Use a dictionary input when related values should travel together as one configuration object. Keep separate typed inputs when each value needs its own type, validation, UI field, or command-line override.

Pass a dictionary from the CLI

The following component accepts one required dictionary named config. The toEnv setting exposes its resolved value as TRAINING_CONFIG:

version: 1.1
kind: component
name: dictionary-config

inputs:
- name: config
  type: dict
  toEnv: TRAINING_CONFIG

run:
  kind: job
  init:
  - file:
      filename: main.py
      content: |
        import json
        import os

        config = json.loads(os.environ["TRAINING_CONFIG"])
        print(f"learning_rate={config['learning_rate']}")
        print(f"epochs={config['epochs']}")
  container:
    image: python:3.11
    workingDir: "{{ globals.artifacts_path }}"
    command: [python3, main.py]

Save the component as config.yaml, then pass the dictionary as JSON:

polyaxon run -f config.yaml \
  -P config='{"learning_rate":0.003,"epochs":20}' \
  -l

The shell's single quotes preserve the JSON string. Polyaxon parses it as a dictionary, validates it against the config input, and serializes it as JSON in TRAINING_CONFIG. The program prints:

learning_rate=0.003
epochs=20

Polyaxon records config as one run input. Its nested keys are not separate inputs, so they cannot be compared or overridden independently. Declare learning_rate and epochs as separate inputs when you need that behavior.

Forward inputs and outputs from another operation

In a DAG, a downstream operation can request the complete input or output dictionary of an upstream operation. The receiving component declares dictionary inputs:

inputs:
- name: training_inputs
  type: dict
- name: training_outputs
  type: dict

The DAG operation maps the upstream contexts to those inputs:

- name: summarize
  dagRef: summarize-component
  params:
    training_inputs:
      ref: ops.training
      value: "{{ inputs }}"
    training_outputs:
      ref: ops.training
      value: "{{ outputs }}"

{{ inputs }} and {{ outputs }} refer to the declared values of ops.training. They do not expose outputs that the current operation may produce later. The upstream operation must make an output available before a downstream operation can consume it.

See Context Params for individual and complete context references. Continue to Pipelines & Orchestration for complete DAG examples.