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​
Option | Description |
---|---|
--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) |
--help | Show 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:
--config
command line argumentTRAINLOOP_CONFIG_FILE
environment variabletrainloop.config.yaml
in current directory- Parent directories up to git root
~/.trainloop/config.yaml
Troubleshooting​
Common Issues​
Port Already in Use​
# Use different port
trainloop studio --port 8080
No Data Showing​
- Check data folder exists:
ls $TRAINLOOP_DATA_FOLDER
- Verify results exist:
ls $TRAINLOOP_DATA_FOLDER/results/
- 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​
- eval - Generate data for Studio UI
- Configuration - Configure Studio behavior
- Architecture - How Studio UI fits into the system