Skip to content

setup.md

Latest commit

 

History

History
188 lines (136 loc) · 4.81 KB

setup.md

File metadata and controls

188 lines (136 loc) · 4.81 KB

AlphaTrion Dashboard Setup Guide

Issue Fixed: UUID Format Error

The "badly formed hexadecimal UUID string" error occurs when the backend expects UUID format for all IDs, but no teams exist in the database yet.

Solution: The dashboard now automatically fetches teams and shows a helpful message if none exist.

Quick Start

1. Create Test Data

First, create a team, user, and experiments for testing:

cd /Users/kerthcet/Workspaces/InftyAI/alphatrion

# Make sure your database is configured
export ALPHATRION_METADATA_DB_URL='postgresql://user:password@localhost/alphatrion'

# Optional: Initialize database tables on first run
export ALPHATRION_INIT_METADATA_TABLES='true'

# Run the test data script
python scripts/create_test_data.py

This will create:

  • ✅ A test team with UUID
  • ✅ A test user with UUID
  • ✅ 3 sample experiments with different configurations and labels
  • ✅ Sample metrics for each experiment

2. Start the Backend

cd /Users/kerthcet/Workspaces/InftyAI/alphatrion
alphatrion

# Or with uvicorn directly:
# uvicorn alphatrion.server.cmd.app:app --reload

The backend will start on http://localhost:8000

3. View the Dashboard

Option A: Production Mode (served by backend)

# Build the dashboard first
cd dashboard
npm run build

# Then access via backend
# Open http://localhost:8000

Option B: Development Mode (with hot-reload)

cd dashboard
npm run dev

# Open http://localhost:5173

Dashboard Features

The dashboard follows a clean hierarchical structure: Experiments → Runs → Metrics

Navigation Structure

Sidebar Menu:

  • 🏠 Dashboard - Overview with quick stats and recent experiments
  • 🧪 Experiments - Browse all experiments
    • Filter experiments by labels
    • Click an experiment → View runs and metrics
  • 📦 Artifacts - Browse ORAS registry artifacts
  • 🔗 Tracing - View roadmap for future tracing features

Key Features

  1. Dashboard Overview

    • Quick stats showing teams and experiments count
    • Visual hierarchy explanation
    • Recent experiments list
    • Getting started guide

  2. Experiments Page

    • List all experiments in your team
    • Filter by labels for organization
    • Click any experiment to view details

  3. Experiment Detail Page

    • View experiment parameters, metadata, and labels
    • See experiment status, duration, and parameters
    • Click to view run details

  4. Run Detail Page

    • View run metadata
    • Real-time metrics charts with auto-refresh
    • View all metrics for the run
    • Compare with other runs

  5. Experiment Comparison

    • Side-by-side parameter diff
    • Metrics overlay chart
    • Highlight differences between experiments

  6. Artifacts Browser

    • Browse ORAS registry
    • View manifests and layers
    • Download artifacts

Troubleshooting

Still getting UUID errors?

Check if teams exist:

import os
os.environ['ALPHATRION_METADATA_DB_URL'] = 'your_db_url'

from alphatrion.storage.sqlstore import SQLStore
db = SQLStore(os.environ['ALPHATRION_METADATA_DB_URL'])
teams = db.list_teams()
print(f"Found {len(teams)} teams")
for team in teams:
    print(f"  Team: {team.name} (ID: {team.uuid})")

If no teams exist: Run the create_test_data.py script above, or create one manually:

import uuid
import alphatrion

team_id = uuid.uuid4()
user_id = uuid.uuid4()

alphatrion.init(team_id=team_id, user_id=user_id)
# Now create experiments...

Dashboard shows "No teams found"

This means no teams exist in your database yet. Run the test data script to create one.

GraphQL errors in browser console

  1. Check that the backend is running on the correct port (default: 8000)
  2. Verify the proxy configuration in dashboard/vite.config.ts matches your backend port
  3. Check browser console for specific error messages

Artifacts not loading

  1. Ensure ALPHATRION_ARTIFACT_REGISTRY_URL is set
  2. Check that ORAS registry is accessible
  3. Verify artifacts exist in the registry

Environment Variables

Required:

ALPHATRION_METADATA_DB_URL='postgresql://user:pass@localhost/dbname'

Optional:

ALPHATRION_INIT_METADATA_TABLES='true'  # Initialize tables on first run
ALPHATRION_ARTIFACT_REGISTRY_URL='your-oras-registry-url'
ALPHATRION_ARTIFACT_INSECURE='false'
ALPHATRION_ENABLE_ARTIFACT_STORAGE='true'
ALPHATRION_ENABLE_TRACING='true'

Next Steps

  1. Create real experiments - Use the AlphaTrion SDK in your ML code
  2. Test polling - Create a long-running experiment and watch real-time updates
  3. Compare experiments - Select multiple experiments and compare parameters
  4. Explore artifacts - Upload and browse experiment artifacts via ORAS

Support

For issues or questions: