Skip to main content

trainloop studio

Launch the Studio UI web interface for interactive visualization and analysis of evaluation results.

Synopsis​

trainloop studio [OPTIONS]

Description​

The trainloop studio command starts a local web server that provides an interactive interface for exploring evaluation results, comparing models, and analyzing LLM performance data.

Options​

OptionDescription
--port <number>HTTP port (default: 3000)
--host <address>Host binding address (default: localhost)
--config <path>Path to configuration file
--local <tar>Path to local Studio build (for development)
--helpShow help message

Examples​

Basic Usage​

# Start Studio UI on default port 3000
trainloop studio

Custom Port​

# Start on port 8080
trainloop studio --port 8080

Custom Host​

# Bind to all interfaces (for Docker/remote access)
trainloop studio --host 0.0.0.0

Development Mode​

# Use local build for development
trainloop studio --local trainloop-studio-runner.tgz

What You'll See​

The Studio UI provides several key interfaces:

Dashboard​

  • Overview of evaluation results
  • Summary statistics and trends
  • Recent evaluation runs

Results Explorer​

  • Interactive tables of evaluation results
  • Filtering and search capabilities
  • Drill-down to individual events

Benchmarks​

  • Model comparison charts
  • Performance vs. cost analysis
  • Side-by-side result comparison

Event Browser​

  • Raw LLM request/response data
  • Tagged event exploration
  • Data quality inspection

Data Requirements​

Studio UI requires:

  • Event data: JSONL files in data/events/
  • Results data: JSON files in data/results/
  • Valid configuration: Accessible trainloop.config.yaml

Configuration​

Studio UI uses the same configuration discovery as other CLI commands:

  1. --config command line argument
  2. TRAINLOOP_CONFIG_FILE environment variable
  3. trainloop.config.yaml in current directory
  4. Parent directories up to git root
  5. ~/.trainloop/config.yaml

Troubleshooting​

Common Issues​

Port Already in Use​

# Use different port
trainloop studio --port 8080

No Data Showing​

  1. Check data folder exists: ls $TRAINLOOP_DATA_FOLDER
  2. Verify results exist: ls $TRAINLOOP_DATA_FOLDER/results/
  3. Run evaluations: trainloop eval

Can't Access from Remote​

# Allow external access
trainloop studio --host 0.0.0.0 --port 3000

Production Deployment​

For production use, consider:

# Secure external access
trainloop studio --host 0.0.0.0 --port 3000

# Use reverse proxy (nginx, Apache)
# Set up authentication
# Configure SSL/TLS

See Also​