dvc params

Commands to display parameters

Description

DVC params commands help you track and compare parameters (hyperparameters, configuration values) across different experiments and commits. Parameters are typically stored in YAML, JSON, TOML, or Python files and define model training configurations.

Subcommands

params diff

Show changes in params between commits in the DVC repository, or between a commit and the workspace.

Compare HEAD with Workspace

dvc params diff

Arguments

a_rev · string · default: "HEAD"

Old Git commit to compare (defaults to HEAD).

b_rev · string · default: "workspace"

New Git commit to compare (defaults to the current workspace).

Options

--targets · list

Specific params file(s) to compare (even if not found as params in dvc.yaml). Shows all tracked params by default.

--all · flag

Show unchanged params as well.

--deps · flag

Show only params that are stage dependencies.

--json · flag

Show output in JSON format.

--md · flag

Show tabulated output in the Markdown format (GFM).

--no-path · flag

Don’t show params path.

Examples

Basic Parameter Comparison

$ dvc params diff

Path         Param              HEAD     workspace
params.yaml  model.learning_rate  0.001   0.01
params.yaml  model.batch_size     32      64
params.yaml  model.epochs         100     150

Compare Specific Files

$ dvc params diff --targets params.yaml config/train.yaml

This compares only the specified parameter files.

Show Only Changed Parameters (Default)

$ dvc params diff HEAD~3 HEAD

Shows parameters that changed in the last 3 commits.

Include Unchanged Parameters

$ dvc params diff --all

Path         Param                HEAD     workspace  
params.yaml  model.learning_rate  0.001    0.01       (changed)
params.yaml  model.batch_size     32       32         
params.yaml  model.epochs         100      150        (changed)

JSON Output for Automation

$ dvc params diff --json

{
  "params.yaml": {
    "model.learning_rate": {
      "old": 0.001,
      "new": 0.01,
      "diff": 0.009
    },
    "model.epochs": {
      "old": 100,
      "new": 150,
      "diff": 50
    }
  }
}

Use --json output with tools like jq for automated parameter validation in CI/CD pipelines.

Show Only Stage Dependencies

$ dvc params diff --deps

Filters to show only parameters that are declared as dependencies in pipeline stages.

The --deps flag is useful when you have many parameters but only want to see those that affect pipeline execution.

Parameter File Formats

DVC supports multiple parameter file formats:

YAML

params.yaml

model:
  learning_rate: 0.001
  batch_size: 32
  epochs: 100

preprocessing:
  normalize: true
  augmentation: false

JSON

params.json

{
  "model": {
    "learning_rate": 0.001,
    "batch_size": 32,
    "epochs": 100
  }
}

TOML

params.toml

[model]
learning_rate = 0.001
batch_size = 32
epochs = 100

Python

params.py

LEARNING_RATE = 0.001
BATCH_SIZE = 32
EPOCHS = 100

Use Cases

Hyperparameter Tracking

Track how hyperparameters change across experiments and their impact on metrics.

Experiment Comparison

Compare parameter configurations between different experiment runs.

Reproducibility

Ensure exact parameter values are recorded for reproducing results.

Configuration Management

Manage different configurations for development, staging, and production.

Parameters must be tracked in dvc.yaml or passed explicitly via --targets to be compared.

Related Commands

• dvc metrics - Compare metrics across experiments
• dvc plots - Visualize how parameters affect outcomes
• dvc exp show - Show both params and metrics together