neurenix run

Run a model training script with specified configuration options

The run command executes a Neurenix model training script with the ability to override configuration parameters from the command line.

Usage

neurenix run [script] [options]

Arguments

Argument	Type	Default	Description
script	string	train.py	Path to the Python training script to execute

Options

Option	Type	Default	Description
--config	string	config.json	Path to configuration file
--device	string	None	Device to use (cpu, cuda, auto)
--batch-size	integer	None	Batch size for training
--epochs	integer	None	Number of training epochs
--learning-rate	float	None	Learning rate for optimizer

Configuration Override

Command-line options override values in the configuration file:

1. Configuration file is loaded from --config path
2. Command-line options (if provided) override config values
3. Updated configuration is written back to the file
4. Script runs with the merged configuration

The script can access the configuration path via the NEURENIX_CONFIG environment variable.

Examples

Run with default settings

neurenix run train.py

Running train.py with configuration from config.json...
Epoch 1/10: loss=0.523, accuracy=0.812
Epoch 2/10: loss=0.412, accuracy=0.856
...
Script completed successfully.

Run with custom script and config

neurenix run scripts/custom_train.py --config configs/experiment1.json

Running scripts/custom_train.py with configuration from configs/experiment1.json...
Script completed successfully.

Override training parameters

neurenix run train.py --epochs 50 --batch-size 64 --learning-rate 0.0001

Running train.py with configuration from config.json...
Epoch 1/50: loss=0.628, accuracy=0.743
...
Script completed successfully.

Force CPU training

neurenix run train.py --device cpu

Running train.py with configuration from config.json...
Using device: cpu
Epoch 1/10: loss=0.523, accuracy=0.812
...
Script completed successfully.

Use GPU with auto-detection

neurenix run train.py --device auto

Running train.py with configuration from config.json...
Using device: cuda:0
Epoch 1/10: loss=0.523, accuracy=0.812
...
Script completed successfully.

Combine multiple overrides

neurenix run train.py \
  --config configs/resnet.json \
  --device cuda \
  --batch-size 128 \
  --epochs 100 \
  --learning-rate 0.001

Configuration File

The configuration file is typically a JSON file with the following structure:

{
  "model": {
    "type": "mlp",
    "layers": [128, 64, 10],
    "activation": "relu"
  },
  "training": {
    "batch_size": 32,
    "epochs": 10,
    "learning_rate": 0.001,
    "optimizer": "adam"
  },
  "hardware": {
    "device": "auto",
    "precision": "float32"
  }
}

Environment Variables

The run command sets the following environment variable for the training script:

• NEURENIX_CONFIG: Absolute path to the configuration file

Your training script can access this:

import os
import neurenix

config_path = os.environ.get('NEURENIX_CONFIG')
config = neurenix.load_config(config_path)

Error Handling

Script not found

neurenix run nonexistent.py

Error: Script 'nonexistent.py' not found.

Configuration file not found

neurenix run train.py --config missing.json

Error: Configuration file 'missing.json' not found.

Script execution error

neurenix run train.py

Running train.py with configuration from config.json...
Traceback (most recent call last):
  File "train.py", line 10, in <module>
    raise ValueError("Invalid configuration")
Error: Script exited with code 1

Best Practices

1. Use configuration files for experiments

Create separate config files for different experiments:

neurenix run train.py --config configs/baseline.json
neurenix run train.py --config configs/with_augmentation.json
neurenix run train.py --config configs/large_model.json

2. Override for quick iterations

Use command-line overrides for quick parameter sweeps:

for lr in 0.001 0.0001 0.00001; do
  neurenix run train.py --learning-rate $lr
done

3. Version control your configs

Keep configuration files in version control to track experiments:

git add configs/experiment_v1.json
git commit -m "Add experiment v1 configuration"

4. Use descriptive script names

neurenix run scripts/train_resnet50.py
neurenix run scripts/finetune_bert.py
neurenix run scripts/train_gan.py

See Also

• Init command - Initialize projects
• Monitor command - Monitor training
• Eval command - Evaluate models
• Save command - Save model state