sparknlp.annotator.GPT2Transformer#

class sparknlp.annotator.GPT2Transformer(classname='com.johnsnowlabs.nlp.annotators.seq2seq.GPT2Transformer', java_model=None)[source]#

Bases: sparknlp.common.AnnotatorModel, sparknlp.common.HasBatchedAnnotate

GPT2: the OpenAI Text-To-Text Transformer

GPT-2 is a large transformer-based language model with 1.5 billion parameters, trained on a dataset of 8 million web pages. GPT-2 is trained with a simple objective: predict the next word, given all of the previous words within some text. The diversity of the dataset causes this simple goal to contain naturally occurring demonstrations of many tasks across diverse domains. GPT-2 is a direct scale-up of GPT, with more than 10X the parameters and trained on more than 10X the amount of data.

GPT-2 displays a broad set of capabilities, including the ability to generate conditional synthetic text samples of unprecedented quality, where we prime the model with an input and have it generate a lengthy continuation. In addition, GPT-2 outperforms other language models trained on specific domains (like Wikipedia, news, or books) without needing to use these domain-specific training datasets. On language tasks like question answering, reading comprehension, summarization, and translation, GPT-2 begins to learn these tasks from the raw text, using no task-specific training data. While scores on these downstream tasks are far from state-of-the-art, they suggest that the tasks can benefit from unsupervised techniques, given sufficient (unlabeled) data and compute.

Pretrained models can be loaded with pretrained() of the companion object:

>>> gpt2 = GPT2Transformer.pretrained() \
...     .setInputCols(["document"]) \
...     .setOutputCol("generation")

The default model is "gpt2", if no name is provided. For available pretrained models please see the Models Hub.

Input Annotation types

Output Annotation type

DOCUMENT

DOCUMENT

Parameters
task

Transformer’s task, e.g. summarize: , by default “”

configProtoBytes

ConfigProto from tensorflow, serialized into byte array.

minOutputLength

Minimum length of the sequence to be generated, by default 0

maxOutputLength

Maximum length of output text, by default 20

doSample

Whether or not to use sampling; use greedy decoding otherwise, by default False

temperature

The value used to module the next token probabilities, by default 1.0

topK

The number of highest probability vocabulary tokens to keep for top-k-filtering, by default 50

topP

Top cumulative probability for vocabulary tokens, by default 1.0

If set to float < 1, only the most probable tokens with probabilities that add up to topP or higher are kept for generation.

repetitionPenalty

The parameter for repetition penalty, 1.0 means no penalty. , by default 1.0

noRepeatNgramSize

If set to int > 0, all ngrams of that size can only occur once, by default 0

ignoreTokenIds

A list of token ids which are ignored in the decoder’s output, by default []

Notes

This is a very computationally expensive module especially on larger sequence. The use of an accelerator such as GPU is recommended.

References

Paper Abstract:

Natural language processing tasks, such as question answering, machine translation, reading comprehension, and summarization, are typically approached with supervised learning on taskspecific datasets. We demonstrate that language models begin to learn these tasks without any explicit supervision when trained on a new dataset of millions of webpages called WebText. When conditioned on a document plus questions, the answers generated by the language model reach F1 on the CoQA dataset - matching or exceeding the performance of 3 out of 4 baseline systems without using the 127,000+ training examples. The capacity of the language model is essential to the success of zero-shot task transfer and increasing it improves performance in a log-linear fashion across tasks. Our largest model, GPT-2, is a 1.5B parameter Transformer that achieves state of the art results on 7 out of 8 tested language modeling datasets in a zero-shot setting but still underfits WebText. Samples from the model reflect these improvements and contain coherent paragraphs of text. These findings suggest a promising path towards building language processing systems which learn to perform tasks from their naturally occurring demonstrations.

Examples

>>> import sparknlp
>>> from sparknlp.base import *
>>> from sparknlp.annotator import *
>>> from pyspark.ml import Pipeline
>>> documentAssembler = DocumentAssembler() \
...     .setInputCol("text") \
...     .setOutputCol("documents")
>>> gpt2 = GPT2Transformer.pretrained("gpt2") \
...     .setInputCols(["documents"]) \
...     .setMaxOutputLength(50) \
...     .setOutputCol("generation")
>>> pipeline = Pipeline().setStages([documentAssembler, gpt2])
>>> data = spark.createDataFrame([["My name is Leonardo."]]).toDF("text")
>>> result = pipeline.fit(data).transform(data)
>>> result.select("summaries.generation").show(truncate=False)
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|result                                                                                                                                                                                              |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|[My name is Leonardo. I am a man of letters. I have been a man for many years. I was born in the year 1776. I came to the United States in 1776, and I have lived in the United Kingdom since 1776.]|
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+

Methods

__init__([classname, java_model])

Initialize this instance with a Java model object.

clear(param)

Clears a param from the param map if it has been explicitly set.

copy([extra])

Creates a copy of this instance with the same uid and some extra params.

explainParam(param)

Explains a single param and returns its name, doc, and optional default value and user-supplied value in a string.

explainParams()

Returns the documentation of all params with their optionally default values and user-supplied values.

extractParamMap([extra])

Extracts the embedded default param values and user-supplied values, and then merges them with extra values from input into a flat param map, where the latter value is used if there exist conflicts, i.e., with ordering: default param values < user-supplied values < extra.

getBatchSize()

Gets current batch size.

getInputCols()

Gets current column names of input annotations.

getLazyAnnotator()

Gets whether Annotator should be evaluated lazily in a RecursivePipeline.

getOrDefault(param)

Gets the value of a param in the user-supplied param map or its default value.

getOutputCol()

Gets output column name of annotations.

getParam(paramName)

Gets a param by its name.

getParamValue(paramName)

Gets the value of a parameter.

hasDefault(param)

Checks whether a param has a default value.

hasParam(paramName)

Tests whether this instance contains a param with a given (string) name.

isDefined(param)

Checks whether a param is explicitly set by user or has a default value.

isSet(param)

Checks whether a param is explicitly set by user.

load(path)

Reads an ML instance from the input path, a shortcut of read().load(path).

loadSavedModel(folder, spark_session)

Loads a locally saved model.

pretrained([name, lang, remote_loc])

Downloads and loads a pretrained model.

read()

Returns an MLReader instance for this class.

save(path)

Save this ML instance to the given path, a shortcut of 'write().save(path)'.

set(param, value)

Sets a parameter in the embedded param map.

setBatchSize(v)

Sets batch size.

setConfigProtoBytes(b)

Sets configProto from tensorflow, serialized into byte array.

setDoSample(value)

Sets whether or not to use sampling, use greedy decoding otherwise.

setIgnoreTokenIds(value)

A list of token ids which are ignored in the decoder's output.

setInputCols(*value)

Sets column names of input annotations.

setLazyAnnotator(value)

Sets whether Annotator should be evaluated lazily in a RecursivePipeline.

setMaxOutputLength(value)

Sets maximum length of output text.

setMinOutputLength(value)

Sets minimum length of the sequence to be generated.

setNoRepeatNgramSize(value)

Sets size of n-grams that can only occur once.

setOutputCol(value)

Sets output column name of annotations.

setParamValue(paramName)

Sets the value of a parameter.

setParams()

setRepetitionPenalty(value)

Sets the parameter for repetition penalty.

setTask(value)

Sets the transformer's task, e.g.

setTemperature(value)

Sets the value used to module the next token probabilities.

setTopK(value)

Sets the number of highest probability vocabulary tokens to keep for top-k-filtering.

setTopP(value)

Sets the top cumulative probability for vocabulary tokens.

transform(dataset[, params])

Transforms the input dataset with optional parameters.

write()

Returns an MLWriter instance for this ML instance.

Attributes

batchSize

configProtoBytes

doSample

getter_attrs

ignoreTokenIds

inputCols

lazyAnnotator

maxOutputLength

minOutputLength

name

noRepeatNgramSize

outputCol

params

Returns all params ordered by name.

repetitionPenalty

task

temperature

topK

topP

clear(param)#

Clears a param from the param map if it has been explicitly set.

copy(extra=None)#

Creates a copy of this instance with the same uid and some extra params. This implementation first calls Params.copy and then make a copy of the companion Java pipeline component with extra params. So both the Python wrapper and the Java pipeline component get copied.

Parameters

extra – Extra parameters to copy to the new instance

Returns

Copy of this instance

explainParam(param)#

Explains a single param and returns its name, doc, and optional default value and user-supplied value in a string.

explainParams()#

Returns the documentation of all params with their optionally default values and user-supplied values.

extractParamMap(extra=None)#

Extracts the embedded default param values and user-supplied values, and then merges them with extra values from input into a flat param map, where the latter value is used if there exist conflicts, i.e., with ordering: default param values < user-supplied values < extra.

Parameters

extra – extra param values

Returns

merged param map

getBatchSize()#

Gets current batch size.

Returns
int

Current batch size

getInputCols()#

Gets current column names of input annotations.

getLazyAnnotator()#

Gets whether Annotator should be evaluated lazily in a RecursivePipeline.

getOrDefault(param)#

Gets the value of a param in the user-supplied param map or its default value. Raises an error if neither is set.

getOutputCol()#

Gets output column name of annotations.

getParam(paramName)#

Gets a param by its name.

getParamValue(paramName)#

Gets the value of a parameter.

Parameters
paramNamestr

Name of the parameter

hasDefault(param)#

Checks whether a param has a default value.

hasParam(paramName)#

Tests whether this instance contains a param with a given (string) name.

isDefined(param)#

Checks whether a param is explicitly set by user or has a default value.

isSet(param)#

Checks whether a param is explicitly set by user.

classmethod load(path)#

Reads an ML instance from the input path, a shortcut of read().load(path).

static loadSavedModel(folder, spark_session)[source]#

Loads a locally saved model.

Parameters
folderstr

Folder of the saved model

spark_sessionpyspark.sql.SparkSession

The current SparkSession

Returns
GPT2Transformer

The restored model

property params#

Returns all params ordered by name. The default implementation uses dir() to get all attributes of type Param.

static pretrained(name='gpt2', lang='en', remote_loc=None)[source]#

Downloads and loads a pretrained model.

Parameters
namestr, optional

Name of the pretrained model, by default “gpt2”

langstr, optional

Language of the pretrained model, by default “en”

remote_locstr, optional

Optional remote address of the resource, by default None. Will use Spark NLPs repositories otherwise.

Returns
GPT2Transformer

The restored model

classmethod read()#

Returns an MLReader instance for this class.

save(path)#

Save this ML instance to the given path, a shortcut of ‘write().save(path)’.

set(param, value)#

Sets a parameter in the embedded param map.

setBatchSize(v)#

Sets batch size.

Parameters
vint

Batch size

setConfigProtoBytes(b)[source]#

Sets configProto from tensorflow, serialized into byte array.

Parameters
bList[int]

ConfigProto from tensorflow, serialized into byte array

setDoSample(value)[source]#

Sets whether or not to use sampling, use greedy decoding otherwise.

Parameters
valuebool

Whether or not to use sampling; use greedy decoding otherwise

setIgnoreTokenIds(value)[source]#

A list of token ids which are ignored in the decoder’s output.

Parameters
valueList[int]

The words to be filtered out

setInputCols(*value)#

Sets column names of input annotations.

Parameters
*valuestr

Input columns for the annotator

setLazyAnnotator(value)#

Sets whether Annotator should be evaluated lazily in a RecursivePipeline.

Parameters
valuebool

Whether Annotator should be evaluated lazily in a RecursivePipeline

setMaxOutputLength(value)[source]#

Sets maximum length of output text.

Parameters
valueint

Maximum length of output text

setMinOutputLength(value)[source]#

Sets minimum length of the sequence to be generated.

Parameters
valueint

Minimum length of the sequence to be generated

setNoRepeatNgramSize(value)[source]#

Sets size of n-grams that can only occur once.

If set to int > 0, all ngrams of that size can only occur once.

Parameters
valueint

N-gram size can only occur once

setOutputCol(value)#

Sets output column name of annotations.

Parameters
valuestr

Name of output column

setParamValue(paramName)#

Sets the value of a parameter.

Parameters
paramNamestr

Name of the parameter

setRepetitionPenalty(value)[source]#

Sets the parameter for repetition penalty. 1.0 means no penalty.

Parameters
valuefloat

The repetition penalty

References

See Ctrl: A Conditional Transformer Language Model For Controllable Generation for more details.

setTask(value)[source]#

Sets the transformer’s task, e.g. summarize:.

Parameters
valuestr

The transformer’s task

setTemperature(value)[source]#

Sets the value used to module the next token probabilities.

Parameters
valuefloat

The value used to module the next token probabilities

setTopK(value)[source]#

Sets the number of highest probability vocabulary tokens to keep for top-k-filtering.

Parameters
valueint

Number of highest probability vocabulary tokens to keep

setTopP(value)[source]#

Sets the top cumulative probability for vocabulary tokens.

If set to float < 1, only the most probable tokens with probabilities that add up to topP or higher are kept for generation.

Parameters
valuefloat

Cumulative probability for vocabulary tokens

transform(dataset, params=None)#

Transforms the input dataset with optional parameters.

Parameters
  • dataset – input dataset, which is an instance of pyspark.sql.DataFrame

  • params – an optional param map that overrides embedded params.

Returns

transformed dataset

New in version 1.3.0.

uid#

A unique id for the object.

write()#

Returns an MLWriter instance for this ML instance.