data_juicer.ops.base_op module

data_juicer.ops.base_op.convert_list_dict_to_dict_list(samples)

data_juicer.ops.base_op.convert_dict_list_to_list_dict(samples)

data_juicer.ops.base_op.convert_arrow_to_python(method)

data_juicer.ops.base_op.catch_map_batches_exception(method, skip_op_error=False, op_name=None)

For batched-map sample-level fault tolerance.

data_juicer.ops.base_op.catch_map_single_exception(method, return_sample=True, skip_op_error=False, op_name=None)

For single-map sample-level fault tolerance. The input sample is expected batch_size = 1.

class data_juicer.ops.base_op.OP(*args, **kwargs)

Bases: object

__init__(*args, **kwargs)

Base class of operators.

Parameters:

• text_key – the key name of field that stores sample texts to be processed.

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

• index_key – index the samples before process if not None

• system_key – the key name of field that stores system prompts

• instruction_key – the key name of field that stores instruction

• index_key – the key name of field that stores index

• batch_size – the batch size for processing

• work_dir – the working directory for this operator

• skip_op_error – whether to skip the error when processing samples

# Ray related parameters :param num_cpus: number of CPUs required for this operator, only used when

running in Ray mode

Parameters:

• num_gpus – number of GPUs required for this operator, only used when running in Ray mode

• memory – memory size required for this operator, only used when running in Ray mode

• runtime_env – runtime environment for this operator, only used when running in Ray mode. More details can be found in Ray documentation.

• ray_execution_mode – execution mode in Ray, can be “actor” or “task” or None, if None, the “actor” mode is used when the operator is a CUDA operator, and the “task” mode is used if the operator is a CPU operator.

use_auto_proc()

is_batched_op()

use_ray_actor()

process(*args, **kwargs)

use_cuda()

runtime_np()

remove_extra_parameters(param_dict, keys=None)

at the beginning of the init of the mapper op, call self.remove_extra_parameters(locals()) to get the init parameter dict of the op for convenience

add_parameters(init_parameter_dict, **extra_param_dict)

add parameters for each sample, need to keep extra_param_dict and init_parameter_dict unchanged.

run(dataset)

empty_history()

class data_juicer.ops.base_op.Mapper(*args, **kwargs)

Bases: OP

__init__(*args, **kwargs)

Base class that conducts data editing.

Parameters:

• text_key – the key name of field that stores sample texts to be processed.

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

process_batched(samples, *args, **kwargs)

process_single(sample)

For sample level, sample –> sample

Parameters:

sample – sample to process

Returns:

processed sample

run(dataset, *, exporter=None, tracer=None)

class data_juicer.ops.base_op.Filter(*args, **kwargs)

Bases: OP

__init__(*args, **kwargs)

Base class that removes specific info.

Parameters:

• text_key – the key name of field that stores sample texts to be processed

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

• min_closed_interval – whether the min_val of the specified filter range is a closed interval. It’s True by default.

• max_closed_interval – whether the max_val of the specified filter range is a closed interval. It’s True by default.

• reversed_range – whether to reverse the target range [min_val, max_val] to (-∞, min_val) or (max_val, +∞). It’s False by default.

get_keep_boolean(val, min_val=None, max_val=None)

compute_stats_batched(samples, *args, **kwargs)

process_batched(samples)

compute_stats_single(sample, context=False)

Compute stats for the sample which is used as a metric to decide whether to filter this sample.

Parameters:

• sample – input sample.

• context – whether to store context information of intermediate vars in the sample temporarily.

Returns:

sample with computed stats

process_single(sample)

For sample level, sample –> Boolean.

Parameters:

sample – sample to decide whether to filter

Returns:

true for keeping and false for filtering

run(dataset, *, exporter=None, tracer=None, reduce=True)

class data_juicer.ops.base_op.Deduplicator(*args, **kwargs)

Bases: OP

__init__(*args, **kwargs)

Base class that conducts deduplication.

Parameters:

• text_key – the key name of field that stores sample texts to be processed

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

compute_hash(sample)

Compute hash values for the sample.

Parameters:

sample – input sample

Returns:

sample with computed hash value.

process(dataset, show_num=0)

For doc-level, dataset –> dataset.

Parameters:

• dataset – input dataset

• show_num – number of traced samples used when tracer is open.

Returns:

deduplicated dataset and the sampled duplicate pairs.

run(dataset, *, exporter=None, tracer=None, reduce=True)

class data_juicer.ops.base_op.Selector(*args, **kwargs)

Bases: OP

__init__(*args, **kwargs)

Base class that conducts selection in dataset-level.

Parameters:

• text_key – the key name of field that stores sample texts to be processed

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

process(dataset)

Dataset –> dataset.

Parameters:

dataset – input dataset

Returns:

selected dataset.

run(dataset, *, exporter=None, tracer=None)

class data_juicer.ops.base_op.Grouper(*args, **kwargs)

Bases: OP

__init__(*args, **kwargs)

Base class that group samples.

Parameters:

• text_key – the key name of field that stores sample texts to be processed

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

process(dataset)

Dataset –> dataset.

Parameters:

dataset – input dataset

Returns:

dataset of batched samples.

run(dataset, *, exporter=None, tracer=None)

class data_juicer.ops.base_op.Aggregator(*args, **kwargs)

Bases: OP

__init__(*args, **kwargs)

Base class that group samples.

Parameters:

• text_key – the key name of field that stores sample texts to be processed

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

process_single(sample)

For sample level, batched sample –> sample, the input must be the output of some Grouper OP.

Parameters:

sample – batched sample to aggregate

Returns:

aggregated sample

run(dataset, *, exporter=None, tracer=None)

class data_juicer.ops.base_op.Pipeline(*args, **kwargs)

Bases: OP

Base class for Operators that represent a data processing pipeline.

__init__(*args, **kwargs)

Base class of operators.

Parameters:

• text_key – the key name of field that stores sample texts to be processed.

• image_key – the key name of field that stores sample image list to be processed

• audio_key – the key name of field that stores sample audio list to be processed

• video_key – the key name of field that stores sample video list to be processed

• image_bytes_key – the key name of field that stores sample image bytes list to be processed

• query_key – the key name of field that stores sample queries

• response_key – the key name of field that stores responses

• history_key – the key name of field that stores history of queries and responses

• index_key – index the samples before process if not None

• batch_size – the batch size for processing

run(dataset)