pyromind-sdk 0.0.26rc1

Lightweight SDK stub for local development and testing of third-party nodes without the full platform codebase

PyroMind Node SDK

A lightweight SDK stub for local development and testing of third-party nodes without the full platform codebase (without app.models.nodes).

In the real platform runtime environment, nodes should prioritize importing base classes from app.models.nodes.

Installation

pip install pyromind-sdk

Usage

YAML-based Node Configuration

Define nodes using YAML files with the parameters format:

from pyromind_sdk import load_nodes_from_yaml

# Load nodes from YAML file
nodes = load_nodes_from_yaml("my_node.yaml")
MyNode = nodes["MyNode"]

# Use the node class
print(MyNode.DESCRIPTION)
print(MyNode.BASE_INPUT_TYPES())

Parameter Format

parameters:
  - name: input0
    type: input
    required_type: required
    dtype: INT
    default: 1
    min: 1
    max: 8
  - name: input1
    type: input
    required_type: optional
    dtype: [STRING, PATH]          # list for union type
  - name: input2
    type: input
    required_type: required
    dtype: STRING
    limit:
      enum: ["train", "eval"]     # dropdown choices
    default: "train"
  - name: output
    type: output
    dtype: STRING

Available dtype values: STRING, INT, FLOAT, BOOLEAN, PATH, MODEL, ENV, ACCELERATE_CONFIG, * (any), ANY.

Union types use a list: [STRING, PATH]. Constraints (min, max, step, enum) go in the limit block and are validated against dtype compatibility.

Example YAML Node Configuration

name: MyNode
description: "My custom node"
base_class: PodExecutionNode

command_template:
  - "sh"
  - "-c"
  - "echo \"Hello, {{name}}!\" > {{output}}"

parameters:
  - name: name
    dtype: STRING
    default: "World"
    type: input
    required_type: required
  - name: score
    dtype: FLOAT
    type: input
    required_type: required
    default: 0.5
    min: 0.0
    max: 1.0
  - name: path_or_model
    dtype: [STRING, PATH]          # union — accepts STRING and non-basic types
    type: input
    required_type: optional
  - name: mode
    dtype: STRING
    type: input
    required_type: required
    limit:
      enum: ["train", "eval", "predict"]
    default: "train"
  - name: output
    dtype: STRING
    type: output

Main Classes

Base Node Classes

Base node classes are available for reference in YAML configurations. You can specify them in your YAML files using the base_class field:

• PodExecutionNode: Base class for Pod execution nodes
• PortPodExecutionNode: Pod execution node with port resource
• DaemonPodExecutionNode: Daemon Pod execution node
• GpuPodExecutionNode: GPU Pod execution node
• JupyterLabPodExecutionNode: Pod execution node with JupyterLab environment
• EndpointNode: Base class for endpoint nodes
• NodeType: Node type enumeration

These base classes are used internally by the YAML loader and should be referenced by name in your YAML configurations, not imported directly in Python code.

YAML Nodes Functions

• load_nodes_from_yaml(yaml_path): Load nodes from a YAML file
• load_all_nodes_from_directory(directory): Load all nodes from a directory
• create_node_class_from_yaml(yaml_config, class_name): Create a node class from YAML config
• yaml_to_node_class(yaml_path): Convert YAML config to Python class object

Python Function Nodes

You can also create nodes that execute Python functions directly:

name: CalculatorNode
description: "A calculator node using Python function"
base_class: JupyterLabPodExecutionNode

# Python function configuration
python_code: "utils/calculator.py"      # Python file path (relative to YAML file or absolute path)
function_name: "calculate"               # Function name

# Execution environment configuration (optional)
python_command: "python3"                # Python execution command (default: python3)
# conda_env: "base"                      # Conda environment name (optional, default: "base")
# workdir: "/workspace/project"          # Working directory (optional)
# environment:                           # Environment variables (optional)
#   PYTHONUNBUFFERED: "1"

parameters:
  - name: input0
    type: input
    dtype: FLOAT
    required_type: required
    default: 0.0
  - name: input1
    type: input
    dtype: FLOAT
    required_type: required
    default: 0.0
  - name: result_input0
    type: output
    dtype: STRING
  - name: result_output0
    type: output
    dtype: STRING

The corresponding Python function (utils/calculator.py):

def calculate(input0: float, input1: float) -> dict:
    """Perform arithmetic operations"""
    output0 = input0 + input1
    return {
        "result_input0": str(input0),
        "result_output0": str(output0),
    }

Auto Generate: Python Function -> YAML

You can generate YAML config directly from a Python function signature and a return dict literal:

from pyromind_sdk import python_function_to_yaml

config = python_function_to_yaml(
    python_file_path="pyromind_sdk/examples/nodes/utils/calculator.py",
    function_name="calculate",
    node_name="PythonCalculatorNode",
    output_path="pyromind_sdk/examples/nodes/python_calculator_node.generated.yaml",
)

Auto-generate rules:

• Inputs are generated from function parameters in order
• Input dtype is inferred from annotations: str → STRING, int → INT, float → FLOAT, bool → BOOLEAN, Path → PATH
• Inputs are generated as required_type: optional with no default
• Outputs are generated only from return { ... } dict literals
• Return dict keys must be string literals
• Unknown annotations are passed through as the annotation name (no fallback to STRING)
• Generated YAML python_code is emitted as an absolute path

CLI 用法（写入到 YAML 文件）:

python -m pyromind_sdk.cli python-to-yaml \
  pyromind_sdk/examples/nodes/utils/calculator.py \
  calculate \
  --node-name PythonCalculatorNode \
  --output pyromind_sdk/examples/nodes/python_calculator_node.generated.yaml

如果不传 --output，会把 YAML 直接打印到 stdout。

Note on Python file paths:

• Relative paths are resolved relative to the YAML file's directory
• Absolute paths are used as-is
• The Python file must exist and be accessible at the specified path

Note on JupyterLab environment:

• When using JupyterLabPodExecutionNode, the Python code will be executed in a JupyterLab environment
• Conda environment activation is handled automatically (default: base environment)
• The command execution uses bash -c with conda activation, so shell operators like && are preserved

Note on accelerate mode:

• Accelerate mode is enabled only when python_command: "accelerate" (exact match after trimming spaces).
• Values that only start with accelerate (for example accelerate launch --num_processes 2) are treated as normal command strings.
• In accelerate mode, the node must inherit GpuPodExecutionNode.
• In accelerate mode, the SDK reads the input parameter with dtype ACCELERATE_CONFIG, writes it into /tmp/accelerate_config_<uuid>.yaml, and starts with accelerate launch --config_file <tmp_file> ....
• ACCELERATE_CONFIG is injected by runtime automatically, so YAML does not need to declare it in parameters.

Advanced Features

Resource Configuration

Configure CPU, memory, and GPU resources:

resources:
  memory_limit: 16      # Memory in GiB
  cpu_limit: 4          # CPU cores
  gpu_min_count: 1      # Minimum GPU count
  gpu_max_count: 8      # Maximum GPU count

Customer Inputs

Mark inputs/outputs for customer use (not used in command templates):

parameters:
  - name: customer_param
    type: input
    dtype: STRING
    required_type: required
    customer_use: true   # Mark as customer use

Multiple Base Classes

Support for multiple inheritance. You can combine multiple base classes to meet your node's requirements:

base_class:
  - GpuPodExecutionNode
  - JupyterLabPodExecutionNode

When to use each base class:

• PodExecutionNode: Basic Pod execution node. Use this for standard command execution without special requirements.

• GpuPodExecutionNode: Required if your node needs GPU resources. This base class provides GPU configuration options (gpu_count, gpu_product) and ensures GPU resources are allocated. If you specify GPU resources in the resources section or need GPU access, you must inherit from this class.

• JupyterLabPodExecutionNode: Required if your node needs to execute in a JupyterLab environment. Use this when you need interactive Python execution, notebook support, or Jupyter-specific features.

• PortPodExecutionNode: Required if your node needs port resources. This base class provides port configuration options for services that need to expose ports.

• DaemonPodExecutionNode: Use for daemon-style Pod execution nodes that run continuously in the background.

• EndpointNode: Use for nodes that return endpoint URLs. This base class automatically sets the return type to STRING with name "endpoint".

Examples:

# Simple node without special requirements
base_class: PodExecutionNode

# GPU-enabled node
base_class: GpuPodExecutionNode

# GPU + JupyterLab environment
base_class:
  - GpuPodExecutionNode
  - JupyterLabPodExecutionNode

# Port resource node
base_class: PortPodExecutionNode

API Reference

Core Functions

Loading Nodes

• load_nodes_from_yaml(yaml_path: str) -> Dict[str, type]: Load nodes from a YAML file
• load_all_nodes_from_directory(directory: str) -> Dict[str, type]: Load all nodes from a directory

Node Creation

• create_node_class_from_yaml(yaml_config: Dict, class_name: str, yaml_file_path: Optional[str] = None) -> type: Create a node class from YAML config

Conversion

• yaml_to_node_class(yaml_path: str) -> type: Convert YAML config to Python class object

Node Validation

• validate_node_class(node_class: type, node_name: str) -> Dict[str, Any]: Validate node class structure
• print_node_info(node_name: str, node_class: type, validation: Dict, execution_result: Optional[Dict] = None): Print detailed node information

Command Execution

• execute_command_template(command_template: List[str], inputs: Optional[Dict] = None, output_names: Optional[List[str]] = None, timeout: int = 300) -> Dict[str, Any]: Execute command template

Type Conversion

• convert_string_to_python_type(value: str, type_spec: Any) -> Any: Convert string value to Python type (supports INT → int, FLOAT → float, BOOLEAN → bool; PATH/MODEL/ENV → str)
• convert_inputs(inputs: Dict, input_types: Dict) -> Dict: Convert input values according to type definitions
• validate_output_type(value: Any, type_spec: str) -> bool: Validate output value type (supports STRING, PATH, MODEL, ENV, INT, FLOAT, BOOLEAN)

Workflow Functions

• WorkflowLiteConverter: Workflow lite format converter
• LayoutGenerator: Auto layout generator
• to_workflow_lite(workflow: Dict) -> Dict: Convert standard workflow to lite format
• to_workflow_standard(workflow: Dict) -> Dict: Convert lite workflow to standard format
• validate_workflow(workflow: Dict, format: str = 'lite') -> ValidationResult: Validate workflow format

Workflow Validation

• validate_lite_format(workflow: Dict) -> ValidationResult: Validate lite format workflow
• validate_standard_format(workflow: Dict) -> ValidationResult: Validate standard format workflow
• validate_workflow_lite(workflow: Dict) -> ValidationResult: Validate lite workflow
• validate_workflow_standard(workflow: Dict) -> ValidationResult: Validate standard workflow
• validate_workflow_legacy(workflow: Dict) -> ValidationResult: Validate legacy format workflow

Exception Classes

• PyroMindAPIError: API error exception
• ValidationError: Workflow validation error
• SchemaValidationError: Workflow schema validation error
• LinkValidationError: Workflow link validation error
• TypeValidationError: Workflow type validation error

Testing

Test your YAML node configurations:

# Test a single YAML file
python -m pyromind_sdk.tests.test_yaml_nodes hello_world_node.yaml

# Test with verbose output
python -m pyromind_sdk.tests.test_yaml_nodes hello_world_node.yaml --verbose

# Execute the command template
python -m pyromind_sdk.tests.test_yaml_nodes hello_world_node.yaml --execute

# Test with custom inputs
python -m pyromind_sdk.tests.test_yaml_nodes hello_world_node.yaml --execute --inputs '{"name": "Alice"}'

# Test all YAML files in a directory
python -m pyromind_sdk.tests.test_yaml_nodes --directory examples

Examples

Check the examples/ directory for more examples:

• hello_world_node.yaml: Basic node example
• echo_node.yaml: Simple command execution
• python_calculator_node.yaml: Python function node with multiple inputs/outputs
• jupyter_gpu_node.yaml: Jupyter GPU execution example
• accelerate_gpu_node.yaml: Accelerate launch example for GPU Python nodes
• multiline_text_node.yaml: Multiline text processing
• customer_inputs_node.yaml: Customer inputs example

Features

• ✅ Base Node Classes: All standard node base classes for local development
• ✅ YAML Configuration: Define nodes using YAML files (Python class definitions are not supported)
• ✅ Dynamic Loading: Load nodes at runtime without code changes
• ✅ Multiple Inheritance: Support for multiple base classes in YAML
• ✅ Python Function Nodes: Execute Python functions directly in nodes via YAML configuration
• ✅ Type Conversion: Automatic type conversion and validation (INT, FLOAT, BOOLEAN, STRING, PATH, MODEL, ENV)
• ✅ Dtype Constraints: Input constraints (min, max, step, enum) validated against dtype compatibility
• ✅ Resource Management: Configure CPU, memory, and GPU resources
• ✅ Customer Inputs: Mark inputs/outputs for customer-specific use
• ✅ Security: Built-in validation and security checks
• ✅ Workflow Conversion: Support conversion between standard and lite formats
• ✅ Workflow Validation: Comprehensive workflow validation

Requirements

• Python >= 3.8
• pyyaml >= 6.0

Development

Project Structure

pyromind_sdk/
├── pyromind_sdk/
│   ├── common/          # Common utilities and base classes
│   ├── nodes/           # Node loading and execution
│   ├── examples/        # Example YAML configurations
│   └── tests/          # Test utilities
├── setup.py
├── pyproject.toml
└── README.md

Contributing

Contributions are welcome! Please ensure:

1. All code comments are in English
2. Follow PEP 8 style guidelines
3. Add tests for new features
4. Update documentation as needed

License

MIT License

Links

• Website: https://pyromind.ai/
• PyPI: https://pypi.org/project/pyromind-sdk/
• Documentation: https://github.com/pyromind/pyromind-sdk