qianfan.dataset package

Library aimed to helping developer to interactive with Dataset

class qianfan.dataset.DataExportDestinationType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: int, Enum

PlatformBos: int = 0

导出到平台 Bos

PrivateBos: int = 1

导出到私有 Bos

class qianfan.dataset.DataProjectType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: int, Enum

Project type used by Qianfan Data

Conversation: int = 20

对话类

GenericText: int = 401

返文本类

QuerySet: int = 402

Query 查询类

Text2Image: int = 705

文生图类

class qianfan.dataset.DataSetType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: int, Enum

MultiModel: int = 7

多模态数据集

TextOnly: int = 4

文本类数据集

class qianfan.dataset.DataSource

Bases: ABC

basic data source class

abstract async afetch(**kwargs: Any) → Union[str, List[str]]

Asynchronously fetch data from source

Args:

**kwargs (Any): optional arguments

Returns:

Union[str, List[str]]: content retrieved from data source

abstract async asave(data: str, **kwargs: Any) → bool

Asynchronously export the data to the data source and return whether the import was successful or failed

Args:

data (str): data need to be saved **kwargs (Any): optional arguments

Returns:

bool: is saving successful

abstract fetch(**kwargs: Any) → Union[str, List[str]]

Fetch data from source

Args:

**kwargs (Any): optional arguments

Returns:

Union[str, List[str]]: content retrieved from data source

abstract format_type() → FormatType

Get format type binding to source

Returns:

FormatType: format type binding to source

abstract save(data: str, **kwargs: Any) → bool

Export the data to the data source and return whether the import was successful or failed

Args:

data (str): data need to be saved **kwargs (Any): optional arguments

Returns:

bool: is saving successful

abstract set_format_type(format_type: FormatType) → None

Set format type binding to source

Args:

format_type (FormatType): format type binding to source

class qianfan.dataset.DataSourceType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: int, Enum

PrivateBos: int = 1

私有 Bos

SharedZipUrl: int = 2

包含 zip 压缩包的分享链接

class qianfan.dataset.DataStorageType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: str, Enum

PrivateBos: str = 'usrBos'

用户私有 Bos

PublicBos: str = 'sysBos'

平台公共 Bos

class qianfan.dataset.DataTemplateType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: int, Enum

Template type used by Qianfan Data

GenericText: int = 40100

泛文本

NonSortedConversation: int = 2000

非排序对话

QuerySet: int = 40200

Query 查询

SortedConversation: int = 2001

含排序对话

Text2Image: int = 70500

文生图

class qianfan.dataset.Dataset(inner_table: Table, inner_data_source_cache: Optional[DataSource] = None, inner_schema_cache: Optional[Schema] = None, input_columns: Optional[List[str]] = None, reference_column: Optional[str] = None, **kwargs: Any)

Bases: Table

add_default_group_column() → Self

add “_group” column to Dataset, the value in “_group” column are sequential incremental

Returns:

Self: Dataset itself

append(elem: Any, add_new_group: bool = False, is_grouped: bool = True) → Self

append element(s) to dataset

Args:

elem (Union[List[List[Dict]], List[Dict], Tuple[Dict], Dict]):

Elements added to dataset

add_new_group (bool):

Whether elem has a new group id. Only used when dataset is grouped.

is_grouped (bool):

Are element in elem in same group. Only used when dataset is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Dataset itself

async atest_using_llm(model_id: Optional[int] = None, model_version_id: Optional[int] = None, service_model: Optional[str] = None, service_endpoint: Optional[str] = None, is_chat_service: bool = True, does_show_latency: bool = True, **kwargs: Any) → Dataset

using arguments to init an llm instance and get output on current dataset from it asynchronously set only model arguments our service arguments to instantiating

Args:

model_id (Optional[int]):

id of your own model, default to None

model_version_id (Optional[int]):

version id of your own model, default to None

service_model (Optional[str]):

name of model you want to use as service, default to None

service_endpoint (Optional[str]):

endpoint of service, default to None

is_chat_service (bool):

the service type of service, default to True. Service will be Completion if False

does_show_latency (bool):

whether result dataset contain latency info column when using Service as evaluated object. Depending on different request mode (stream and non-stream), it will contains request_complete_latency or (first_token_latency, request_complete_latency) combo. Default to True

**kwargs (Any):

optional argument dict

Returns:

Dataset: A dataset contains inputs, reference outputs and llm outputs

col_append(elem: Any) → Self

append a row to dataset

Args:

elem (Dict[str, List]): a dict containing element added to dataset, which

must has column name “name” and column data list “data”

Returns:

Self: Dataset itself

col_delete(index: Union[int, str]) → Self

delete an column from dataset

Args:

index (str): column name to delete

Returns:

Self: Dataset itself

col_filter(op: Callable[[Any], bool]) → Self

filter on dataset’s column

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Dataset itself

col_insert(elem: Any, index: Any) → Self

append a row to dataset

Args:

elem (Dict[str, List]): dict containing element added to dataset

must has column name “name” and column data list “data”

index (int): where to insert new column

Returns:

Self: Dataset itself

col_list(by: Optional[Union[slice, int, str, List[int], Tuple[int], List[str], Tuple[str]]] = None) → Any

get column(s) from dataset

Args:

by (Optional[Union[int, str, Sequence[int], Sequence[str]]]):

index or indices for columns, default to None, in which case return a python list of dataset column

Returns:

Any: dataset column list

col_map(op: Callable[[Any], Any]) → Self

map on dataset’s column

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Dataset itself

col_names() → List[str]

get column name list

Returns:

List[str]: column name list

col_renames(new_names: List[str]) → Self

rename all dataset column

Args:

new_names (List[str]): All new names for columns

Returns:

Self: A brand-new Dataset with new name

classmethod create_from_pyarrow_table(table: Table, schema: Optional[Schema] = None, **kwargs: Any) → Dataset

create a dataset from pyarrow table

Args:

table (pyarrow):

pyarrow table object used to create dataset。

schema (Optional[Schema]):

schema used to validate before exporting data, default to None

**kwargs (Any):

optional arguments

Returns:

Dataset: a dataset instance

classmethod create_from_pyobj(data: Union[List[Dict[str, Any]], Dict[str, List]], schema: Optional[Schema] = None, **kwargs: Any) → Dataset

create a dataset from python dict or list

Args:

data (Union[List[Dict[str, Any]], Dict[str, List]]):

python object used to create dataset。

schema (Optional[Schema]):

schema used to validate before exporting data, default to None

**kwargs (Any):

optional arguments

Returns:

Dataset: a dataset instance

delete(index: Union[int, str]) → Self

delete an element from dataset

Args:

index (Union[int, str]): element index to delete

Returns:

Self: Dataset itself

delete_group_column() → Self

remove “_group” column from Dataset

Returns:

Self: Dataset itself

filter(op: Callable[[Any], bool]) → Self

filter on dataset

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Dataset itself

property get_input_data: Dict[str, List[Any]]

get input columns data in dataset

Returns:

Dict[str, List[Any]]: a dict

which indicates the “column name-column data” pairs

property get_reference_data: List[Any]

get reference data in dataset

Returns:

List[Any]: list of output data column

insert(elem: Any, index: Any, group_id: int = -1, add_new_group: bool = False, is_grouped: bool = True) → Self

insert element(s) to dataset

Args:

elem (Union[List[List[Dict]], List[Dict], Tuple[Dict], Dict]):

Elements added to dataset

index (int): where to insert element(s) group_id (int):

which group id you want to apply to new element(s). Default to -1, which means let group id be automatically inferred from table.

add_new_group (bool):

Whether elem has a new group id. Only used when dataset is grouped and group_id is -1

is_grouped (bool):

Are element in elem in same group. Only used when dataset is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Dataset itself

is_dataset_generic_text() → bool

tell whether current dataset is generic text dataset

Returns:

bool: whether current dataset is generic text dataset

is_dataset_located_in_qianfan() → bool

tell whether current dataset is cloud-based dataset

Returns:

bool: whether current dataset is cloud-based dataset

list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None, **kwargs: Any) → Any

get element(s) from dataset

Args:

by (Optional[Union[slice, int, Sequence[int]]]):

index or indices for elements, default to None, in which case return a python list of dataset row

Returns:

Any: dataset row list

classmethod load(source: Optional[DataSource] = None, data_file: Optional[str] = None, qianfan_dataset_id: Optional[int] = None, bos_load_args: Optional[Dict[str, Any]] = None, huggingface_dataset: Optional[Any] = None, schema: Optional[Schema] = None, organize_data_as_group: bool = False, **kwargs: Any) → Dataset

Read data from the source or create a source from the parameters and create a Table instance. If a schema is specified, perform validation after importing.

Args:

source (Optional[DataSource]): where dataset load from,

default to None，in which case, a datasource will be created inside dataset using parameters below

data_file (Optional[str]):

dataset local file path, default to None

qianfan_dataset_id (Optional[int]):

qianfan dataset ID, default to None

bos_load_args: (Optional[Dict[str, Any]]):

create a dataset and import initial dataset content from args

huggingface_dataset (Optional[Dict[str, Any], Any]):

Huggingface dataset object, only support DatasetDict and Dataset of Huggingface datasets.

schema (Optional[Schema]):

schema used to validate loaded data, default to None

organize_data_as_group (bool):

only available when data source’s format is FormatType.Jsonl. Indicates whether organize data within dataset in group format, default to False, and when it’s True, the default format will be a group-based 2D structure.

**kwargs (Any): optional arguments

Returns:

Dataset: a dataset instance

map(op: Callable[[Any], Any]) → Self

map on dataset

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Dataset itself

online_data_process(operators: List[QianfanOperator]) → Dict[str, Any]

create an online ETL task on qianfan

Args:

operators (List[QianfanOperator]): operators applied to ETL task

Returns:

Dict[str, Any]: ETL task info, contains 3 field:

is_succeeded (bool): whether ETL task succeed etl_task_id (Optional[int]): etl task id, only

exists when etl task is created successfully

new_dataset_id (Optional[int]): dataset id which

stores data after etl, only exists when etl task is succeeded

row_number() → int

get pyarrow table row count。

Returns:

int: row count。

save(destination: Optional[DataSource] = None, data_file: Optional[str] = None, qianfan_dataset_id: Optional[int] = None, qianfan_dataset_create_args: Optional[Dict[str, Any]] = None, schema: Optional[Schema] = None, replace_source: bool = False, **kwargs: Any) → bool

Write data to source if a schema has been passed, validate data before exporting

Args:

destination (Optional[DataSource]):

data source where dataset exports，default to None. in which case, a datasource will be created inside dataset using parameters below

data_file (Optional[str]):

dataset local file path, default to None

qianfan_dataset_id (Optional[int]):

qianfan dataset ID, default to None

qianfan_dataset_create_args: (Optional[Dict[str: Any]]):

create arguments for creating a bare dataset on qianfan, default to None

schema: (Optional[Schema]):

schema used to validate before exporting data, default to None

replace_source: (bool):

if replace the original source, default to False

kwargs (Any): optional arguments

Returns:

bool: is saving succeeded

start_online_data_process_task(operators: List[QianfanOperator]) → int

create an online ETL task on qianfan

Args:

operators (List[QianfanOperator]): operators applied to ETL task

Returns:

int: etl task id

test_using_llm(model_id: Optional[int] = None, model_version_id: Optional[int] = None, service_model: Optional[str] = None, service_endpoint: Optional[str] = None, is_chat_service: bool = True, does_show_latency: bool = True, **kwargs: Any) → Dataset

using arguments to init an llm instance and get output on current dataset from it set only model arguments our service arguments to instantiating

Args:

model_id (Optional[int]):

id of your own model, default to None

model_version_id (Optional[int]):

version id of your own model, default to None

service_model (Optional[str]):

name of model you want to use as service, default to None

service_endpoint (Optional[str]):

endpoint of service, default to None

is_chat_service (bool):

the service type of service, default to True. Service will be Completion if False

does_show_latency (bool):

whether result dataset contain latency info column when using Service as evaluated object. Depending on different request mode (stream and non-stream), it will contains request_complete_latency or (first_token_latency, request_complete_latency) combo. Default to True

**kwargs (Any):

optional argument dict

Returns:

Dataset: A dataset contains inputs, reference outputs and llm outputs

class qianfan.dataset.FileDataSource(*, path: str, file_format: Optional[FormatType] = None, save_as_folder: bool = False)

Bases: DataSource, BaseModel

file data source

async afetch(**kwargs: Any) → Union[str, List[str]]

Asynchronously Read data from file. Not available currently

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]:

String or list of string containing the data read from the file.

async asave(data: Union[str, List[str]], **kwargs: Any) → bool

Asynchronously Write data to file。 Not available currently

Args:

data (Union[str, List[str]]): data waiting to be written。 **kwargs (Any): optional arguments。

Returns:

bool: has data been written successfully

fetch(**kwargs: Any) → Union[str, List[str]]

Read data from file.

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]:

String or list of string containing the data read from the file.

file_format: Optional[FormatType]

format_type() → FormatType

Get format type binding to source

Returns:

FormatType: format type binding to source

path: str

save(data: Union[str, List[str]], **kwargs: Any) → bool

Write data to file。

Args:

data (Union[str, List[str]]): data waiting to be written。 **kwargs (Any): optional arguments。

Returns:

bool: has data been written successfully

save_as_folder: bool

set_format_type(format_type: FormatType) → None

Set format type binding to source

Args:

format_type (FormatType): format type binding to source

class qianfan.dataset.QianfanDataSource(*, id: int, group_id: int, name: str, set_type: DataSetType, project_type: DataProjectType, template_type: DataTemplateType, version: int, storage_type: DataStorageType, storage_id: str, storage_path: str, storage_raw_path: Optional[str] = None, storage_name: str, storage_region: Optional[str] = None, info: Dict[str, Any] = {}, download_when_init: bool = False, data_format_type: FormatType, ak: Optional[str] = None, sk: Optional[str] = None)

Bases: DataSource, BaseModel

Qianfan data source

async afetch(**kwargs: Any) → Union[str, List[str]]

Asynchronously read data from qianfan or local cache。 Not available currently

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]: content retrieved from data source

ak: Optional[str]

async asave(data: str, is_annotated: bool = False, **kwargs: Any) → bool

Asynchronously write data to qianfan currently only support to write to user BOS storage

Not available currently

Args:

data (str): data waiting to be uploaded。 is_annotated (bool): has data been annotated **kwargs (Any): optional arguments。

Returns:

bool: has data been uploaded successfully

classmethod create_bare_dataset(name: str, template_type: DataTemplateType, storage_type: DataStorageType = DataStorageType.PublicBos, storage_id: Optional[str] = None, storage_path: Optional[str] = None, addition_info: Optional[Dict[str, Any]] = None, ak: Optional[str] = None, sk: Optional[str] = None, **kwargs: Any) → QianfanDataSource

create bare dataset on qianfan as data source, which is empty Args:

name (str): dataset name you want template_type (DataTemplateType): template type applying to data set storage_type (Optional[DataStorageType]):

data storage type used to store your data, default to PublicBos

storage_id (Optional[str]): private BOS bucket name，

needed when storage_type is PrivateBos, default to None

storage_path (Optional[str]): private BOS file path，

needed when storage_type is PrivateBos, default to None

addition_info (Optional[Dict[str, Any]]):

additional info you want to have，default to None

ak (Optional[str]):

console ak related to your dataset and bos，default to None

sk (Optional[str]):

console sk related to your dataset and bos，default to None

kwargs (Any): other arguments

Returns:

QianfanDataSource: A datasource represents your dataset on Qianfan

classmethod create_from_bos_file(name: str, template_type: DataTemplateType, storage_id: str, storage_path: str, file_name: str, is_data_annotated: bool, storage_type: DataStorageType = DataStorageType.PrivateBos, addition_info: Optional[Dict[str, Any]] = None, ak: Optional[str] = None, sk: Optional[str] = None, is_download_to_local: bool = True, **kwargs: Any) → QianfanDataSource

create a dataset on qianfan as data source, which will import data from specific bos Args:

name (str): dataset name you want template_type (DataTemplateType): template type applying to data set storage_id (str): private BOS bucket name storage_path (str): private BOS file path file_name (str): file need to upload is_data_annotated (bool): is data in bos annotated storage_type (Optional[DataStorageType]):

data storage type used to store your data, default to PrivateBos

addition_info (Optional[Dict[str, Any]]):

additional info you want to have，default to None

ak (Optional[str]):

console ak related to your dataset and bos，default to None

sk (Optional[str]):

console sk related to your dataset and bos，default to None

is_download_to_local (bool):

does dataset download file when initialize object，default to True

kwargs (Any): other arguments

Returns:

QianfanDataSource: A datasource represents your dataset on Qianfan

data_format_type: FormatType

download_when_init: bool

fetch(**kwargs: Any) → Union[str, List[str]]

Read data from qianfan or local cache。

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]: content retrieved from data source

format_type() → FormatType

Get format type binding to qianfan data source

Returns:

FormatType: format type binding to qianfan data source

classmethod get_existed_dataset(dataset_id: int, is_download_to_local: bool = True, ak: Optional[str] = None, sk: Optional[str] = None, **kwargs: Any) → QianfanDataSource

Load a dataset from qianfan as data source

Args:

dataset_id (int): dataset id on Qianfan, show as “数据集版本 ID” is_download_to_local (bool):

does dataset download file when initialize object，default to True

ak (Optional[str]):

console ak related to your dataset and bos，default to None

sk (Optional[str]):

console sk related to your dataset and bos，default to None

kwargs (Any): other arguments

Returns:

QianfanDataSource: A datasource represents your dataset on Qianfan

group_id: int

id: int

info: Dict[str, Any]

name: str

project_type: DataProjectType

release_dataset(**kwargs: Any) → bool

make a dataset released

Returns:

bool: Whether releasing succeeded

save(data: Optional[str] = None, zip_file_path: Optional[str] = None, is_annotated: bool = False, does_release: bool = False, sup_storage_id: str = '', sup_storage_path: str = '', sup_storage_region: str = '', **kwargs: Any) → bool

Write data to qianfan Currently only support to write to user BOS storage

Args:

data (str): data waiting to be uploaded. Default to None zip_file_path (Optional[str]):

zip file path which contains data files, default to None.

is_annotated (bool): has data been annotated, default to False does_release (bool):

does release dataset after saving successfully, default to False

sup_storage_id (Optional[str]):

bos bucket name used for uploading, we recommend to use this parameter when your destination dataset on qianfan is stored in public BOS. Default to empty str

sup_storage_path (Optional[str]):

bos bucket file path used for uploading, we recommend to use this parameter when your destination dataset on qianfan is stored in public BOS. Default to empty str

sup_storage_region (Optional[str]):

bos bucket region used for uploading, we recommend to use this parameter when your destination dataset on qianfan is stored in public BOS. Default to empty str

**kwargs (Any): optional arguments。

Returns:

bool: has data been uploaded successfully

set_format_type(format_type: FormatType) → None

Set format type binding to qianfan data source Not available

TextOnly -> Jsonl MultiModel -> Json

set_type: DataSetType

sk: Optional[str]

storage_id: str

storage_name: str

storage_path: str

storage_raw_path: Optional[str]

storage_region: Optional[str]

storage_type: DataStorageType

template_type: DataTemplateType

version: int

class qianfan.dataset.Table(inner_table: Table)

Bases: Addable, Listable, Processable

dataset representation on memory inherited from pyarrow.Table，implementing interface in process_interface.py

append(elem: Any, add_new_group: bool = False, is_grouped: bool = True) → Self

append an element to pyarrow table

Args:

elem (Union[List[Dict], Tuple[Dict], Dict]): Elements added to pyarrow table add_new_group (bool):

Whether elem has a new group id. Only used when table is grouped.

is_grouped (bool):

Are element in elem in same group. Only used when table is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Table itself

col_append(elem: Any) → Self

append a row to pyarrow table

Args:

elem (Dict[str, List]): dict containing element added to pyarrow table

must has column name “name” and column data list “data”

Returns:

Self: Table itself

col_delete(index: Union[int, str]) → Self

delete a column from pyarrow table

Args:

index (str): column name to delete

Returns:

Self: Table itself

col_filter(op: Callable[[Any], bool]) → Self

filter on pyarrow table’s column

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Table itself

col_insert(elem: Any, index: Any) → Self

append a row to pyarrow table

Args:

elem (Dict[str, List]): dict containing element added to pyarrow table

must has column name “name” and column data list “data”

index (int): where to insert new column

Returns:

Self: Table itself

col_list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None) → Any

get column(s) from pyarrow table

Args:

by (Optional[Union[int, str, Sequence[int], Sequence[str]]]):

index or indices for columns, default to None, in which case return a python list of pyarrow table column

Returns:

Any: pyarrow table column list

col_map(op: Callable[[Any], Any]) → Self

map on pyarrow table’s column

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Table itself

col_names() → List[str]

get column name list

Returns:

List[str]: column name list

col_renames(new_names: List[str]) → Self

rename all dataset column

Args:

new_names (List[str]): All new names for columns

Returns:

Self: A brand-new Table with new name

column_number() → int

get pyarrow table column count。

Returns:

int: column count。

delete(index: Union[int, str]) → Self

delete an element from pyarrow table

Args:

index (Union[int, str]): element index to delete

Returns:

Self: Table itself

filter(op: Callable[[Any], bool]) → Self

filter on pyarrow table’s row

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Table itself

insert(elem: Any, index: Any, group_id: int = -1, add_new_group: bool = False, is_grouped: bool = True) → Self

insert an element to pyarrow table

Args:

elem (Union[List[Dict], Tuple[Dict], Dict]): Elements added to pyarrow table index (int): where to insert element(s) group_id (int):

which group id you want to apply to new element(s). Default to -1, which means let group id be automatically inferred from table.

add_new_group (bool):

Whether elem has a new group id. Only used when table is grouped and group_id is -1

is_grouped (bool):

Are element in elem in same group. Only used when table is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Table itself

is_dataset_grouped() → bool

is_dataset_packed() → bool

list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None) → Any

get element(s) from pyarrow table

Args:

by (Optional[Union[slice, int, Sequence[int]]]):

index or indices for elements, default to None, in which case return a python list of pyarrow table row

Returns:

Any: pyarrow table row list

map(op: Callable[[Any], Any]) → Self

map on pyarrow table’s row

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Table itself

pack() → bool

pack all group into 1 row and make table array-like with single column

Returns:

bool: whether packing succeeded

row_number() → int

get pyarrow table row count。

Returns:

int: row count。

to_pydict() → Dict

convert a pyarrow table to dict

Returns:

Dict: a dict

to_pylist() → List

convert a pyarrow table to list

Returns:

List: a list

unpack() → bool

unpack all element in the row “_pack” make sure the element in the column “_pack” is Sequence[Dict[str, Any]]

Returns:

bool: whether unpacking succeeded

Submodules

qianfan.dataset.consts module

constants for dataset using

qianfan.dataset.data_operator module

data operator for qianfan online not available currently

class qianfan.dataset.data_operator.DeduplicationSimhash(*, operator_name: str = 'deduplication_simhash', operator_type: str = 'deduplication', distance: float)

Bases: Deduplicator

Deduplicator class to deduplicate by simhash

distance: float

operator_name: str

class qianfan.dataset.data_operator.Deduplicator(*, operator_name: str, operator_type: str = 'deduplication')

Bases: QianfanOperator

Deduplicator class for online ETL operator

operator_type: str

class qianfan.dataset.data_operator.DesensitizationProcessor(*, operator_name: str, operator_type: str = 'desensitization')

Bases: QianfanOperator

Sensitive data processor class for online ETL operator

operator_type: str

class qianfan.dataset.data_operator.ExceptionRegulator(*, operator_name: str, operator_type: str = 'clean')

Bases: QianfanOperator

Exception class for online ETL operator

operator_type: str

class qianfan.dataset.data_operator.Filter(*, operator_name: str, operator_type: str = 'filter')

Bases: QianfanOperator

Filter class for online ETL operator

operator_type: str

class qianfan.dataset.data_operator.FilterCheckCharacterRepetitionRemoval(*, operator_name: str = 'filter_check_character_repetition_removal', operator_type: str = 'filter', default_character_repetition_max_cutoff: float)

Bases: Filter

Filter class to check character repetition removal

default_character_repetition_max_cutoff: float

operator_name: str

class qianfan.dataset.data_operator.FilterCheckFlaggedWords(*, operator_name: str = 'filter_check_flagged_words', operator_type: str = 'filter', flagged_words_max_cutoff: float)

Bases: Filter

Filter class to check flagged words

flagged_words_max_cutoff: float

operator_name: str

class qianfan.dataset.data_operator.FilterCheckLangId(*, operator_name: str = 'filter_check_lang_id', operator_type: str = 'filter', lang_id_min_cutoff: float)

Bases: Filter

Filter class to check lang id

lang_id_min_cutoff: float

operator_name: str

class qianfan.dataset.data_operator.FilterCheckNumberWords(*, operator_name: str = 'filter_check_number_words', operator_type: str = 'filter', number_words_min_cutoff: int = 1, number_words_max_cutoff: int = 10000)

Bases: Filter

Filter class to check number of words

number_words_max_cutoff: int

number_words_min_cutoff: int

operator_name: str

class qianfan.dataset.data_operator.FilterCheckPerplexity(*, operator_name: str = 'filter_check_perplexity', operator_type: str = 'filter', perplexity_max_cutoff: int)

Bases: Filter

Filter class to check perplexity

operator_name: str

perplexity_max_cutoff: int

class qianfan.dataset.data_operator.FilterCheckSpecialCharacters(*, operator_name: str = 'filter_check_special_characters', operator_type: str = 'filter', special_characters_max_cutoff: float)

Bases: Filter

Filter class to check special characters

operator_name: str

special_characters_max_cutoff: float

class qianfan.dataset.data_operator.FilterCheckWordRepetitionRemoval(*, operator_name: str = 'filter_check_word_repetition_removal', operator_type: str = 'filter', word_repetition_max_cutoff: float)

Bases: Filter

Filter class to check word repetition removal

operator_name: str

word_repetition_max_cutoff: float

class qianfan.dataset.data_operator.QianfanOperator(*, operator_name: str, operator_type: str)

Bases: BaseModel

Basic class for online ETL operator

operator_name: str

operator_type: str

class qianfan.dataset.data_operator.RemoveEmoji(*, operator_name: str = 'remove_emoji', operator_type: str = 'clean')

Bases: ExceptionRegulator

Exception class to remove emoji

operator_name: str

class qianfan.dataset.data_operator.RemoveInvisibleCharacter(*, operator_name: str = 'remove_invisible_character', operator_type: str = 'clean')

Bases: ExceptionRegulator

Exception class to remove invisible character

operator_name: str

class qianfan.dataset.data_operator.RemoveNonMeaningCharacters(*, operator_name: str = 'remove_non_meaning_characters', operator_type: str = 'clean')

Bases: ExceptionRegulator

Exception class to remove non-meaning characters

operator_name: str

class qianfan.dataset.data_operator.RemoveWebIdentifiers(*, operator_name: str = 'remove_web_identifiers', operator_type: str = 'clean')

Bases: ExceptionRegulator

Exception class to remove web identifiers

operator_name: str

class qianfan.dataset.data_operator.ReplaceEmails(*, operator_name: str = 'replace_emails', operator_type: str = 'desensitization')

Bases: DesensitizationProcessor

Sensitive data processor class to replace emails

operator_name: str

class qianfan.dataset.data_operator.ReplaceIdentifier(*, operator_name: str = 'replace_identifier', operator_type: str = 'desensitization')

Bases: DesensitizationProcessor

Sensitive data processor class to replace identifier

operator_name: str

class qianfan.dataset.data_operator.ReplaceIp(*, operator_name: str = 'replace_ip', operator_type: str = 'desensitization')

Bases: DesensitizationProcessor

Sensitive data processor class to replace ip

operator_name: str

class qianfan.dataset.data_operator.ReplaceTraditionalChineseToSimplified(*, operator_name: str = 'replace_traditional_chinese_to_simplified', operator_type: str = 'clean')

Bases: ExceptionRegulator

Exception class to replace traditional chinese to simplified

operator_name: str

class qianfan.dataset.data_operator.ReplaceUniformWhitespace(*, operator_name: str = 'replace_uniform_whitespace', operator_type: str = 'clean')

Bases: ExceptionRegulator

Exception class to replace uniform whitespace

operator_name: str

qianfan.dataset.data_source module

data source which is related to download/upload

class qianfan.dataset.data_source.DataSource

Bases: ABC

basic data source class

abstract async afetch(**kwargs: Any) → Union[str, List[str]]

Asynchronously fetch data from source

Args:

**kwargs (Any): optional arguments

Returns:

Union[str, List[str]]: content retrieved from data source

abstract async asave(data: str, **kwargs: Any) → bool

Asynchronously export the data to the data source and return whether the import was successful or failed

Args:

data (str): data need to be saved **kwargs (Any): optional arguments

Returns:

bool: is saving successful

abstract fetch(**kwargs: Any) → Union[str, List[str]]

Fetch data from source

Args:

**kwargs (Any): optional arguments

Returns:

Union[str, List[str]]: content retrieved from data source

abstract format_type() → FormatType

Get format type binding to source

Returns:

FormatType: format type binding to source

abstract save(data: str, **kwargs: Any) → bool

Export the data to the data source and return whether the import was successful or failed

Args:

data (str): data need to be saved **kwargs (Any): optional arguments

Returns:

bool: is saving successful

abstract set_format_type(format_type: FormatType) → None

Set format type binding to source

Args:

format_type (FormatType): format type binding to source

class qianfan.dataset.data_source.FileDataSource(*, path: str, file_format: Optional[FormatType] = None, save_as_folder: bool = False)

Bases: DataSource, BaseModel

file data source

async afetch(**kwargs: Any) → Union[str, List[str]]

Asynchronously Read data from file. Not available currently

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]:

String or list of string containing the data read from the file.

async asave(data: Union[str, List[str]], **kwargs: Any) → bool

Asynchronously Write data to file。 Not available currently

Args:

data (Union[str, List[str]]): data waiting to be written。 **kwargs (Any): optional arguments。

Returns:

bool: has data been written successfully

fetch(**kwargs: Any) → Union[str, List[str]]

Read data from file.

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]:

String or list of string containing the data read from the file.

file_format: Optional[FormatType]

format_type() → FormatType

Get format type binding to source

Returns:

FormatType: format type binding to source

path: str

save(data: Union[str, List[str]], **kwargs: Any) → bool

Write data to file。

Args:

data (Union[str, List[str]]): data waiting to be written。 **kwargs (Any): optional arguments。

Returns:

bool: has data been written successfully

save_as_folder: bool

set_format_type(format_type: FormatType) → None

Set format type binding to source

Args:

format_type (FormatType): format type binding to source

class qianfan.dataset.data_source.FormatType(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)

Bases: Enum

Enum for data source format type

Csv = 'csv'

Json = 'json'

Jsonl = 'jsonl'

Text = 'txt'

class qianfan.dataset.data_source.QianfanDataSource(*, id: int, group_id: int, name: str, set_type: DataSetType, project_type: DataProjectType, template_type: DataTemplateType, version: int, storage_type: DataStorageType, storage_id: str, storage_path: str, storage_raw_path: Optional[str] = None, storage_name: str, storage_region: Optional[str] = None, info: Dict[str, Any] = {}, download_when_init: bool = False, data_format_type: FormatType, ak: Optional[str] = None, sk: Optional[str] = None)

Bases: DataSource, BaseModel

Qianfan data source

async afetch(**kwargs: Any) → Union[str, List[str]]

Asynchronously read data from qianfan or local cache。 Not available currently

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]: content retrieved from data source

ak: Optional[str]

async asave(data: str, is_annotated: bool = False, **kwargs: Any) → bool

Asynchronously write data to qianfan currently only support to write to user BOS storage

Not available currently

Args:

data (str): data waiting to be uploaded。 is_annotated (bool): has data been annotated **kwargs (Any): optional arguments。

Returns:

bool: has data been uploaded successfully

classmethod create_bare_dataset(name: str, template_type: DataTemplateType, storage_type: DataStorageType = DataStorageType.PublicBos, storage_id: Optional[str] = None, storage_path: Optional[str] = None, addition_info: Optional[Dict[str, Any]] = None, ak: Optional[str] = None, sk: Optional[str] = None, **kwargs: Any) → QianfanDataSource

create bare dataset on qianfan as data source, which is empty Args:

name (str): dataset name you want template_type (DataTemplateType): template type applying to data set storage_type (Optional[DataStorageType]):

data storage type used to store your data, default to PublicBos

storage_id (Optional[str]): private BOS bucket name，

needed when storage_type is PrivateBos, default to None

storage_path (Optional[str]): private BOS file path，

needed when storage_type is PrivateBos, default to None

addition_info (Optional[Dict[str, Any]]):

additional info you want to have，default to None

ak (Optional[str]):

console ak related to your dataset and bos，default to None

sk (Optional[str]):

console sk related to your dataset and bos，default to None

kwargs (Any): other arguments

Returns:

QianfanDataSource: A datasource represents your dataset on Qianfan

classmethod create_from_bos_file(name: str, template_type: DataTemplateType, storage_id: str, storage_path: str, file_name: str, is_data_annotated: bool, storage_type: DataStorageType = DataStorageType.PrivateBos, addition_info: Optional[Dict[str, Any]] = None, ak: Optional[str] = None, sk: Optional[str] = None, is_download_to_local: bool = True, **kwargs: Any) → QianfanDataSource

create a dataset on qianfan as data source, which will import data from specific bos Args:

name (str): dataset name you want template_type (DataTemplateType): template type applying to data set storage_id (str): private BOS bucket name storage_path (str): private BOS file path file_name (str): file need to upload is_data_annotated (bool): is data in bos annotated storage_type (Optional[DataStorageType]):

data storage type used to store your data, default to PrivateBos

addition_info (Optional[Dict[str, Any]]):

additional info you want to have，default to None

ak (Optional[str]):

console ak related to your dataset and bos，default to None

sk (Optional[str]):

console sk related to your dataset and bos，default to None

is_download_to_local (bool):

does dataset download file when initialize object，default to True

kwargs (Any): other arguments

Returns:

QianfanDataSource: A datasource represents your dataset on Qianfan

data_format_type: FormatType

download_when_init: bool

fetch(**kwargs: Any) → Union[str, List[str]]

Read data from qianfan or local cache。

Args:

**kwargs (Any): Arbitrary keyword arguments.

Returns:

Union[str, List[str]]: content retrieved from data source

format_type() → FormatType

Get format type binding to qianfan data source

Returns:

FormatType: format type binding to qianfan data source

classmethod get_existed_dataset(dataset_id: int, is_download_to_local: bool = True, ak: Optional[str] = None, sk: Optional[str] = None, **kwargs: Any) → QianfanDataSource

Load a dataset from qianfan as data source

Args:

dataset_id (int): dataset id on Qianfan, show as “数据集版本 ID” is_download_to_local (bool):

does dataset download file when initialize object，default to True

ak (Optional[str]):

console ak related to your dataset and bos，default to None

sk (Optional[str]):

console sk related to your dataset and bos，default to None

kwargs (Any): other arguments

Returns:

QianfanDataSource: A datasource represents your dataset on Qianfan

group_id: int

id: int

info: Dict[str, Any]

name: str

project_type: DataProjectType

release_dataset(**kwargs: Any) → bool

make a dataset released

Returns:

bool: Whether releasing succeeded

save(data: Optional[str] = None, zip_file_path: Optional[str] = None, is_annotated: bool = False, does_release: bool = False, sup_storage_id: str = '', sup_storage_path: str = '', sup_storage_region: str = '', **kwargs: Any) → bool

Write data to qianfan Currently only support to write to user BOS storage

Args:

data (str): data waiting to be uploaded. Default to None zip_file_path (Optional[str]):

zip file path which contains data files, default to None.

is_annotated (bool): has data been annotated, default to False does_release (bool):

does release dataset after saving successfully, default to False

sup_storage_id (Optional[str]):

bos bucket name used for uploading, we recommend to use this parameter when your destination dataset on qianfan is stored in public BOS. Default to empty str

sup_storage_path (Optional[str]):

bos bucket file path used for uploading, we recommend to use this parameter when your destination dataset on qianfan is stored in public BOS. Default to empty str

sup_storage_region (Optional[str]):

bos bucket region used for uploading, we recommend to use this parameter when your destination dataset on qianfan is stored in public BOS. Default to empty str

**kwargs (Any): optional arguments。

Returns:

bool: has data been uploaded successfully

set_format_type(format_type: FormatType) → None

Set format type binding to qianfan data source Not available

TextOnly -> Jsonl MultiModel -> Json

set_type: DataSetType

sk: Optional[str]

storage_id: str

storage_name: str

storage_path: str

storage_raw_path: Optional[str]

storage_region: Optional[str]

storage_type: DataStorageType

template_type: DataTemplateType

version: int

qianfan.dataset.dataset module

dataset core concept, a wrap of data processing, data transmission and data validation

class qianfan.dataset.dataset.Dataset(inner_table: Table, inner_data_source_cache: Optional[DataSource] = None, inner_schema_cache: Optional[Schema] = None, input_columns: Optional[List[str]] = None, reference_column: Optional[str] = None, **kwargs: Any)

Bases: Table

add_default_group_column() → Self

add “_group” column to Dataset, the value in “_group” column are sequential incremental

Returns:

Self: Dataset itself

append(elem: Any, add_new_group: bool = False, is_grouped: bool = True) → Self

append element(s) to dataset

Args:

elem (Union[List[List[Dict]], List[Dict], Tuple[Dict], Dict]):

Elements added to dataset

add_new_group (bool):

Whether elem has a new group id. Only used when dataset is grouped.

is_grouped (bool):

Are element in elem in same group. Only used when dataset is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Dataset itself

async atest_using_llm(model_id: Optional[int] = None, model_version_id: Optional[int] = None, service_model: Optional[str] = None, service_endpoint: Optional[str] = None, is_chat_service: bool = True, does_show_latency: bool = True, **kwargs: Any) → Dataset

using arguments to init an llm instance and get output on current dataset from it asynchronously set only model arguments our service arguments to instantiating

Args:

model_id (Optional[int]):

id of your own model, default to None

model_version_id (Optional[int]):

version id of your own model, default to None

service_model (Optional[str]):

name of model you want to use as service, default to None

service_endpoint (Optional[str]):

endpoint of service, default to None

is_chat_service (bool):

the service type of service, default to True. Service will be Completion if False

does_show_latency (bool):

whether result dataset contain latency info column when using Service as evaluated object. Depending on different request mode (stream and non-stream), it will contains request_complete_latency or (first_token_latency, request_complete_latency) combo. Default to True

**kwargs (Any):

optional argument dict

Returns:

Dataset: A dataset contains inputs, reference outputs and llm outputs

col_append(elem: Any) → Self

append a row to dataset

Args:

elem (Dict[str, List]): a dict containing element added to dataset, which

must has column name “name” and column data list “data”

Returns:

Self: Dataset itself

col_delete(index: Union[int, str]) → Self

delete an column from dataset

Args:

index (str): column name to delete

Returns:

Self: Dataset itself

col_filter(op: Callable[[Any], bool]) → Self

filter on dataset’s column

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Dataset itself

col_insert(elem: Any, index: Any) → Self

append a row to dataset

Args:

elem (Dict[str, List]): dict containing element added to dataset

must has column name “name” and column data list “data”

index (int): where to insert new column

Returns:

Self: Dataset itself

col_list(by: Optional[Union[slice, int, str, List[int], Tuple[int], List[str], Tuple[str]]] = None) → Any

get column(s) from dataset

Args:

by (Optional[Union[int, str, Sequence[int], Sequence[str]]]):

index or indices for columns, default to None, in which case return a python list of dataset column

Returns:

Any: dataset column list

col_map(op: Callable[[Any], Any]) → Self

map on dataset’s column

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Dataset itself

col_names() → List[str]

get column name list

Returns:

List[str]: column name list

col_renames(new_names: List[str]) → Self

rename all dataset column

Args:

new_names (List[str]): All new names for columns

Returns:

Self: A brand-new Dataset with new name

classmethod create_from_pyarrow_table(table: Table, schema: Optional[Schema] = None, **kwargs: Any) → Dataset

create a dataset from pyarrow table

Args:

table (pyarrow):

pyarrow table object used to create dataset。

schema (Optional[Schema]):

schema used to validate before exporting data, default to None

**kwargs (Any):

optional arguments

Returns:

Dataset: a dataset instance

classmethod create_from_pyobj(data: Union[List[Dict[str, Any]], Dict[str, List]], schema: Optional[Schema] = None, **kwargs: Any) → Dataset

create a dataset from python dict or list

Args:

data (Union[List[Dict[str, Any]], Dict[str, List]]):

python object used to create dataset。

schema (Optional[Schema]):

schema used to validate before exporting data, default to None

**kwargs (Any):

optional arguments

Returns:

Dataset: a dataset instance

delete(index: Union[int, str]) → Self

delete an element from dataset

Args:

index (Union[int, str]): element index to delete

Returns:

Self: Dataset itself

delete_group_column() → Self

remove “_group” column from Dataset

Returns:

Self: Dataset itself

filter(op: Callable[[Any], bool]) → Self

filter on dataset

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Dataset itself

property get_input_data: Dict[str, List[Any]]

get input columns data in dataset

Returns:

Dict[str, List[Any]]: a dict

which indicates the “column name-column data” pairs

property get_reference_data: List[Any]

get reference data in dataset

Returns:

List[Any]: list of output data column

insert(elem: Any, index: Any, group_id: int = -1, add_new_group: bool = False, is_grouped: bool = True) → Self

insert element(s) to dataset

Args:

elem (Union[List[List[Dict]], List[Dict], Tuple[Dict], Dict]):

Elements added to dataset

index (int): where to insert element(s) group_id (int):

which group id you want to apply to new element(s). Default to -1, which means let group id be automatically inferred from table.

add_new_group (bool):

Whether elem has a new group id. Only used when dataset is grouped and group_id is -1

is_grouped (bool):

Are element in elem in same group. Only used when dataset is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Dataset itself

is_dataset_generic_text() → bool

tell whether current dataset is generic text dataset

Returns:

bool: whether current dataset is generic text dataset

is_dataset_located_in_qianfan() → bool

tell whether current dataset is cloud-based dataset

Returns:

bool: whether current dataset is cloud-based dataset

list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None, **kwargs: Any) → Any

get element(s) from dataset

Args:

by (Optional[Union[slice, int, Sequence[int]]]):

index or indices for elements, default to None, in which case return a python list of dataset row

Returns:

Any: dataset row list

classmethod load(source: Optional[DataSource] = None, data_file: Optional[str] = None, qianfan_dataset_id: Optional[int] = None, bos_load_args: Optional[Dict[str, Any]] = None, huggingface_dataset: Optional[Any] = None, schema: Optional[Schema] = None, organize_data_as_group: bool = False, **kwargs: Any) → Dataset

Read data from the source or create a source from the parameters and create a Table instance. If a schema is specified, perform validation after importing.

Args:

source (Optional[DataSource]): where dataset load from,

default to None，in which case, a datasource will be created inside dataset using parameters below

data_file (Optional[str]):

dataset local file path, default to None

qianfan_dataset_id (Optional[int]):

qianfan dataset ID, default to None

bos_load_args: (Optional[Dict[str, Any]]):

create a dataset and import initial dataset content from args

huggingface_dataset (Optional[Dict[str, Any], Any]):

Huggingface dataset object, only support DatasetDict and Dataset of Huggingface datasets.

schema (Optional[Schema]):

schema used to validate loaded data, default to None

organize_data_as_group (bool):

only available when data source’s format is FormatType.Jsonl. Indicates whether organize data within dataset in group format, default to False, and when it’s True, the default format will be a group-based 2D structure.

**kwargs (Any): optional arguments

Returns:

Dataset: a dataset instance

map(op: Callable[[Any], Any]) → Self

map on dataset

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Dataset itself

online_data_process(operators: List[QianfanOperator]) → Dict[str, Any]

create an online ETL task on qianfan

Args:

operators (List[QianfanOperator]): operators applied to ETL task

Returns:

Dict[str, Any]: ETL task info, contains 3 field:

is_succeeded (bool): whether ETL task succeed etl_task_id (Optional[int]): etl task id, only

exists when etl task is created successfully

new_dataset_id (Optional[int]): dataset id which

stores data after etl, only exists when etl task is succeeded

row_number() → int

get pyarrow table row count。

Returns:

int: row count。

save(destination: Optional[DataSource] = None, data_file: Optional[str] = None, qianfan_dataset_id: Optional[int] = None, qianfan_dataset_create_args: Optional[Dict[str, Any]] = None, schema: Optional[Schema] = None, replace_source: bool = False, **kwargs: Any) → bool

Write data to source if a schema has been passed, validate data before exporting

Args:

destination (Optional[DataSource]):

data source where dataset exports，default to None. in which case, a datasource will be created inside dataset using parameters below

data_file (Optional[str]):

dataset local file path, default to None

qianfan_dataset_id (Optional[int]):

qianfan dataset ID, default to None

qianfan_dataset_create_args: (Optional[Dict[str: Any]]):

create arguments for creating a bare dataset on qianfan, default to None

schema: (Optional[Schema]):

schema used to validate before exporting data, default to None

replace_source: (bool):

if replace the original source, default to False

kwargs (Any): optional arguments

Returns:

bool: is saving succeeded

start_online_data_process_task(operators: List[QianfanOperator]) → int

create an online ETL task on qianfan

Args:

operators (List[QianfanOperator]): operators applied to ETL task

Returns:

int: etl task id

test_using_llm(model_id: Optional[int] = None, model_version_id: Optional[int] = None, service_model: Optional[str] = None, service_endpoint: Optional[str] = None, is_chat_service: bool = True, does_show_latency: bool = True, **kwargs: Any) → Dataset

using arguments to init an llm instance and get output on current dataset from it set only model arguments our service arguments to instantiating

Args:

model_id (Optional[int]):

id of your own model, default to None

model_version_id (Optional[int]):

version id of your own model, default to None

service_model (Optional[str]):

name of model you want to use as service, default to None

service_endpoint (Optional[str]):

endpoint of service, default to None

is_chat_service (bool):

the service type of service, default to True. Service will be Completion if False

does_show_latency (bool):

whether result dataset contain latency info column when using Service as evaluated object. Depending on different request mode (stream and non-stream), it will contains request_complete_latency or (first_token_latency, request_complete_latency) combo. Default to True

**kwargs (Any):

optional argument dict

Returns:

Dataset: A dataset contains inputs, reference outputs and llm outputs

qianfan.dataset.dataset_utils module

utilities dataset needs

qianfan.dataset.dataset_utils.log_latency_info(result: QfResponse, index: int, stream_index: int = 1) → Tuple

qianfan.dataset.process_interface module

interface file

class qianfan.dataset.process_interface.Addable

Bases: ABC

make object ‘addable’

abstract append(elem: Any) → Self

append an element at Appendable object

Args:

elem (Any): element to append

Returns:

Self: a new Addable object after appending

abstract insert(elem: Any, index: Any) → Self

insert an element to Appendable object

Args:

elem (Any): element(s) to insert index (Any): where to insert element(s)

Returns:

Self: a new Addable object after inserting

class qianfan.dataset.process_interface.Listable

Bases: ABC

make object ‘listable’

abstract list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None) → Any

get an element from object

Args:

by (Optional[Union[slice, int, str, Sequence[int], Sequence[str]]):

index used to get data or data list, default to None

Returns:

Any: elements

class qianfan.dataset.process_interface.Processable

Bases: ABC

make object ‘processable’

abstract delete(index: Union[int, str]) → Self

delete an element from Processable object

Args:

index (Union[int, str]): element index to delete

Returns:

Self: a new Processable object after delete

abstract filter(op: Callable[[Any], bool]) → Self

filter on a Processable object

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: a new Processable object after filtering

abstract map(op: Callable[[Any], Any]) → Self

map on a Processable object

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: a new Processable object after mapping

qianfan.dataset.schema module

schema for validation currently qianfan schema only

class qianfan.dataset.schema.QianfanGenericText

Bases: QianfanSchema

validator for generic text dataset

validate(table: Table) → bool

validate a table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

class qianfan.dataset.schema.QianfanNonSortedConversation

Bases: QianfanSchema

validator for non-sorted, conversational dataset

validate(table: Table) → bool

validate a table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

class qianfan.dataset.schema.QianfanQuerySet

Bases: QianfanSchema

validator for query set dataset

validate(table: Table) → bool

validate a table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

class qianfan.dataset.schema.QianfanSchema

Bases: Schema

validate(table: Table) → bool

validate a dataset.Table object currently check field and type only, not content in table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

class qianfan.dataset.schema.QianfanSortedConversation

Bases: QianfanSchema

validator for sorted, conversational dataset

validate(table: Table) → bool

validate a table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

class qianfan.dataset.schema.QianfanText2Image

Bases: QianfanSchema

validator for text to image dataset

validate(table: Table) → bool

validate a table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

class qianfan.dataset.schema.Schema

Bases: ABC

abstract validate(table: Table) → bool

validate a dataset.Table object currently check field and type only, not content in table

Args:

table (Table): table need to be validated

Returns:

bool:whether table is valid

qianfan.dataset.table module

wrapper for pyarrow.Table

class qianfan.dataset.table.Table(inner_table: Table)

Bases: Addable, Listable, Processable

dataset representation on memory inherited from pyarrow.Table，implementing interface in process_interface.py

append(elem: Any, add_new_group: bool = False, is_grouped: bool = True) → Self

append an element to pyarrow table

Args:

elem (Union[List[Dict], Tuple[Dict], Dict]): Elements added to pyarrow table add_new_group (bool):

Whether elem has a new group id. Only used when table is grouped.

is_grouped (bool):

Are element in elem in same group. Only used when table is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Table itself

col_append(elem: Any) → Self

append a row to pyarrow table

Args:

elem (Dict[str, List]): dict containing element added to pyarrow table

must has column name “name” and column data list “data”

Returns:

Self: Table itself

col_delete(index: Union[int, str]) → Self

delete a column from pyarrow table

Args:

index (str): column name to delete

Returns:

Self: Table itself

col_filter(op: Callable[[Any], bool]) → Self

filter on pyarrow table’s column

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Table itself

col_insert(elem: Any, index: Any) → Self

append a row to pyarrow table

Args:

elem (Dict[str, List]): dict containing element added to pyarrow table

must has column name “name” and column data list “data”

index (int): where to insert new column

Returns:

Self: Table itself

col_list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None) → Any

get column(s) from pyarrow table

Args:

by (Optional[Union[int, str, Sequence[int], Sequence[str]]]):

index or indices for columns, default to None, in which case return a python list of pyarrow table column

Returns:

Any: pyarrow table column list

col_map(op: Callable[[Any], Any]) → Self

map on pyarrow table’s column

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Table itself

col_names() → List[str]

get column name list

Returns:

List[str]: column name list

col_renames(new_names: List[str]) → Self

rename all dataset column

Args:

new_names (List[str]): All new names for columns

Returns:

Self: A brand-new Table with new name

column_number() → int

get pyarrow table column count。

Returns:

int: column count。

delete(index: Union[int, str]) → Self

delete an element from pyarrow table

Args:

index (Union[int, str]): element index to delete

Returns:

Self: Table itself

filter(op: Callable[[Any], bool]) → Self

filter on pyarrow table’s row

Args:

op (Callable[[Any], bool]): handler used to filter

Returns:

Self: Table itself

insert(elem: Any, index: Any, group_id: int = -1, add_new_group: bool = False, is_grouped: bool = True) → Self

insert an element to pyarrow table

Args:

elem (Union[List[Dict], Tuple[Dict], Dict]): Elements added to pyarrow table index (int): where to insert element(s) group_id (int):

which group id you want to apply to new element(s). Default to -1, which means let group id be automatically inferred from table.

add_new_group (bool):

Whether elem has a new group id. Only used when table is grouped and group_id is -1

is_grouped (bool):

Are element in elem in same group. Only used when table is grouped and elem is Sequence and add_new_group was set True. Default to True, all elements will be in same group. If it’s True, each element will have sequential incremental group id from last available group id.

Returns:

Self: Table itself

is_dataset_grouped() → bool

is_dataset_packed() → bool

list(by: Optional[Union[slice, int, str, Sequence[int], Sequence[str]]] = None) → Any

get element(s) from pyarrow table

Args:

by (Optional[Union[slice, int, Sequence[int]]]):

index or indices for elements, default to None, in which case return a python list of pyarrow table row

Returns:

Any: pyarrow table row list

map(op: Callable[[Any], Any]) → Self

map on pyarrow table’s row

Args:

op (Callable[[Any], Any]): handler used to map

Returns:

Self: Table itself

pack() → bool

pack all group into 1 row and make table array-like with single column

Returns:

bool: whether packing succeeded

row_number() → int

get pyarrow table row count。

Returns:

int: row count。

to_pydict() → Dict

convert a pyarrow table to dict

Returns:

Dict: a dict

to_pylist() → List

convert a pyarrow table to list

Returns:

List: a list

unpack() → bool

unpack all element in the row “_pack” make sure the element in the column “_pack” is Sequence[Dict[str, Any]]

Returns:

bool: whether unpacking succeeded

qianfan.dataset.table_utils module

utilities table needs