Custom Argilla with auto-backup to HF Dataset

Dockerfile

@@ -1,10 +1,19 @@
-FROM argilla/argilla-hf-spaces:v2.8.0
+FROM argilla/argilla-server:latest
 
-# Copy the auth config section
-COPY .oauth.yaml /home/argilla/
+# Install additional dependencies
+USER root
+RUN pip install huggingface_hub datasets
 
-# Comment this line to disable annotation progress sharing feature
-ENV ARGILLA_ENABLE_SHARE_YOUR_PROGRESS=1
+# Create exports directory
+RUN mkdir -p /exports && chmod 777 /exports
 
-# Uncoment this line to remove the persistence storage warning
-#ENV ARGILLA_SHOW_HUGGINGFACE_SPACE_PERSISTENT_STORAGE_WARNING=false
+# Copy backup script
+COPY backup_to_hub.py /app/backup_to_hub.py
+COPY start.sh /app/start.sh
+RUN chmod +x /app/start.sh
+
+USER argilla
+WORKDIR /home/argilla
+
+# Start both Argilla and backup script
+CMD ["/app/start.sh"]

README.md

@@ -1,20 +1,8 @@
 ---
 title: Vuln Detection Preferences
-emoji: ✍
-colorFrom: pink
-colorTo: pink
+emoji: 🔒
+colorFrom: blue
+colorTo: purple
 sdk: docker
 app_port: 6900
-fullWidth: true
-hf_oauth: true
-tags:
-- argilla
-pinned: true
-license: mit
 ---
-Argilla is a free and open source tool to build and iterate on datasets for AI. It can be deployed on the Hub with a few clicks and Hugging Face OAuth enabled, perfect for running community annotation initiatives!
-
-This is the Argilla Space for:
-
-- Creating your own Argilla Spaces, check the [quickstart guide](http://docs.argilla.io/latest/getting_started/quickstart/) and the [Hugging Face Spaces configuration](http://docs.argilla.io/latest/getting_started/how-to-configure-argilla-on-huggingface/) for more details.
-- Discovering the Argilla UI, sign in with your Hugging Face account!

backup_to_hub.py

@@ -0,0 +1,88 @@
+"""
+Automatic backup of Argilla annotations to HF Dataset.
+Runs as background process, syncs every 10 minutes.
+"""
+import os
+import json
+import time
+from pathlib import Path
+from datetime import datetime
+from huggingface_hub import CommitScheduler, login
+import argilla as rg
+
+# Configuration
+EXPORT_DIR = Path("/exports")
+EXPORT_DIR.mkdir(exist_ok=True)
+BACKUP_REPO = os.environ.get("BACKUP_DATASET_REPO", "argus-systems/vuln-preferences-data")
+SYNC_INTERVAL = int(os.environ.get("SYNC_INTERVAL_MINUTES", "10"))
+ARGILLA_API_URL = "http://localhost:6900"
+
+# Login to HF (uses HF_TOKEN secret)
+hf_token = os.environ.get("HF_TOKEN")
+if hf_token:
+    login(token=hf_token)
+
+# Setup CommitScheduler
+scheduler = CommitScheduler(
+    repo_id=BACKUP_REPO,
+    repo_type="dataset",
+    folder_path=EXPORT_DIR,
+    every=SYNC_INTERVAL,
+)
+
+
+def wait_for_argilla():
+    """Wait for Argilla server to be ready."""
+    import urllib.request
+    max_retries = 30
+    for i in range(max_retries):
+        try:
+            urllib.request.urlopen(f"{ARGILLA_API_URL}/api/v1/status")
+            print("Argilla server is ready")
+            return True
+        except Exception:
+            print(f"Waiting for Argilla... ({i+1}/{max_retries})")
+            time.sleep(2)
+    return False
+
+
+def export_annotations():
+    """Export all annotations from Argilla to JSON files."""
+    try:
+        api_key = os.environ.get("ARGILLA_API_KEY", "argilla.apikey")
+        client = rg.Argilla(api_url=ARGILLA_API_URL, api_key=api_key)
+
+        for dataset in client.datasets:
+            records = list(dataset.records)
+
+            # Filter to records with responses (annotations)
+            annotated = [r.to_dict() for r in records if r.responses]
+
+            if annotated:
+                export_file = EXPORT_DIR / f"{dataset.name}_{datetime.now().strftime('%Y%m%d_%H%M%S')}.json"
+                with scheduler.lock:
+                    with open(export_file, "w") as f:
+                        json.dump({
+                            "dataset": dataset.name,
+                            "exported_at": datetime.now().isoformat(),
+                            "num_annotations": len(annotated),
+                            "records": annotated
+                        }, f, indent=2, default=str)
+                print(f"Exported {len(annotated)} annotations from {dataset.name}")
+    except Exception as e:
+        print(f"Export error: {e}")
+
+
+def main():
+    print("Starting Argilla backup service...")
+
+    if not wait_for_argilla():
+        print("Warning: Argilla not responding, will retry exports")
+
+    while True:
+        export_annotations()
+        time.sleep(SYNC_INTERVAL * 60)  # Convert to seconds
+
+
+if __name__ == "__main__":
+    main()

start.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+# Start backup script in background
+python /app/backup_to_hub.py &
+
+# Start Argilla server (original command)
+python -m argilla server start --host 0.0.0.0 --port 6900