu', compute_type='int8', download_root=os.environ['WHISPER_MODEL_DIR'])" && \
+    python -c "import os; from chromadb.utils import embedding_functions; sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(model_name=os.environ['RAG_EMBEDDING_MODEL'], device='cpu')"; \
+    else \
+    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu --no-cache-dir && \
+    pip3 install -r requirements.txt --no-cache-dir && \
+    python -c "import os; from faster_whisper import WhisperModel; WhisperModel(os.environ['WHISPER_MODEL'], device='cpu', compute_type='int8', download_root=os.environ['WHISPER_MODEL_DIR'])" && \
+    python -c "import os; from chromadb.utils import embedding_functions; sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(model_name=os.environ['RAG_EMBEDDING_MODEL'], device='cpu')"; \
+    fi
+
+
+RUN if [ "$USE_OLLAMA" = "true" ]; then \
+    apt-get update && \
+    # Install pandoc and netcat
+    apt-get install -y --no-install-recommends pandoc netcat-openbsd && \
+    # for RAG OCR
+    apt-get install -y --no-install-recommends ffmpeg libsm6 libxext6 && \
+    # install helper tools
+    apt-get install -y --no-install-recommends curl && \
+    # install ollama
+    curl -fsSL https://ollama.com/install.sh | sh && \
+    # cleanup
+    rm -rf /var/lib/apt/lists/*; \
+    else \
+    apt-get update && \
+    # Install pandoc and netcat
+    apt-get install -y --no-install-recommends pandoc netcat-openbsd && \
+    # for RAG OCR
+    apt-get install -y --no-install-recommends ffmpeg libsm6 libxext6 && \
+    # cleanup
+    rm -rf /var/lib/apt/lists/*; \
+    fi
 
-RUN pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu --no-cache-dir
-RUN pip3 install -r requirements.txt --no-cache-dir
 
-# Install pandoc and netcat
-# RUN python -c "import pypandoc; pypandoc.download_pandoc()"
-RUN apt-get update \
-    && apt-get install -y pandoc netcat-openbsd \
-    && rm -rf /var/lib/apt/lists/*
-
-# preload embedding model
-RUN python -c "import os; from chromadb.utils import embedding_functions; sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(model_name=os.environ['RAG_EMBEDDING_MODEL'], device=os.environ['RAG_EMBEDDING_MODEL_DEVICE_TYPE'])"
-# preload tts model
-RUN python -c "import os; from faster_whisper import WhisperModel; WhisperModel(os.environ['WHISPER_MODEL'], device='auto', compute_type='int8', download_root=os.environ['WHISPER_MODEL_DIR'])"
 
 # copy embedding weight from build
-RUN mkdir -p /root/.cache/chroma/onnx_models/all-MiniLM-L6-v2
-COPY --from=build /app/onnx /root/.cache/chroma/onnx_models/all-MiniLM-L6-v2/onnx
+# RUN mkdir -p /root/.cache/chroma/onnx_models/all-MiniLM-L6-v2
+# COPY --from=build /app/onnx /root/.cache/chroma/onnx_models/all-MiniLM-L6-v2/onnx
 
 # copy built frontend files
 COPY --from=build /app/build /app/build
@@ -82,4 +120,6 @@ COPY --from=build /app/package.json /app/package.json
 # copy backend files
 COPY ./backend .
 
-CMD [ "bash", "start.sh"]
+EXPOSE 8080
+
+CMD [ "bash", "start.sh"]
\ No newline at end of file

README.md

@@ -113,6 +113,65 @@ Don't forget to explore our sibling project, [Open WebUI Community](https://open
 
 - After installation, you can access Open WebUI at [http://localhost:3000](http://localhost:3000). Enjoy! 😄
 
+- **If you want to customize your build with additional args**, use this commands:
+
+  > [!NOTE]  
+  > If you only want to use Open WebUI with Ollama included or CUDA acelleration it's recomented to use our official images with the tags :cuda or :with-ollama
+  > If you want a combination of both or more customisation options like a different embedding model and/or CUDA version you need to build the image yourself following the instructions below.
+
+  **For the build:**
+
+  ```bash
+  docker build -t open-webui
+  ```
+
+  Optional build ARGS (use them in the docker build command below if needed):
+
+      e.g.
+
+  ```bash
+  --build-arg="USE_EMBEDDING_MODEL=intfloat/multilingual-e5-large"
+  ```
+
+  For "intfloat/multilingual-e5-large" custom embedding model (default is all-MiniLM-L6-v2), only works with [sentence transforer models](https://huggingface.co/models?library=sentence-transformers). Current [Leaderbord](https://huggingface.co/spaces/mteb/leaderboard) of embedding models.
+
+  ```bash
+  --build-arg="USE_OLLAMA=true"
+  ```
+
+  For including ollama in the image.
+
+  ```bash
+  --build-arg="USE_CUDA=true"
+  ```
+
+  To use CUDA exeleration for the embedding and whisper models.
+
+  > [!NOTE]
+  > You need to install the [Nvidia CUDA container toolkit](https://docs.nvidia.com/dgx/nvidia-container-runtime-upgrade/) on your machine to be able to set CUDA as the Docker engine. Only works with Linux - use WSL for Windows!
+
+  ```bash
+  --build-arg="USE_CUDA_VER=cu117"
+  ```
+
+  For CUDA 11 (default is CUDA 12)
+
+  **To run the image:**
+
+  - **If you DID NOT use the USE_CUDA=true build ARG**, use this command:
+
+  ```bash
+    docker run -d -p 3000:8080 -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main
+  ```
+
+  - **If you DID use the USE_CUDA=true build ARG**, use this command:
+
+  ```bash
+    docker run --gpus all -d -p 3000:8080 -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main
+  ```
+
+  - After installation, you can access Open WebUI at [http://localhost:3000](http://localhost:3000). Enjoy! 😄
+
 #### Open WebUI: Server Connection Error
 
 If you're experiencing connection issues, it’s often due to the WebUI docker container not being able to reach the Ollama server at 127.0.0.1:11434 (host.docker.internal:11434) inside the container . Use the `--network=host` flag in your docker command to resolve this. Note that the port changes from 3000 to 8080, resulting in the link: `http://localhost:8080`.

backend/apps/audio/main.py

@@ -28,6 +28,7 @@ from config import (
     UPLOAD_DIR,
     WHISPER_MODEL,
     WHISPER_MODEL_DIR,
+    DEVICE_TYPE,
 )
 
 log = logging.getLogger(__name__)
@@ -42,6 +43,10 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
+# setting device type for whisper model
+whisper_device_type = DEVICE_TYPE if DEVICE_TYPE and DEVICE_TYPE == "cuda" else "cpu"
+log.info(f"whisper_device_type: {whisper_device_type}")
+
 
 @app.post("/transcribe")
 def transcribe(
@@ -66,7 +71,7 @@ def transcribe(
 
         model = WhisperModel(
             WHISPER_MODEL,
-            device="auto",
+            device=whisper_device_type,
             compute_type="int8",
             download_root=WHISPER_MODEL_DIR,
         )

backend/apps/ollama/main.py

@@ -215,7 +215,8 @@ async def get_ollama_versions(url_idx: Optional[int] = None):
 
         if len(responses) > 0:
             lowest_version = min(
-                responses, key=lambda x: tuple(map(int, x["version"].split(".")))
+                responses,
+                key=lambda x: tuple(map(int, x["version"].split("-")[0].split("."))),
             )
 
             return {"version": lowest_version["version"]}

backend/apps/rag/main.py

@@ -58,8 +58,8 @@ from config import (
     UPLOAD_DIR,
     DOCS_DIR,
     RAG_EMBEDDING_MODEL,
-    RAG_EMBEDDING_MODEL_DEVICE_TYPE,
     RAG_EMBEDDING_MODEL_AUTO_UPDATE,
+    DEVICE_TYPE,
     CHROMA_CLIENT,
     CHUNK_SIZE,
     CHUNK_OVERLAP,
@@ -86,7 +86,7 @@ app.state.TOP_K = 4
 app.state.sentence_transformer_ef = (
     embedding_functions.SentenceTransformerEmbeddingFunction(
         model_name=app.state.RAG_EMBEDDING_MODEL_PATH,
-        device=RAG_EMBEDDING_MODEL_DEVICE_TYPE,
+        device=DEVICE_TYPE,
     )
 )
 
@@ -154,7 +154,7 @@ async def update_embedding_model(
         app.state.sentence_transformer_ef = (
             embedding_functions.SentenceTransformerEmbeddingFunction(
                 model_name=app.state.RAG_EMBEDDING_MODEL_PATH,
-                device=RAG_EMBEDDING_MODEL_DEVICE_TYPE,
+                device=DEVICE_TYPE,
             )
         )
     except Exception as e:
@@ -471,25 +471,11 @@ def store_doc(
 
     log.info(f"file.content_type: {file.content_type}")
     try:
-        is_valid_filename = True
         unsanitized_filename = file.filename
-        if re.search(r'[\\/:"\*\?<>|\n\t ]', unsanitized_filename) is not None:
-            is_valid_filename = False
+        filename = os.path.basename(unsanitized_filename)
 
-        unvalidated_file_path = f"{UPLOAD_DIR}/{unsanitized_filename}"
-        dereferenced_file_path = str(Path(unvalidated_file_path).resolve(strict=False))
-        if not dereferenced_file_path.startswith(UPLOAD_DIR):
-            is_valid_filename = False
+        file_path = f"{UPLOAD_DIR}/{filename}"
 
-        if is_valid_filename:
-            file_path = dereferenced_file_path
-        else:
-            raise HTTPException(
-                status_code=status.HTTP_400_BAD_REQUEST,
-                detail=ERROR_MESSAGES.DEFAULT(),
-            )
-
-        filename = file.filename
         contents = file.file.read()
         with open(file_path, "wb") as f:
             f.write(contents)
@@ -500,7 +486,7 @@ def store_doc(
             collection_name = calculate_sha256(f)[:63]
         f.close()
 
-        loader, known_type = get_loader(file.filename, file.content_type, file_path)
+        loader, known_type = get_loader(filename, file.content_type, file_path)
         data = loader.load()
 
         try:

backend/apps/web/models/auths.py

@@ -86,6 +86,7 @@ class SignupForm(BaseModel):
     name: str
     email: str
     password: str
+    profile_image_url: Optional[str] = "/user.png"
 
 
 class AuthsTable:
@@ -94,7 +95,12 @@ class AuthsTable:
         self.db.create_tables([Auth])
 
     def insert_new_auth(
-        self, email: str, password: str, name: str, role: str = "pending"
+        self,
+        email: str,
+        password: str,
+        name: str,
+        profile_image_url: str = "/user.png",
+        role: str = "pending",
     ) -> Optional[UserModel]:
         log.info("insert_new_auth")
 
@@ -105,7 +111,7 @@ class AuthsTable:
         )
         result = Auth.create(**auth.model_dump())
 
-        user = Users.insert_new_user(id, name, email, role)
+        user = Users.insert_new_user(id, name, email, profile_image_url, role)
 
         if result and user:
             return user

backend/apps/web/models/chats.py

@@ -206,6 +206,18 @@ class ChatTable:
         except:
             return None
 
+    def get_chat_by_share_id(self, id: str) -> Optional[ChatModel]:
+        try:
+            chat = Chat.get(Chat.share_id == id)
+
+            if chat:
+                chat = Chat.get(Chat.id == id)
+                return ChatModel(**model_to_dict(chat))
+            else:
+                return None
+        except:
+            return None
+
     def get_chat_by_id_and_user_id(self, id: str, user_id: str) -> Optional[ChatModel]:
         try:
             chat = Chat.get(Chat.id == id, Chat.user_id == user_id)

backend/apps/web/models/users.py

@@ -31,7 +31,7 @@ class UserModel(BaseModel):
     name: str
     email: str
     role: str = "pending"
-    profile_image_url: str = "/user.png"
+    profile_image_url: str
     timestamp: int  # timestamp in epoch
     api_key: Optional[str] = None
 
@@ -59,7 +59,12 @@ class UsersTable:
         self.db.create_tables([User])
 
     def insert_new_user(
-        self, id: str, name: str, email: str, role: str = "pending"
+        self,
+        id: str,
+        name: str,
+        email: str,
+        profile_image_url: str = "/user.png",
+        role: str = "pending",
     ) -> Optional[UserModel]:
         user = UserModel(
             **{
@@ -67,7 +72,7 @@ class UsersTable:
                 "name": name,
                 "email": email,
                 "role": role,
-                "profile_image_url": "/user.png",
+                "profile_image_url": profile_image_url,
                 "timestamp": int(time.time()),
             }
         )

backend/apps/web/routers/auths.py

@@ -163,7 +163,11 @@ async def signup(request: Request, form_data: SignupForm):
         )
         hashed = get_password_hash(form_data.password)
         user = Auths.insert_new_auth(
-            form_data.email.lower(), hashed, form_data.name, role
+            form_data.email.lower(),
+            hashed,
+            form_data.name,
+            form_data.profile_image_url,
+            role,
         )
 
         if user:

backend/apps/web/routers/chats.py

@@ -251,7 +251,15 @@ async def delete_shared_chat_by_id(id: str, user=Depends(get_current_user)):
 
 @router.get("/share/{share_id}", response_model=Optional[ChatResponse])
 async def get_shared_chat_by_id(share_id: str, user=Depends(get_current_user)):
-    chat = Chats.get_chat_by_id(share_id)
+    if user.role == "pending":
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED, detail=ERROR_MESSAGES.NOT_FOUND
+        )
+
+    if user.role == "user":
+        chat = Chats.get_chat_by_share_id(share_id)
+    elif user.role == "admin":
+        chat = Chats.get_chat_by_id(share_id)
 
     if chat:
         return ChatResponse(**{**chat.model_dump(), "chat": json.loads(chat.chat)})

backend/apps/web/routers/utils.py

-from fastapi import APIRouter, UploadFile, File, BackgroundTasks
+from fastapi import APIRouter, UploadFile, File, Response
 from fastapi import Depends, HTTPException, status
 from starlette.responses import StreamingResponse, FileResponse
-
-
 from pydantic import BaseModel
 
 
+from fpdf import FPDF
 import markdown
-import requests
-import os
-import aiohttp
-import json
 
 
 from utils.utils import get_admin_user
@@ -18,7 +13,7 @@ from utils.misc import calculate_sha256, get_gravatar_url
 
 from config import OLLAMA_BASE_URLS, DATA_DIR, UPLOAD_DIR
 from constants import ERROR_MESSAGES
-
+from typing import List
 
 router = APIRouter()
 
@@ -41,6 +36,59 @@ async def get_html_from_markdown(
     return {"html": markdown.markdown(form_data.md)}
 
 
+class ChatForm(BaseModel):
+    title: str
+    messages: List[dict]
+
+
+@router.post("/pdf")
+async def download_chat_as_pdf(
+    form_data: ChatForm,
+):
+    pdf = FPDF()
+    pdf.add_page()
+
+    STATIC_DIR = "./static"
+    FONTS_DIR = f"{STATIC_DIR}/fonts"
+
+    pdf.add_font("NotoSans", "", f"{FONTS_DIR}/NotoSans-Regular.ttf")
+    pdf.add_font("NotoSans", "b", f"{FONTS_DIR}/NotoSans-Bold.ttf")
+    pdf.add_font("NotoSans", "i", f"{FONTS_DIR}/NotoSans-Italic.ttf")
+    pdf.add_font("NotoSansKR", "", f"{FONTS_DIR}/NotoSansKR-Regular.ttf")
+    pdf.add_font("NotoSansJP", "", f"{FONTS_DIR}/NotoSansJP-Regular.ttf")
+
+    pdf.set_font("NotoSans", size=12)
+    pdf.set_fallback_fonts(["NotoSansKR", "NotoSansJP"])
+
+    pdf.set_auto_page_break(auto=True, margin=15)
+
+    # Adjust the effective page width for multi_cell
+    effective_page_width = (
+        pdf.w - 2 * pdf.l_margin - 10
+    )  # Subtracted an additional 10 for extra padding
+
+    # Add chat messages
+    for message in form_data.messages:
+        role = message["role"]
+        content = message["content"]
+        pdf.set_font("NotoSans", "B", size=14)  # Bold for the role
+        pdf.multi_cell(effective_page_width, 10, f"{role.upper()}", 0, "L")
+        pdf.ln(1)  # Extra space between messages
+
+        pdf.set_font("NotoSans", size=10)  # Regular for content
+        pdf.multi_cell(effective_page_width, 6, content, 0, "L")
+        pdf.ln(1.5)  # Extra space between messages
+
+    # Save the pdf with name .pdf
+    pdf_bytes = pdf.output()
+
+    return Response(
+        content=bytes(pdf_bytes),
+        media_type="application/pdf",
+        headers={"Content-Disposition": f"attachment;filename=chat.pdf"},
+    )
+
+
 @router.get("/db/download")
 async def download_db(user=Depends(get_admin_user)):
 

backend/config.py

@@ -257,6 +257,7 @@ OLLAMA_API_BASE_URL = os.environ.get(
 
 OLLAMA_BASE_URL = os.environ.get("OLLAMA_BASE_URL", "")
 K8S_FLAG = os.environ.get("K8S_FLAG", "")
+USE_OLLAMA_DOCKER = os.environ.get("USE_OLLAMA_DOCKER", "false")
 
 if OLLAMA_BASE_URL == "" and OLLAMA_API_BASE_URL != "":
     OLLAMA_BASE_URL = (
@@ -266,9 +267,13 @@ if OLLAMA_BASE_URL == "" and OLLAMA_API_BASE_URL != "":
     )
 
 if ENV == "prod":
-    if OLLAMA_BASE_URL == "/ollama":
-        OLLAMA_BASE_URL = "http://host.docker.internal:11434"
-
+    if OLLAMA_BASE_URL == "/ollama" and not K8S_FLAG:
+        if USE_OLLAMA_DOCKER.lower() == "true":
+            # if you use all-in-one docker container (Open WebUI + Ollama)
+            # with the docker build arg USE_OLLAMA=true (--build-arg="USE_OLLAMA=true") this only works with http://localhost:11434
+            OLLAMA_BASE_URL = "http://localhost:11434"
+        else:
+            OLLAMA_BASE_URL = "http://host.docker.internal:11434"
     elif K8S_FLAG:
         OLLAMA_BASE_URL = "http://ollama-service.open-webui.svc.cluster.local:11434"
 
@@ -391,13 +396,21 @@ if WEBUI_AUTH and WEBUI_SECRET_KEY == "":
 CHROMA_DATA_PATH = f"{DATA_DIR}/vector_db"
 # this uses the model defined in the Dockerfile ENV variable. If you dont use docker or docker based deployments such as k8s, the default embedding model will be used (all-MiniLM-L6-v2)
 RAG_EMBEDDING_MODEL = os.environ.get("RAG_EMBEDDING_MODEL", "all-MiniLM-L6-v2")
-# device type ebbeding models - "cpu" (default), "cuda" (nvidia gpu required) or "mps" (apple silicon) - choosing this right can lead to better performance
-RAG_EMBEDDING_MODEL_DEVICE_TYPE = os.environ.get(
-    "RAG_EMBEDDING_MODEL_DEVICE_TYPE", "cpu"
-)
+log.info(f"Embedding model set: {RAG_EMBEDDING_MODEL}"),
 RAG_EMBEDDING_MODEL_AUTO_UPDATE = False
 if os.environ.get("RAG_EMBEDDING_MODEL_AUTO_UPDATE", "").lower() == "true":
     RAG_EMBEDDING_MODEL_AUTO_UPDATE = True
+
+
+# device type ebbeding models - "cpu" (default), "cuda" (nvidia gpu required) or "mps" (apple silicon) - choosing this right can lead to better performance
+USE_CUDA = os.environ.get("USE_CUDA_DOCKER", "false")
+
+if USE_CUDA.lower() == "true":
+    DEVICE_TYPE = "cuda"
+else:
+    DEVICE_TYPE = "cpu"
+
+
 CHROMA_CLIENT = chromadb.PersistentClient(
     path=CHROMA_DATA_PATH,
     settings=Settings(allow_reset=True, anonymized_telemetry=False),

backend/requirements.txt

@@ -42,6 +42,8 @@ xlrd
 opencv-python-headless
 rapidocr-onnxruntime
 
+fpdf2
+
 faster-whisper
 
 PyJWT

backend/start.sh

@@ -7,16 +7,26 @@ KEY_FILE=.webui_secret_key
 
 PORT="${PORT:-8080}"
 if test "$WEBUI_SECRET_KEY $WEBUI_JWT_SECRET_KEY" = " "; then
-  echo No WEBUI_SECRET_KEY provided
+  echo "No WEBUI_SECRET_KEY provided"
 
   if ! [ -e "$KEY_FILE" ]; then
-    echo Generating WEBUI_SECRET_KEY
+    echo "Generating WEBUI_SECRET_KEY"
     # Generate a random value to use as a WEBUI_SECRET_KEY in case the user didn't provide one.
-    echo $(head -c 12 /dev/random | base64) > $KEY_FILE
+    echo $(head -c 12 /dev/random | base64) > "$KEY_FILE"
   fi
 
-  echo Loading WEBUI_SECRET_KEY from $KEY_FILE
-  WEBUI_SECRET_KEY=`cat $KEY_FILE`
+  echo "Loading WEBUI_SECRET_KEY from $KEY_FILE"
+  WEBUI_SECRET_KEY=$(cat "$KEY_FILE")
 fi
 
-WEBUI_SECRET_KEY="$WEBUI_SECRET_KEY" exec uvicorn main:app --host 0.0.0.0 --port "$PORT" --forwarded-allow-ips '*'
\ No newline at end of file
+if [ "$USE_OLLAMA_DOCKER" = "true" ]; then
+    echo "USE_OLLAMA is set to true, starting ollama serve."
+    ollama serve &
+fi
+
+if [ "$USE_CUDA_DOCKER" = "true" ]; then
+  echo "CUDA is enabled, appending LD_LIBRARY_PATH to include torch/cudnn & cublas libraries."
+  export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/lib/python3.11/site-packages/torch/lib:/usr/local/lib/python3.11/site-packages/nvidia/cudnn/lib"
+fi
+
+WEBUI_SECRET_KEY="$WEBUI_SECRET_KEY" exec uvicorn main:app --host 0.0.0.0 --port "$PORT" --forwarded-allow-ips '*'