Class InferenceConfig

src/google/adk/evaluation/base_eval_service.py:58–95

Contains configurations need to run inferences.

class InferenceConfig(BaseModel):
  """Contains configurations need to run inferences."""

  model_config = ConfigDict(
      alias_generator=alias_generators.to_camel,
      populate_by_name=True,
  )

  labels: Optional[dict[str, str]] = Field(
      default=None,
      description="""Labels with user-defined metadata to break down billed
charges.""",
  )

  parallelism: int = Field(
      default=4,
      description="""Number of parallel inferences to run during an Eval. Few
factors to consider while changing this value:

1) Your available quota with the model. Models tend to enforce per-minute or
per-second SLAs. Using a larger value could result in the eval quickly consuming
the quota.

2) The tools used by the Agent could also have their SLA. Using a larger value
could also overwhelm those tools.""",
  )

  use_live: bool = Field(
      default=False,
      description="""Whether to use live (bidirectional streaming) mode for
inference. This is required for Live API models (e.g., gemini-*-live-*).""",
  )

  live_timeout_seconds: int = Field(
      default=DEFAULT_LIVE_TIMEOUT_SECONDS,
      description="""Timeout in seconds for waiting for model turn completion in
live mode.""",
  )


class InferenceRequest(BaseModel):