modeling

modeling#

class AutoBackbone(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoBackbone.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoBackbone. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoBackbone.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoBackbone

# Name of built-in pretrained model
model = AutoBackbone.from_pretrained("google/bit-50")
print(type(model))
# <class 'paddlenlp.transformers.bit.modeling.BitBackbone'>


# Load from local directory path
model = AutoBackbone.from_pretrained("./bit-50")
print(type(model))
# <class 'paddlenlp.transformers.bit.modeling.BitBackbone'>

class AutoModel(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoClass can help you automatically retrieve the relevant model given the provided pretrained weights/vocabulary. AutoModel is a generic model class that will be instantiated as one of the base model classes when created with the from_pretrained() classmethod.

classmethod from_pretrained(pretrained_model_name_or_path, task=None, *model_args, **kwargs)[source]#

Creates an instance of AutoModel. Model weights are loaded by specifying name of a built-in pretrained model, a pretrained model on HF, a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) –
Name of pretrained model or dir path to load from. The string can be:
- Name of a built-in pretrained model
- Name of a community-contributed pretrained model.
- Local directory path which contains model weights file(“model_state.pdparams”) and model config file (“model_config.json”).
task (str) – Specify a downstream task. Task can be ‘Model’, ‘ForPretraining’, ‘ForSequenceClassification’, ‘ForTokenClassification’, ‘ForQuestionAnswering’, ‘ForMultipleChoice’, ‘ForMaskedLM’, ‘ForCausalLM’, ‘Encoder’, ‘Decoder’, ‘Generator’, ‘Discriminator’, ‘ForConditionalGeneration’. We only support specify downstream tasks in AutoModel. Defaults to None.
*args (tuple) – Position arguments for model __init__. If provided, use these as position argument values for model initialization.
**kwargs (dict) – Keyword arguments for model __init__. If provided, use these to update pre-defined keyword argument values for model initialization. If the keyword is in __init__ argument names of base model, update argument values of the base model; else update argument values of derived model.

Returns:

An instance of AutoModel.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModel

# Name of built-in pretrained model
model = AutoModel.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModel'>

# Name of community-contributed pretrained model
model = AutoModel.from_pretrained('yingyibiao/bert-base-uncased-sst-2-finetuned')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModel'>

# Load from local directory path
model = AutoModel.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModel'>

# choose task
model = AutoModel.from_pretrained('bert-base-uncased', task='ForPretraining')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertForPretraining'>

class AutoModelForPretraining(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForPretraining.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForPretraining. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForPretraining.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForPretraining

# Name of built-in pretrained model
model = AutoModelForPretraining.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForPretraining'>

# Name of community-contributed pretrained model
model = AutoModelForPretraining.from_pretrained('iverxin/bert-base-japanese')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForPretraining'>

# Load from local directory path
model = AutoModelForPretraining.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForPretraining'>

class AutoModelForSequenceClassification(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForSequenceClassification.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForSequenceClassification. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForSequenceClassification.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForSequenceClassification

# Name of built-in pretrained model
model = AutoModelForSequenceClassification.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForSequenceClassification'>

# Name of community-contributed pretrained model
model = AutoModelForSequenceClassification.from_pretrained('iverxin/bert-base-japanese')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForSequenceClassification'>

# Load from local directory path
model = AutoModelForSequenceClassification.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForSequenceClassification'>

class AutoModelForTokenClassification(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForTokenClassification.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForTokenClassification. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForTokenClassification.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForTokenClassification

# Name of built-in pretrained model
model = AutoModelForTokenClassification.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForTokenClassification'>

# Name of community-contributed pretrained model
model = AutoModelForTokenClassification.from_pretrained('iverxin/bert-base-japanese')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForTokenClassification'>

# Load from local directory path
model = AutoModelForTokenClassification.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForTokenClassification'>

class AutoModelForQuestionAnswering(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForQuestionAnswering.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForQuestionAnswering. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForQuestionAnswering.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForQuestionAnswering

# Name of built-in pretrained model
model = AutoModelForQuestionAnswering.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForQuestionAnswering'>

# Name of community-contributed pretrained model
model = AutoModelForQuestionAnswering.from_pretrained('iverxin/bert-base-japanese')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForQuestionAnswering'>

# Load from local directory path
model = AutoModelForQuestionAnswering.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForQuestionAnswering'>

class AutoModelForMultipleChoice(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForMultipleChoice.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForMultipleChoice. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForMultipleChoice.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForMultipleChoice

# Name of built-in pretrained model
model = AutoModelForMultipleChoice.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForMultipleChoice'>

# Name of community-contributed pretrained model
model = AutoModelForMultipleChoice.from_pretrained('iverxin/bert-base-japanese')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForMultipleChoice'>

# Load from local directory path
model = AutoModelForMultipleChoice.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForMultipleChoice'>

class AutoModelForMaskedLM(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForMaskedLM.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForMaskedLM. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForMaskedLM.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForMaskedLM

# Name of built-in pretrained model
model = AutoModelForMaskedLM.from_pretrained('bert-base-uncased')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForMaskedLM'>

# Name of community-contributed pretrained model
model = AutoModelForMaskedLM.from_pretrained('iverxin/bert-base-japanese')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForMaskedLM'>

# Load from local directory path
model = AutoModelForMaskedLM.from_pretrained('./my_bert/')
print(type(model))
# <class 'paddlenlp.transformers.bert.modeling.BertModelForMaskedLM'>

class AutoModelForCausalLM(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForCausalLM.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForCausalLM. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForCausalLM.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForCausalLM

# Name of built-in pretrained model
model = AutoModelForCausalLM.from_pretrained('gpt2-en')
print(type(model))
# <class 'paddlenlp.transformers.gpt.modeling.GPTLMHeadModel'>

# Name of community-contributed pretrained model
model = AutoModelForCausalLM.from_pretrained('junnyu/distilgpt2')
print(type(model))
# <class 'paddlenlp.transformers.gpt.modeling.GPTLMHeadModel'>

# Load from local directory path
model = AutoModelForCausalLM.from_pretrained('./my_gpt/')
print(type(model))
# <class 'paddlenlp.transformers.gpt.modeling.GPTLMHeadModel'>

class AutoModelForCausalLMPipe(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

Pipeline model for AutoModelForCausalLM.

class AutoEncoder(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoEncoder.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoEncoder. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoEncoder.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoEncoder

# Name of built-in pretrained model
model = AutoEncoder.from_pretrained('bart-base',vocab_size=20000)
print(type(model))
# <class 'paddlenlp.transformers.bart.modeling.BartEncoder'>

# Load from local directory path
model = AutoEncoder.from_pretrained('./my_bart/')
print(type(model))
# <class 'paddlenlp.transformers.bart.modeling.BartEncoder'>

class AutoDecoder(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoDecoder.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoDecoder. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoDecoder.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoDecoder

# Name of built-in pretrained model
model = AutoDecoder.from_pretrained('bart-base', vocab_size=20000)
print(type(model))
# <class 'paddlenlp.transformers.bart.modeling.BartEncoder'>

# Load from local directory path
model = AutoDecoder.from_pretrained('./my_bart/')
print(type(model))
# <class 'paddlenlp.transformers.bart.modeling.BartEncoder'>

class AutoGenerator(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoGenerator.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoGenerator. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoGenerator.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoGenerator

# Name of built-in pretrained model
model = AutoGenerator.from_pretrained('electra-small')
print(type(model))
# <class 'paddlenlp.transformers.electra.modeling.ElectraGenerator'>

# Name of community-contributed pretrained model
model = AutoGenerator.from_pretrained('junnyu/hfl-chinese-legal-electra-small-generator')
print(type(model))
# <class 'paddlenlp.transformers.electra.modeling.ElectraGenerator'>

# Load from local directory path
model = AutoGenerator.from_pretrained('./my_electra/')
print(type(model))
# <class 'paddlenlp.transformers.electra.modeling.ElectraGenerator'>

class AutoDiscriminator(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoDiscriminator.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoDiscriminator. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoDiscriminator.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoDiscriminator

# Name of built-in pretrained model
model = AutoDiscriminator.from_pretrained('electra-small')
print(type(model))
# <class 'paddlenlp.transformers.electra.modeling.ElectraDiscriminator'>

# Name of community-contributed pretrained model
model = AutoDiscriminator.from_pretrained('junnyu/hfl-chinese-legal-electra-small-generator')
print(type(model))
# <class 'paddlenlp.transformers.electra.modeling.ElectraDiscriminator'>

# Load from local directory path
model = AutoDiscriminator.from_pretrained('./my_electra/')
print(type(model))
# <class 'paddlenlp.transformers.electra.modeling.ElectraDiscriminator'>

class AutoModelForConditionalGeneration(*args, **kwargs)[source]#

Bases: _BaseAutoModelClass

AutoModelForConditionalGeneration.

classmethod from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs)[source]#

Creates an instance of AutoModelForConditionalGeneration. Model weights are loaded by specifying name of a built-in pretrained model, or a community contributed model, or a local file directory path.

Parameters:

pretrained_model_name_or_path (str) – See AutoModel.
*args (tuple) – See AutoModel.
**kwargs (dict) – See AutoModel.

Returns:

An instance of AutoModelForConditionalGeneration.

Return type:

PretrainedModel

Example

from paddlenlp.transformers import AutoModelForConditionalGeneration

# Name of built-in pretrained model
model = AutoModelForConditionalGeneration.from_pretrained('bart-base')
print(type(model))
# <class 'paddlenlp.transformers.bart.modeling.BartForConditionalGeneration'>


# Load from local directory path
model = AutoModelForConditionalGeneration.from_pretrained('./my_bart/')
print(type(model))
# <class 'paddlenlp.transformers.bart.modeling.BartForConditionalGeneration'>

modeling

Contents

modeling#