dragon.ai.inference.config.InferenceConfig

class InferenceConfig[source] 

Bases: object

Master configuration for the entire inference pipeline.

Composes the model, hardware, batching, guardrails, and dynamic worker sections consumed by dragon.ai.inference.Inference. Only model is required for direct construction. All other fields have defaults suitable for a single shared backend in agentic pipelines.

Note

Direct construction and from_dict() have different defaults for guardrails and dynamic workers. Direct construction defaults both to disabled through this class’s default factories. The YAML-compatible from_dict() path defaults missing guardrails.toggle_on and dynamic_inf_wrkr.toggle_on values to True to match config.sample.

Parameters:

model (ModelConfig) – Required model and generation settings.
hardware (HardwareConfig) – Node and GPU allocation settings.
batching (BatchingConfig) – Request batching settings.
guardrails (GuardrailsConfig) – Prompt guardrails settings.
dynamic_worker (DynamicWorkerConfig) – Dynamic inference worker lifecycle settings.
flask_secret_key (str ) – Secret key retained for compatibility with application configurations that include a Flask service.
run_type (str ) – Application run mode label used by drivers and examples.
token (str ) – Application token string used by drivers that require one.

__init__(model: ModelConfig, hardware: HardwareConfig = <factory>, batching: BatchingConfig = <factory>, guardrails: GuardrailsConfig = <factory>, dynamic_worker: DynamicWorkerConfig = <factory>, flask_secret_key: str = '', run_type: str = 'backend_only', token: str = '') → None 

Methods

`__init__`(model, hardware, batching, ...)
`from_dict`(config_dict)	Create InferenceConfig from dictionary (loaded from YAML).
`validate_all`(all_nodes)	Validate all configuration sections.

Attributes

`flask_secret_key`
`run_type`
`token`
`model`
`hardware`
`batching`
`guardrails`
`dynamic_worker`

model: ModelConfig

hardware: HardwareConfig

batching: BatchingConfig

guardrails: GuardrailsConfig

dynamic_worker: DynamicWorkerConfig

flask_secret_key: str = ''

run_type: str = 'backend_only'

token: str = ''

classmethod from_dict(config_dict: dict ) → InferenceConfig[source] 

Create InferenceConfig from dictionary (loaded from YAML).

Parameters:: config_dict (dict ) – Configuration dictionary loaded from YAML.
Returns:: InferenceConfig instance.
Return type:: InferenceConfig

validate_all(all_nodes: dict ) → None [source] 

Validate all configuration sections.

Parameters:: all_nodes (dict ) – Dictionary of all available nodes in the cluster. Keys are hostnames, values are dragon.native.machine.Node objects.
Raises:: ValueError – If any configuration parameter is invalid.

__init__(model: ModelConfig, hardware: HardwareConfig = <factory>, batching: BatchingConfig = <factory>, guardrails: GuardrailsConfig = <factory>, dynamic_worker: DynamicWorkerConfig = <factory>, flask_secret_key: str = '', run_type: str = 'backend_only', token: str = '') → None 