Skip to main contentsynth_ai.data.enums
Centralized enums for Synth AI SDK.
This module defines all enum types used across the SDK. Enums are pure data
with no dependencies on other modules.
Classes
JobType
Types of training/optimization jobs.
JobStatus
Status of a job.
InferenceMode
Inference modes for policy evaluation.
ProviderName
LLM provider names.
RewardSource
Source of reward signal for training.
RewardType
Category of reward signal.
RewardScope
Granularity level for reward observations.
ObjectiveKey
Canonical objective keys.
ObjectiveDirection
Optimization direction for an objective.
GraphType
Graph categories for training and inference.
OptimizationMode
Optimization strategy selection.
VerifierMode
Verifier evaluation mode.
TrainingType
Training/optimization category labels.
AdaptiveCurriculumLevel
Preset levels for adaptive pooling curriculum.
AdaptiveBatchLevel
Preset levels for adaptive batch curriculum (GEPA only).
SynthModelName
Synth-hosted model names for routing through Synth backend.
SuccessStatus
Infrastructure/runtime success status (orthogonal to reward).
SuccessStatus indicates whether a rollout completed without infrastructure
or runtime failures. This is orthogonal to reward - a rollout can have
SUCCESS status but low reward (bad AI output), or FAILURE status
(infrastructure error) regardless of AI output quality.
NEVER computed from reward, rubric scores, or task outcomes.
ONLY computed from infrastructure/runtime execution status.
OutputMode
Controls how the policy expects model outputs.
- TOOL_CALLS: Use function/tool calling (default, current behavior)
- TEXT: Plain text in message.content
- STRUCTURED: JSON via response_format (OpenAI json_schema, Groq json_object, Gemini responseSchema)
