# 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 - | \
# 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
- 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):
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:
-**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`.
@@ -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)
# 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}"),
# device type ebbeding models - "cpu" (default), "cuda" (nvidia gpu required) or "mps" (apple silicon) - choosing this right can lead to better performance
