neurenix save

Save the current project state including models, configurations, and data

The save command saves the current project state by creating a checkpoint that includes models, configurations, and optionally data and logs.

Usage

neurenix save [options]

Options

Option	Type	Default	Description
--name	string	checkpoint_{timestamp}	Name for the checkpoint
--include-data	flag	false	Include data directory in the checkpoint
--include-logs	flag	false	Include logs directory in the checkpoint
--output-dir	string	checkpoints	Output directory for checkpoints

Checkpoint Contents

A checkpoint always includes:

• models/: All saved model files
• configs/: Configuration files (or config.json if it exists)
• metadata.json: Checkpoint information (timestamp, name, Neurenix version, options)

Optionally includes:

• data/: Training/validation data (with --include-data)
• logs/: Training logs (with --include-logs)

Examples

Save basic checkpoint

neurenix save

Saving models...
Saving configurations...
Project state saved to checkpoints/checkpoint_1678901234

Save with custom name

neurenix save --name best_model_v1

Saving models...
Saving configurations...
Project state saved to checkpoints/best_model_v1

Save with data and logs

neurenix save --name experiment_1 --include-data --include-logs

Saving models...
Saving configurations...
Saving data...
Saving logs...
Project state saved to checkpoints/experiment_1

Save to custom directory

neurenix save --name milestone_v2 --output-dir backup/checkpoints

Saving models...
Saving configurations...
Project state saved to backup/checkpoints/milestone_v2

Full backup before deployment

neurenix save \
  --name production_deployment \
  --include-data \
  --include-logs \
  --output-dir backups

Saving models...
Saving configurations...
Saving data...
Saving logs...
Project state saved to backups/production_deployment

Checkpoint Metadata

Each checkpoint includes a metadata.json file with information about the save:

{
  "timestamp": 1678901234.567,
  "name": "best_model_v1",
  "neurenix_version": "0.1.0",
  "include_data": false,
  "include_logs": false
}

Use Cases

1. Save training milestones

Save checkpoints at key training milestones:

neurenix save --name epoch_50
neurenix save --name best_accuracy
neurenix save --name final_model

2. Version control for models

Create versioned checkpoints for model iterations:

neurenix save --name model_v1.0
neurenix save --name model_v1.1
neurenix save --name model_v2.0

3. Pre-deployment backup

Create a full backup before deploying to production:

neurenix save \
  --name pre_deployment_$(date +%Y%m%d) \
  --include-data \
  --include-logs

4. Experiment tracking

Save results of different experiments:

neurenix save --name baseline_model
neurenix save --name with_augmentation
neurenix save --name larger_batch_size

Restoring from Checkpoint

To restore a project from a checkpoint, copy the contents back to your project directory:

cp -r checkpoints/best_model_v1/* .

Or selectively restore components:

# Restore only models
cp -r checkpoints/best_model_v1/models/* models/

# Restore only configuration
cp checkpoints/best_model_v1/configs/config.json config.json

Best Practices

1. Use descriptive names

Give checkpoints meaningful names that indicate their purpose:

neurenix save --name baseline_acc_0.92
neurenix save --name after_hyperparameter_tuning
neurenix save --name final_production_model

2. Regular checkpoints during long training

Save periodic checkpoints during extended training sessions:

# In a training loop
for epoch in {10..100..10}; do
  neurenix save --name epoch_$epoch
done

3. Include data for reproducibility

When saving experiment results, include data to ensure reproducibility:

neurenix save --name paper_submission --include-data

4. Clean up old checkpoints

Regularly remove outdated checkpoints to save disk space:

# Keep only the 5 most recent checkpoints
ls -t checkpoints | tail -n +6 | xargs -I {} rm -r checkpoints/{}

See Also

• Run command - Train models
• Export command - Export to different formats
• Optimize command - Optimize saved models