Parameter Reference

synkro.generate()

Generate training traces from a policy document.

synkro.generate(
    policy: str | Policy,
    traces: int = 20,
    turns: int | str = "auto",
    dataset_type: DatasetType = DatasetType.CONVERSATION,
    generation_model: str = "gpt-5-mini",
    grading_model: str = "gpt-5.2",
    max_iterations: int = 3,
    skip_grading: bool = False,
    reporter: ProgressReporter | None = None,
    return_logic_map: bool = False,
    enable_hitl: bool = True,
    base_url: str | None = None,
    temperature: float = 0.7,
) -> Dataset | GenerationResult

Parameter	Type	Default	Description
`policy`	`str \| Policy`	required	Policy text or Policy object
`traces`	`int`	`20`	Number of traces to generate
`turns`	`int \| str`	`"auto"`	Conversation turns. `"auto"` for policy complexity-driven
`dataset_type`	`DatasetType`	`CONVERSATION`	Type of dataset
`generation_model`	`str`	`"gpt-5-mini"`	Model for generating responses
`grading_model`	`str`	`"gpt-5.2"`	Model for grading (use stronger model)
`max_iterations`	`int`	`3`	Max refinement iterations per trace
`skip_grading`	`bool`	`False`	Skip grading phase for faster generation
`reporter`	`ProgressReporter`	`None`	Progress reporter (default: RichReporter)
`return_logic_map`	`bool`	`False`	Return GenerationResult with Logic Map
`enable_hitl`	`bool`	`True`	Enable Human-in-the-Loop editing
`base_url`	`str`	`None`	API base URL for local LLM providers
`temperature`	`float`	`0.7`	Sampling temperature (0.0-2.0)

synkro.generate_scenarios()

Generate eval scenarios without synthetic responses.

synkro.generate_scenarios(
    policy: str | Policy,
    count: int = 100,
    generation_model: str = "gpt-4o-mini",
    temperature: float = 0.8,
    reporter: ProgressReporter | None = None,
    enable_hitl: bool = False,
    base_url: str | None = None,
) -> ScenariosResult

Parameter	Type	Default	Description
`policy`	`str \| Policy`	required	Policy text or Policy object
`count`	`int`	`100`	Number of scenarios to generate
`generation_model`	`str`	`"gpt-4o-mini"`	Model for generation
`temperature`	`float`	`0.8`	Sampling temperature
`reporter`	`ProgressReporter`	`None`	Progress reporter
`enable_hitl`	`bool`	`False`	Enable Human-in-the-Loop editing
`base_url`	`str`	`None`	API base URL

synkro.grade()

Grade a response against a scenario and policy.

synkro.grade(
    response: str,
    scenario: EvalScenario,
    policy: str | Policy,
    model: str = "gpt-4o",
    base_url: str | None = None,
) -> GradeResult

Parameter	Type	Default	Description
`response`	`str`	required	Model response to grade
`scenario`	`EvalScenario`	required	Eval scenario with expected outcome
`policy`	`str \| Policy`	required	Policy for grading context
`model`	`str`	`"gpt-4o"`	LLM for grading
`base_url`	`str`	`None`	API base URL

synkro.create_pipeline()

Create a customized generation pipeline.

synkro.create_pipeline(
    model: str = "gpt-4o-mini",
    grading_model: str = "gpt-4o",
    dataset_type: DatasetType = DatasetType.CONVERSATION,
    max_iterations: int = 3,
    skip_grading: bool = False,
    reporter: ProgressReporter | None = None,
    tools: list[ToolDefinition] | None = None,
    checkpoint_dir: str | None = None,
    enable_hitl: bool = True,
    base_url: str | None = None,
    temperature: float = 0.7,
    thinking: bool = False,
) -> Pipeline

Parameter	Type	Default	Description
`model`	`str`	`"gpt-4o-mini"`	Model for generating responses
`grading_model`	`str`	`"gpt-4o"`	Model for grading
`dataset_type`	`DatasetType`	`CONVERSATION`	Type of dataset
`max_iterations`	`int`	`3`	Max refinement iterations
`skip_grading`	`bool`	`False`	Skip grading phase
`reporter`	`ProgressReporter`	`None`	Progress reporter
`tools`	`list[ToolDefinition]`	`None`	Tool definitions for TOOL_CALL
`checkpoint_dir`	`str`	`None`	Directory for checkpointing
`enable_hitl`	`bool`	`True`	Enable Human-in-the-Loop
`base_url`	`str`	`None`	API base URL
`temperature`	`float`	`0.7`	Sampling temperature
`thinking`	`bool`	`False`	Enable `<think>` tags

Dataset.save()

Save dataset to a JSONL file.

dataset.save(
    path: str | Path | None = None,
    format: str = "messages",
    pretty_print: bool = False,
) -> Dataset

Parameter	Type	Default	Description
`path`	`str \| Path`	`None`	Output path (auto-generated if None)
`format`	`str`	`"messages"`	Output format
`pretty_print`	`bool`	`False`	Format with indentation

Format options: "messages", "qa", "langsmith", "langfuse", "tool_call", "chatml", "bert", "bert:<task>"

Dataset.filter()

Filter traces by criteria.

dataset.filter(
    passed: bool | None = None,
    category: str | None = None,
    min_length: int | None = None,
) -> Dataset

Parameter	Type	Default	Description
`passed`	`bool`	`None`	Filter by grade status
`category`	`str`	`None`	Filter by category
`min_length`	`int`	`None`	Minimum response length

Dataset.dedupe()

Remove duplicates.

dataset.dedupe(
    threshold: float = 0.85,
    method: str = "semantic",
    field: str = "user",
) -> Dataset

Parameter	Type	Default	Description
`threshold`	`float`	`0.85`	Similarity threshold (0-1)
`method`	`str`	`"semantic"`	Method: `"exact"` or `"semantic"`
`field`	`str`	`"user"`	Field: `"user"`, `"assistant"`, or `"both"`

Dataset.push_to_hub()

Push to HuggingFace Hub.

dataset.push_to_hub(
    repo_id: str,
    format: str = "messages",
    private: bool = False,
    split: str = "train",
    token: str | None = None,
) -> str

Parameter	Type	Default	Description
`repo_id`	`str`	required	HuggingFace repo ID
`format`	`str`	`"messages"`	Output format
`private`	`bool`	`False`	Private repo
`split`	`str`	`"train"`	Dataset split
`token`	`str`	`None`	HuggingFace token

ToolDefinition

Define a tool for TOOL_CALL datasets.

ToolDefinition(
    name: str,
    description: str,
    parameters: dict = {"type": "object", "properties": {}},
    examples: list[dict] = [],
    mock_responses: list[str] = [],
)

Parameter	Type	Default	Description
`name`	`str`	required	Tool name
`description`	`str`	required	Tool description
`parameters`	`dict`	`{}`	JSON Schema for parameters
`examples`	`list[dict]`	`[]`	Example tool calls
`mock_responses`	`list[str]`	`[]`	Mock responses for simulation

FileLoggingReporter

Log to file with optional console output.

FileLoggingReporter(
    delegate: ProgressReporter | None = None,
    log_dir: str = ".",
    log_filename: str | None = None,
)

Parameter	Type	Default	Description
`delegate`	`ProgressReporter`	`RichReporter()`	Reporter for display
`log_dir`	`str`	`"."`	Log directory
`log_filename`	`str`	`None`	Custom filename (auto if None)

CallbackReporter

Custom callbacks for progress events.

CallbackReporter(
    on_progress: Callable[[str, dict], None] | None = None,
    on_start: Callable[[int, str, str], None] | None = None,
    on_plan_complete: Callable[[Plan], None] | None = None,
    on_scenario_progress: Callable[[int, int], None] | None = None,
    on_scenarios_complete: Callable[[list[Scenario]], None] | None = None,
    on_response_progress: Callable[[int, int], None] | None = None,
    on_responses_complete: Callable[[list[Trace]], None] | None = None,
    on_grading_progress: Callable[[int, int], None] | None = None,
    on_grading_complete: Callable[[list[Trace], float], None] | None = None,
    on_complete: Callable[[int, float, float | None], None] | None = None,
)

Get Started

Core Concepts

Dataset Types

Guides

API Reference

Advanced

Examples

Reference

Parameter Reference

synkro.generate()

synkro.generate_scenarios()

synkro.grade()

synkro.create_pipeline()

Dataset.save()

Dataset.filter()

Dataset.dedupe()

Dataset.push_to_hub()

ToolDefinition

FileLoggingReporter

CallbackReporter

Get Started

Core Concepts

Dataset Types

Guides

API Reference

Advanced

Examples

Reference

​synkro.generate()

​synkro.generate_scenarios()

​synkro.grade()

​synkro.create_pipeline()

​Dataset.save()

​Dataset.filter()

​Dataset.dedupe()

​Dataset.push_to_hub()

​ToolDefinition

​FileLoggingReporter

​CallbackReporter

synkro.generate()

synkro.generate_scenarios()

synkro.grade()

synkro.create_pipeline()

Dataset.save()

Dataset.filter()

Dataset.dedupe()

Dataset.push_to_hub()

ToolDefinition

FileLoggingReporter

CallbackReporter