Abstract module for generating data. The generate_text function is the primary entrypoint for creating text.

class gretel_synthetics.generate.BaseGenerator

Do not use directly.

Specific generation modules should have a subclass of this ABC that implements the core logic for generating data

class gretel_synthetics.generate.GenText(valid: bool = None, text: str = None, explain: str = None, delimiter: str = None)

alias of pred_string

class gretel_synthetics.generate.SeedingGenerator(config: None, *, seed_list: List[str], line_validator: Callable | None = None, max_invalid: int = 1000)

A single threaded line / text generator that is specifically for using with a list of seeds. This also exposes the Settings class back to the caller so the actual seed list can be directly accessed, which controls the underlying progression of the main text generator.

This is useful when you need to manipulate the actual seed list as data is being generated.

class gretel_synthetics.generate.Settings(config: None, start_string: str | List[str] | None = None, multi_seed: bool = False, line_validator: Callable | None = None, max_invalid: int = 1000, tokenizer: BaseTokenizer | None = None, generator: BaseGenerator | None = None)

Do not use directly.

Arguments for a generator generating lines of text.

This class contains basic settings for a generation process. It is separated from the Generator class for ensuring reliable serializability without an excess amount of code tied to it.

This class also will take a provided start string and validate that it can be utilized for text generation. If the start_string is something other than the default, we have to do a couple things:

  1. If the config utilizes a field delimiter, the start_string MUST end with that delimiter

  2. Convert the user-facing delim char into the special delim token specified in the config

class gretel_synthetics.generate.gen_text(valid: bool | None = None, text: str | None = None, explain: str | None = None, delimiter: str | None = None)

A record that is yielded from the Generator.generate_next generator.


True, False, or None. If the line passed a validation function, then this will be True. If the validation function raised an exception then this will be automatically set to False. If no validation function is used, then this value will be None.




The actual record as a string




A string that describes why a record failed validation. This is the string representation of the Exception that is thrown in a validation function. This will only be set if validation fails, otherwise will be None.




If the generated text are column/field based records. This will hold the delimiter used to separate the fields from each other.



as_dict() dict

Serialize the generated record to a dictionary

values_as_list() List[str] | None

Attempt to split the generated text on the provided delimiter


A list of values that are separated by the object’s delimiter or None is there is no delimiter in the text

gretel_synthetics.generate.generate_text(config: None, start_string: str | List[str] | None = None, line_validator: Callable | None = None, max_invalid: int = 1000, num_lines: int | None = None, parallelism: int = 0) Iterator[GenText]

A generator that will load a model and start creating records.

  • config – A configuration object, which you must have created previously

  • start_string

    A prefix string that is used to seed the record generation. By default we use a newline, but you may substitue any initial value here which will influence how the generator predicts what to generate. If you are working with a field delimiter, and you want to seed more than one column value, then you MUST utilize the field delimiter specified in your config. An example would be “foo,bar,baz,”. Also, if using a field delimiter, the string MUST end with the delimiter value.


    This param may also be a list of prefixes. If this is provided, then the generator will attempt to create exactly 1 record for each seed in the list. The num_lines param will be implicity set to the size of the list and this number of records will be created at a 1:1 ratio between prefix strings and valid records.

  • line_validator – An optional callback validator function that will take the raw string value from the generator as a single argument. This validator can executue arbitrary code with the raw string value. The validator function may return a bool to indicate line validity. This boolean value will be set on the yielded gen_text object. Additionally, if the validator throws an exception, the gen_text object will be set with a failed validation. If the validator returns None, we will assume successful validation.

  • max_invalid – If using a line_validator, this is the maximum number of invalid lines to generate. If the number of invalid lines exceeds this value a RunTimeError will be raised.

  • num_lines

    If not None, this will override the gen_lines value that is provided in the config. .. note:

    If ``start_string`` is a list, this value will be set to the length of that list and any other
    values for the param are ignored.

  • parallelism – The number of concurrent workers to use. 1 (the default) disables parallelization, while a non-positive value means “number of CPUs + x” (i.e., use 0 for using as many workers as there are CPUs). A floating-point value is interpreted as a fraction of the available CPUs, rounded down.

Simple validator example:

def my_validator(raw_line: str):
    parts = raw_line.split(',')
    if len(parts) != 5:
        raise Exception('record does not have 5 fields')


gen_lines from the config is important for this function. If a line validator is not provided, each line will count towards the number of total generated lines. When the total lines generated is >= gen_lines we stop. If a line validator is provided, only valid lines will count towards the total number of lines generated. When the total number of valid lines generated is >= gen_lines, we stop.


gen_chars, controls the possible maximum number of characters a single generated line can have. If a newline character has not been generated before reaching this number, then the line will be returned. For example if gen_chars is 180 and a newline has not been generated, once 180 chars have been created, the line will be returned no matter what. As a note, if this value is 0, then each line will generate until a newline is observed.


A GenText object for each record that is generated. The generator will stop after the max number of lines is reached (based on your config).


A RunTimeError if the max_invalid number of lines is generated