Initial commit

Dockerfile

@@ -0,0 +1,15 @@
+# Building from source is too slow, so let's just customize the existing image
+FROM ghcr.io/open-webui/open-webui:git-143ac08
+
+WORKDIR /app
+
+USER 0:0
+
+# HACK for huggingface.co iframe
+RUN sed -i "s|set_cookie(|set_cookie(samesite='none',secure=True,|g" backend/open_webui/apps/webui/routers/auths.py
+
+RUN pip install "litellm[proxy]==1.44.27" && chown -R 1000:0 /app
+USER 1000:0
+
+COPY ./start.sh /start.sh
+CMD [ "bash", "/start.sh" ]

README.md

@@ -0,0 +1,12 @@
+---
+title: Open WebUI
+emoji: 🌍
+colorFrom: indigo
+colorTo: purple
+sdk: docker
+pinned: false
+app_port: 8080
+---
+
+A simple, opinionated Open WebUI setup for Hugging Face Spaces.
+It opts to maximize ease-of-use over customizability.

start.sh

@@ -0,0 +1,109 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+pids=()
+
+generate_litellm_provider() {
+	provider="$1"
+	key_env="$2"
+
+	if [[ -n "${!key_env:-}" ]]; then
+		>&2 echo "[!] Found key ${key_env} for ${provider}"
+		cat <<EOF
+- model_name: "${provider}/*"
+  litellm_params:
+    model: "${provider}/*"
+    api_key: "os.environ/${key_env}"
+EOF
+	fi
+}
+
+generate_litellm_config() {
+	cat <<EOF
+model_list:
+- model_name: huggingface/google/gemma-2-2b-it
+  litellm_params:
+    model: huggingface/google/gemma-2-2b-it
+EOF
+	generate_litellm_provider "openai" "OPENAI_API_KEY"
+	generate_litellm_provider "perplexity" "PERPLEXITY_API_KEY"
+	generate_litellm_provider "anthropic" "ANTHROPIC_API_KEY"
+
+	if [[ -n "${LITELLM_MODELS_BASE64:-}" ]]; then
+		echo "${LITELLM_MODELS_BASE64}" | base64 -d
+	fi
+}
+
+start_litellm() {
+	(
+	# KISS: No persistence for LiteLLM
+	unset DATABASE_URL
+
+	litellm \
+		--host "127.0.0.1" \
+		--port "4000" \
+		--config /tmp/litellm_config.yaml
+	) &
+	pids+=("$!")
+}
+
+start_open_webui() {
+	(
+	# By default, we expect it to be deployed to a private HF Space.
+	# You can enable WEBUI_AUTH as needed
+	if [[ -z "${WEBUI_AUTH:-}" ]]; then
+		>&2 echo "[!] Enabling single user mode"
+		export WEBUI_AUTH="False"
+	fi
+
+	# The less the user needs to configure, the better :)
+	if [[ -n "${OPENAI_API_KEY:-}" ]] && [[ -z "${WEBUI_SECRET_KEY:-}${WEBUI_JWT_SECRET_KEY:-}" ]]; then
+		>&2 echo "[!] Using OpenAI API key as Web UI secret key"
+		export WEBUI_SECRET_KEY="${OPENAI_API_KEY}"
+	fi
+
+	if [[ -z "${ENABLE_RAG_WEB_SEARCH:-}${RAG_WEB_SEARCH_ENGINE:-}" ]]; then
+		if [[ -n "${BRAVE_SEARCH_API_KEY:-}" ]]; then
+			export RAG_WEB_SEARCH_ENGINE="brave"
+			export ENABLE_RAG_WEB_SEARCH="True"
+		fi
+	fi
+
+	if [[ -z "${DATABASE_URL:-}" ]]; then
+		unset DATABASE_URL
+	fi
+
+	export ENABLE_OLLAMA_API="${ENABLE_OLLAMA_API:-False}"
+
+	export OPENAI_API_BASE_URLS="http://localhost:4000"
+	export OPENAI_API_KEY="sk-unused"
+
+	export ENABLE_IMAGE_GENERATION="True"
+	export IMAGES_OPENAI_API_BASE_URL="http://localhost:4000"
+	export IMAGES_OPENAI_API_KEY="sk-unused"
+
+	export PGSSLCERT=/tmp/postgresql.crt
+
+	env
+	/app/backend/start.sh
+	) &
+	pids+=("$!")
+}
+
+wait_litellm() {
+	while ! curl -s http://localhost:4000 >/dev/null; do
+		>&2 echo "[!] Waiting for LiteLLM..."
+		sleep 1
+	done
+}
+
+generate_litellm_config >/tmp/litellm_config.yaml
+>&2 cat /tmp/litellm_config.yaml
+
+start_litellm
+wait_litellm
+start_open_webui
+
+for pid in "${pids[@]}"; do
+    wait "$pid"
+done