RNAErnie
Pre-trained model on non-coding RNA (ncRNA) using a multi-stage masked language modeling (MLM) objective.
Statement
Multi-purpose RNA language modelling with motif-aware pretraining and type-guided fine-tuning is published in Nature Machine Intelligence, which is a Closed Access / Author-Fee journal.
Machine learning has been at the forefront of the movement for free and open access to research.
We see no role for closed access or author-fee publication in the future of machine learning research and believe the adoption of these journals as an outlet of record for the machine learning community would be a retrograde step.
The MultiMolecule team is committed to the principles of open access and open science.
We do NOT endorse the publication of manuscripts in Closed Access / Author-Fee journals and encourage the community to support Open Access journals and conferences.
Please consider signing the Statement on Nature Machine Intelligence.
Disclaimer
This is an UNOFFICIAL implementation of the RNAErnie: An RNA Language Model with Structure-enhanced Representations by Ning Wang, Jiang Bian,
Haoyi Xiong, et al.
The OFFICIAL repository of RNAErnie is at CatIIIIIIII/RNAErnie.
Warning
The MultiMolecule team is unable to confirm that the provided model and checkpoints are producing the same intermediate representations as the original implementation.
This is because
The proposed method is published in a Closed Access / Author-Fee journal.
The team releasing RNAErnie did not write this model card for this model so this model card has been written by the MultiMolecule team.
Model Details
RNAErnie is a bert-style model pre-trained on a large corpus of non-coding RNA sequences in a self-supervised fashion. This means that the model was trained on the raw nucleotides of RNA sequences only, with an automatic process to generate inputs and labels from those texts. Please refer to the Training Details section for more information on the training process.
Note that during the conversion process, additional tokens such as [IND] and ncRNA class symbols are removed.
Model Specification
| Num Layers |
Hidden Size |
Num Heads |
Intermediate Size |
Num Parameters (M) |
FLOPs (G) |
MACs (G) |
Max Num Tokens |
| 12 |
768 |
12 |
3072 |
86.06 |
22.36 |
11.17 |
512 |
Links
Usage
The model file depends on the multimolecule library. You can install it using pip:
| Bash |
|---|
| pip install multimolecule
|
Direct Use
You can use this model directly with a pipeline for masked language modeling:
| Python |
|---|
| >>> import multimolecule # you must import multimolecule to register models
>>> from transformers import pipeline
>>> unmasker = pipeline("fill-mask", model="multimolecule/rnaernie")
>>> unmasker("gguc<mask>cucugguuagaccagaucugagccu")
[{'score': 0.09252794831991196,
'token': 8,
'token_str': 'G',
'sequence': 'G G U C G C U C U G G U U A G A C C A G A U C U G A G C C U'},
{'score': 0.09062391519546509,
'token': 11,
'token_str': 'R',
'sequence': 'G G U C R C U C U G G U U A G A C C A G A U C U G A G C C U'},
{'score': 0.08875908702611923,
'token': 6,
'token_str': 'A',
'sequence': 'G G U C A C U C U G G U U A G A C C A G A U C U G A G C C U'},
{'score': 0.07809742540121078,
'token': 20,
'token_str': 'V',
'sequence': 'G G U C V C U C U G G U U A G A C C A G A U C U G A G C C U'},
{'score': 0.07325706630945206,
'token': 13,
'token_str': 'S',
'sequence': 'G G U C S C U C U G G U U A G A C C A G A U C U G A G C C U'}]
|
Downstream Use
Here is how to use this model to get the features of a given sequence in PyTorch:
| Python |
|---|
| from multimolecule import RnaTokenizer, RnaErnieModel
tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnaernie")
model = RnaErnieModel.from_pretrained("multimolecule/rnaernie")
text = "UAGCUUAUCAGACUGAUGUUGA"
input = tokenizer(text, return_tensors="pt")
output = model(**input)
|
Sequence Classification / Regression
Note: This model is not fine-tuned for any specific task. You will need to fine-tune the model on a downstream task to use it for sequence classification or regression.
Here is how to use this model as backbone to fine-tune for a sequence-level task in PyTorch:
| Python |
|---|
| import torch
from multimolecule import RnaTokenizer, RnaErnieForSequencePrediction
tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnaernie")
model = RnaErnieForSequencePrediction.from_pretrained("multimolecule/rnaernie")
text = "UAGCUUAUCAGACUGAUGUUGA"
input = tokenizer(text, return_tensors="pt")
label = torch.tensor([1])
output = model(**input, labels=label)
|
Token Classification / Regression
Note: This model is not fine-tuned for any specific task. You will need to fine-tune the model on a downstream task to use it for nucleotide classification or regression.
Here is how to use this model as backbone to fine-tune for a nucleotide-level task in PyTorch:
| Python |
|---|
| import torch
from multimolecule import RnaTokenizer, RnaErnieForTokenPrediction
tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnaernie")
model = RnaErnieForTokenPrediction.from_pretrained("multimolecule/rnaernie")
text = "UAGCUUAUCAGACUGAUGUUGA"
input = tokenizer(text, return_tensors="pt")
label = torch.randint(2, (len(text), ))
output = model(**input, labels=label)
|
Note: This model is not fine-tuned for any specific task. You will need to fine-tune the model on a downstream task to use it for contact classification or regression.
Here is how to use this model as backbone to fine-tune for a contact-level task in PyTorch:
| Python |
|---|
| import torch
from multimolecule import RnaTokenizer, RnaErnieForContactPrediction
tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnaernie")
model = RnaErnieForContactPrediction.from_pretrained("multimolecule/rnaernie")
text = "UAGCUUAUCAGACUGAUGUUGA"
input = tokenizer(text, return_tensors="pt")
label = torch.randint(2, (len(text), len(text)))
output = model(**input, labels=label)
|
Training Details
RNAErnie used Masked Language Modeling (MLM) as the pre-training objective: taking a sequence, the model randomly masks 15% of the tokens in the input then runs the entire masked sentence through the model and has to predict the masked tokens. This is comparable to the Cloze task in language modeling.
Training Data
The RNAErnie model was pre-trained on RNAcentral.
RNAcentral is a free, public resource that offers integrated access to a comprehensive and up-to-date set of non-coding RNA sequences provided by a collaborating group of Expert Databases representing a broad range of organisms and RNA types.
RNAErnie used a subset of RNAcentral for pre-training. The subset contains 23 million sequences.
RNAErnie preprocessed all tokens by replacing “T”s with “S”s.
Note that RnaTokenizer will convert “T”s to “U”s for you, you may disable this behaviour by passing replace_T_with_U=False.
Training Procedure
Preprocessing
RNAErnie used masked language modeling (MLM) as the pre-training objective. The masking procedure is similar to the one used in BERT:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by
<mask>.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
PreTraining
RNAErnie uses a special 3-stage training pipeline to pre-train the model, each with a different masking strategy:
Base-level Masking: The masking applies to each nucleotide in the sequence.
Subsequence-level Masking: The masking applies to subsequences of 4-8bp in the sequence.
Motif-level Masking: The model is trained on motif datasets.
The model was trained on 4 NVIDIA V100 GPUs with 32GiB memories.
- Batch size: 50
- Learning rate: 1e-4
- Weight decay: 0.01
- Optimizer: AdamW
- Steps: 2,580,000
- Learning rate warm-up: 129,000 steps
- Learning rate cool-down: 129,000 steps
- Minimum learning rate: 5e-5
Citation
Citation information is not available for papers published in Closed Access / Author-Fee journals.
Please use GitHub issues of MultiMolecule for any questions or comments on the model card.
Please contact the authors of the RNAErnie paper for questions or comments on the paper/model.
License
This model is licensed under the AGPL-3.0 License.
| Text Only |
|---|
| SPDX-License-Identifier: AGPL-3.0-or-later
|
Bases: Tokenizer
Tokenizer for RNA sequences.
Parameters:
| Name |
Type |
Description |
Default |
alphabet
|
Alphabet | str | List[str] | None
|
alphabet to use for tokenization.
- If is
None, the standard RNA alphabet will be used.
- If is a
string, it should correspond to the name of a predefined alphabet. The options include
standardextendedstreamlinenucleobase
- If is an alphabet or a list of characters, that specific alphabet will be used.
|
None
|
nmers
|
int
|
Size of kmer to tokenize.
|
1
|
codon
|
bool
|
Whether to tokenize into codons.
|
False
|
replace_T_with_U
|
bool
|
Whether to replace T with U.
|
True
|
do_upper_case
|
bool
|
Whether to convert input to uppercase.
|
True
|
Examples:
Python Console Session>>> from multimolecule import RnaTokenizer
>>> tokenizer = RnaTokenizer()
>>> tokenizer('<pad><cls><eos><unk><mask><null>ACGUNRYSWKMBDHV.X*-I')["input_ids"]
[1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 2]
>>> tokenizer('acgu')["input_ids"]
[1, 6, 7, 8, 9, 2]
>>> tokenizer('acgt')["input_ids"]
[1, 6, 7, 8, 9, 2]
>>> tokenizer = RnaTokenizer(replace_T_with_U=False)
>>> tokenizer('acgt')["input_ids"]
[1, 6, 7, 8, 3, 2]
>>> tokenizer = RnaTokenizer(nmers=3)
>>> tokenizer('uagcuuauc')["input_ids"]
[1, 83, 17, 64, 49, 96, 84, 22, 2]
>>> tokenizer = RnaTokenizer(codon=True)
>>> tokenizer('uagcuuauc')["input_ids"]
[1, 83, 49, 22, 2]
>>> tokenizer('uagcuuauca')["input_ids"]
Traceback (most recent call last):
ValueError: length of input sequence must be a multiple of 3 for codon tokenization, but got 10
Source code in multimolecule/tokenisers/rna/tokenization_rna.py
| Python |
|---|
| class RnaTokenizer(Tokenizer):
"""
Tokenizer for RNA sequences.
Args:
alphabet: alphabet to use for tokenization.
- If is `None`, the standard RNA alphabet will be used.
- If is a `string`, it should correspond to the name of a predefined alphabet. The options include
+ `standard`
+ `extended`
+ `streamline`
+ `nucleobase`
- If is an alphabet or a list of characters, that specific alphabet will be used.
nmers: Size of kmer to tokenize.
codon: Whether to tokenize into codons.
replace_T_with_U: Whether to replace T with U.
do_upper_case: Whether to convert input to uppercase.
Examples:
>>> from multimolecule import RnaTokenizer
>>> tokenizer = RnaTokenizer()
>>> tokenizer('<pad><cls><eos><unk><mask><null>ACGUNRYSWKMBDHV.X*-I')["input_ids"]
[1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 2]
>>> tokenizer('acgu')["input_ids"]
[1, 6, 7, 8, 9, 2]
>>> tokenizer('acgt')["input_ids"]
[1, 6, 7, 8, 9, 2]
>>> tokenizer = RnaTokenizer(replace_T_with_U=False)
>>> tokenizer('acgt')["input_ids"]
[1, 6, 7, 8, 3, 2]
>>> tokenizer = RnaTokenizer(nmers=3)
>>> tokenizer('uagcuuauc')["input_ids"]
[1, 83, 17, 64, 49, 96, 84, 22, 2]
>>> tokenizer = RnaTokenizer(codon=True)
>>> tokenizer('uagcuuauc')["input_ids"]
[1, 83, 49, 22, 2]
>>> tokenizer('uagcuuauca')["input_ids"]
Traceback (most recent call last):
ValueError: length of input sequence must be a multiple of 3 for codon tokenization, but got 10
"""
model_input_names = ["input_ids", "attention_mask"]
def __init__(
self,
alphabet: Alphabet | str | List[str] | None = None,
nmers: int = 1,
codon: bool = False,
replace_T_with_U: bool = True,
do_upper_case: bool = True,
additional_special_tokens: List | Tuple | None = None,
**kwargs,
):
if codon and (nmers > 1 and nmers != 3):
raise ValueError("Codon and nmers cannot be used together.")
if codon:
nmers = 3 # set to 3 to get correct vocab
if not isinstance(alphabet, Alphabet):
alphabet = get_alphabet(alphabet, nmers=nmers)
super().__init__(
alphabet=alphabet,
nmers=nmers,
codon=codon,
replace_T_with_U=replace_T_with_U,
do_upper_case=do_upper_case,
additional_special_tokens=additional_special_tokens,
**kwargs,
)
self.replace_T_with_U = replace_T_with_U
self.nmers = nmers
self.condon = codon
def _tokenize(self, text: str, **kwargs):
if self.do_upper_case:
text = text.upper()
if self.replace_T_with_U:
text = text.replace("T", "U")
if self.condon:
if len(text) % 3 != 0:
raise ValueError(
f"length of input sequence must be a multiple of 3 for codon tokenization, but got {len(text)}"
)
return [text[i : i + 3] for i in range(0, len(text), 3)]
if self.nmers > 1:
return [text[i : i + self.nmers] for i in range(len(text) - self.nmers + 1)] # noqa: E203
return list(text)
|
Bases: PreTrainedConfig
This is the configuration class to store the configuration of a
RnaErnieModel. It is used to instantiate a RnaErnie model according to the
specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a
similar configuration to that of the RnaErnie Bruce-ywj/rnaernie
architecture.
Configuration objects inherit from PreTrainedConfig and can be used to
control the model outputs. Read the documentation from PreTrainedConfig
for more information.
Parameters:
| Name |
Type |
Description |
Default |
vocab_size
|
int
|
Vocabulary size of the RnaErnie model. Defines the number of different tokens that can be represented by
the inputs_ids passed when calling [RnaErnieModel].
|
26
|
hidden_size
|
int
|
Dimensionality of the encoder layers and the pooler layer.
|
768
|
num_hidden_layers
|
int
|
Number of hidden layers in the Transformer encoder.
|
12
|
num_attention_heads
|
int
|
Number of attention heads for each attention layer in the Transformer encoder.
|
12
|
|
|
int
|
Dimensionality of the “intermediate” (often named feed-forward) layer in the Transformer encoder.
|
3072
|
hidden_dropout
|
float
|
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
|
0.1
|
attention_dropout
|
float
|
The dropout ratio for the attention probabilities.
|
0.1
|
max_position_embeddings
|
int
|
The maximum sequence length that this model might ever be used with. Typically set this to something large
just in case (e.g., 512 or 1024 or 2048).
|
513
|
initializer_range
|
float
|
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
|
0.02
|
layer_norm_eps
|
float
|
The epsilon used by the layer normalization layers.
|
1e-12
|
Examples:
Python Console Session>>> from multimolecule import RnaErnieModel, RnaErnieConfig
>>> # Initializing a rnaernie multimolecule/rnaernie style configuration
>>> configuration = RnaErnieConfig()
>>> # Initializing a model (with random weights) from the multimolecule/rnaernie style configuration
>>> model = RnaErnieModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
Source code in multimolecule/models/rnaernie/configuration_rnaernie.py
| Python |
|---|
| class RnaErnieConfig(PreTrainedConfig):
r"""
This is the configuration class to store the configuration of a
[`RnaErnieModel`][multimolecule.models.RnaErnieModel]. It is used to instantiate a RnaErnie model according to the
specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a
similar configuration to that of the RnaErnie [Bruce-ywj/rnaernie](https://github.com/Bruce-ywj/rnaernie)
architecture.
Configuration objects inherit from [`PreTrainedConfig`][multimolecule.models.PreTrainedConfig] and can be used to
control the model outputs. Read the documentation from [`PreTrainedConfig`][multimolecule.models.PreTrainedConfig]
for more information.
Args:
vocab_size:
Vocabulary size of the RnaErnie model. Defines the number of different tokens that can be represented by
the `inputs_ids` passed when calling [`RnaErnieModel`].
hidden_size:
Dimensionality of the encoder layers and the pooler layer.
num_hidden_layers:
Number of hidden layers in the Transformer encoder.
num_attention_heads:
Number of attention heads for each attention layer in the Transformer encoder.
intermediate_size:
Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.
hidden_dropout:
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
attention_dropout:
The dropout ratio for the attention probabilities.
max_position_embeddings:
The maximum sequence length that this model might ever be used with. Typically set this to something large
just in case (e.g., 512 or 1024 or 2048).
initializer_range:
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
layer_norm_eps:
The epsilon used by the layer normalization layers.
Examples:
>>> from multimolecule import RnaErnieModel, RnaErnieConfig
>>> # Initializing a rnaernie multimolecule/rnaernie style configuration
>>> configuration = RnaErnieConfig()
>>> # Initializing a model (with random weights) from the multimolecule/rnaernie style configuration
>>> model = RnaErnieModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
"""
model_type = "rnaernie"
def __init__(
self,
vocab_size: int = 26,
hidden_size: int = 768,
num_hidden_layers: int = 12,
num_attention_heads: int = 12,
intermediate_size: int = 3072,
hidden_act: str = "relu",
hidden_dropout: float = 0.1,
attention_dropout: float = 0.1,
max_position_embeddings: int = 513,
initializer_range: float = 0.02,
layer_norm_eps: float = 1e-12,
position_embedding_type: str = "absolute",
is_decoder: bool = False,
use_cache: bool = True,
head: HeadConfig | None = None,
lm_head: MaskedLMHeadConfig | None = None,
**kwargs,
):
super().__init__(**kwargs)
self.vocab_size = vocab_size
self.type_vocab_size = 2
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.hidden_act = hidden_act
self.hidden_dropout = hidden_dropout
self.attention_dropout = attention_dropout
self.max_position_embeddings = max_position_embeddings
self.initializer_range = initializer_range
self.layer_norm_eps = layer_norm_eps
self.position_embedding_type = position_embedding_type
self.is_decoder = is_decoder
self.use_cache = use_cache
self.head = HeadConfig(**head if head is not None else {})
self.lm_head = MaskedLMHeadConfig(**lm_head if lm_head is not None else {})
|
Bases: RnaErniePreTrainedModel
Examples:
Python Console Session>>> from multimolecule import RnaErnieConfig, RnaErnieForContactPrediction, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForContactPrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=torch.randint(2, (1, 5, 5)))
>>> output["logits"].shape
torch.Size([1, 5, 5, 2])
>>> output["loss"]
tensor(..., grad_fn=<NllLossBackward0>)
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| class RnaErnieForContactPrediction(RnaErniePreTrainedModel):
"""
Examples:
>>> from multimolecule import RnaErnieConfig, RnaErnieForContactPrediction, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForContactPrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=torch.randint(2, (1, 5, 5)))
>>> output["logits"].shape
torch.Size([1, 5, 5, 2])
>>> output["loss"] # doctest:+ELLIPSIS
tensor(..., grad_fn=<NllLossBackward0>)
"""
def __init__(self, config: RnaErnieConfig):
super().__init__(config)
self.rnaernie = RnaErnieModel(config, add_pooling_layer=True)
self.contact_head = ContactPredictionHead(config)
self.head_config = self.contact_head.config
# Initialize weights and apply final processing
self.post_init()
def forward(
self,
input_ids: Tensor | NestedTensor,
attention_mask: Tensor | None = None,
position_ids: Tensor | None = None,
head_mask: Tensor | None = None,
inputs_embeds: Tensor | NestedTensor | None = None,
labels: Tensor | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> Tuple[Tensor, ...] | ContactPredictorOutput:
if output_attentions is False:
warn("output_attentions must be True for contact classification and will be ignored.")
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.rnaernie(
input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=True,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
**kwargs,
)
output = self.contact_head(outputs, attention_mask, input_ids, labels)
logits, loss = output.logits, output.loss
if not return_dict:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return ContactPredictorOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
|
Bases: RnaErniePreTrainedModel
Examples:
Python Console Session>>> from multimolecule import RnaErnieConfig, RnaErnieForMaskedLM, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForMaskedLM(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=input["input_ids"])
>>> output["logits"].shape
torch.Size([1, 7, 26])
>>> output["loss"]
tensor(..., grad_fn=<NllLossBackward0>)
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| class RnaErnieForMaskedLM(RnaErniePreTrainedModel):
"""
Examples:
>>> from multimolecule import RnaErnieConfig, RnaErnieForMaskedLM, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForMaskedLM(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=input["input_ids"])
>>> output["logits"].shape
torch.Size([1, 7, 26])
>>> output["loss"] # doctest:+ELLIPSIS
tensor(..., grad_fn=<NllLossBackward0>)
"""
_tied_weights_keys = ["lm_head.decoder.bias", "lm_head.decoder.weight"]
def __init__(self, config: RnaErnieConfig):
super().__init__(config)
if config.is_decoder:
logger.warning(
"If you want to use `RnaErnieForMaskedLM` make sure `config.is_decoder=False` for "
"bi-directional self-attention."
)
self.rnaernie = RnaErnieModel(config, add_pooling_layer=False)
self.lm_head = MaskedLMHead(config)
# Initialize weights and apply final processing
self.post_init()
def get_output_embeddings(self):
return self.lm_head.decoder
def set_output_embeddings(self, new_embeddings):
self.lm_head.decoder = new_embeddings
def forward(
self,
input_ids: Tensor | NestedTensor,
attention_mask: Tensor | None = None,
position_ids: Tensor | None = None,
head_mask: Tensor | None = None,
inputs_embeds: Tensor | NestedTensor | None = None,
encoder_hidden_states: Tensor | None = None,
encoder_attention_mask: Tensor | None = None,
labels: Tensor | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> Tuple[Tensor, ...] | MaskedLMOutput:
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.rnaernie(
input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
**kwargs,
)
output = self.lm_head(outputs, labels)
logits, loss = output.logits, output.loss
if not return_dict:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return MaskedLMOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
|
Bases: RnaErniePreTrainedModel
Examples:
Python Console Session>>> from multimolecule import RnaErnieConfig, RnaErnieForSequencePrediction, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForSequencePrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=torch.tensor([[1]]))
>>> output["logits"].shape
torch.Size([1, 2])
>>> output["loss"]
tensor(..., grad_fn=<NllLossBackward0>)
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| class RnaErnieForSequencePrediction(RnaErniePreTrainedModel):
"""
Examples:
>>> from multimolecule import RnaErnieConfig, RnaErnieForSequencePrediction, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForSequencePrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=torch.tensor([[1]]))
>>> output["logits"].shape
torch.Size([1, 2])
>>> output["loss"] # doctest:+ELLIPSIS
tensor(..., grad_fn=<NllLossBackward0>)
"""
def __init__(self, config):
super().__init__(config)
self.rnaernie = RnaErnieModel(config)
self.sequence_head = SequencePredictionHead(config)
self.head_config = self.sequence_head.config
# Initialize weights and apply final processing
self.post_init()
def forward(
self,
input_ids: Tensor | NestedTensor,
attention_mask: Tensor | None = None,
position_ids: Tensor | None = None,
head_mask: Tensor | None = None,
inputs_embeds: Tensor | NestedTensor | None = None,
labels: Tensor | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> Tuple[Tensor, ...] | SequencePredictorOutput:
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.rnaernie(
input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
**kwargs,
)
output = self.sequence_head(outputs, labels)
logits, loss = output.logits, output.loss
if not return_dict:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return SequencePredictorOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
|
Bases: RnaErniePreTrainedModel
Examples:
Python Console Session>>> from multimolecule import RnaErnieConfig, RnaErnieForTokenPrediction, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForTokenPrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=torch.randint(2, (1, 5)))
>>> output["logits"].shape
torch.Size([1, 5, 2])
>>> output["loss"]
tensor(..., grad_fn=<NllLossBackward0>)
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| class RnaErnieForTokenPrediction(RnaErniePreTrainedModel):
"""
Examples:
>>> from multimolecule import RnaErnieConfig, RnaErnieForTokenPrediction, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieForTokenPrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels=torch.randint(2, (1, 5)))
>>> output["logits"].shape
torch.Size([1, 5, 2])
>>> output["loss"] # doctest:+ELLIPSIS
tensor(..., grad_fn=<NllLossBackward0>)
"""
def __init__(self, config: RnaErnieConfig):
super().__init__(config)
self.rnaernie = RnaErnieModel(config, add_pooling_layer=True)
self.token_head = TokenPredictionHead(config)
self.head_config = self.token_head.config
# Initialize weights and apply final processing
self.post_init()
def forward(
self,
input_ids: Tensor | NestedTensor,
attention_mask: Tensor | None = None,
position_ids: Tensor | None = None,
head_mask: Tensor | None = None,
inputs_embeds: Tensor | NestedTensor | None = None,
labels: Tensor | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> Tuple[Tensor, ...] | TokenPredictorOutput:
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.rnaernie(
input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
**kwargs,
)
output = self.token_head(outputs, attention_mask, input_ids, labels)
logits, loss = output.logits, output.loss
if not return_dict:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return TokenPredictorOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
|
Bases: RnaErniePreTrainedModel
Examples:
Python Console Session>>> from multimolecule import RnaErnieConfig, RnaErnieModel, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieModel(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input)
>>> output["last_hidden_state"].shape
torch.Size([1, 7, 768])
>>> output["pooler_output"].shape
torch.Size([1, 768])
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| class RnaErnieModel(RnaErniePreTrainedModel):
"""
Examples:
>>> from multimolecule import RnaErnieConfig, RnaErnieModel, RnaTokenizer
>>> config = RnaErnieConfig()
>>> model = RnaErnieModel(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input)
>>> output["last_hidden_state"].shape
torch.Size([1, 7, 768])
>>> output["pooler_output"].shape
torch.Size([1, 768])
"""
def __init__(self, config: RnaErnieConfig, add_pooling_layer: bool = True):
super().__init__(config)
self.pad_token_id = config.pad_token_id
self.embeddings = RnaErnieEmbeddings(config)
self.encoder = RnaErnieEncoder(config)
self.pooler = RnaErniePooler(config) if add_pooling_layer else None
# Initialize weights and apply final processing
self.post_init()
def get_input_embeddings(self):
return self.embeddings.word_embeddings
def set_input_embeddings(self, value):
self.embeddings.word_embeddings = value
def _prune_heads(self, heads_to_prune):
"""
Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
class PreTrainedModel
"""
for layer, heads in heads_to_prune.items():
self.encoder.layer[layer].attention.prune_heads(heads)
def forward(
self,
input_ids: Tensor | NestedTensor,
attention_mask: Tensor | None = None,
position_ids: Tensor | None = None,
head_mask: Tensor | None = None,
inputs_embeds: Tensor | NestedTensor | None = None,
encoder_hidden_states: Tensor | None = None,
encoder_attention_mask: Tensor | None = None,
past_key_values: Tuple[Tuple[Tensor, Tensor, Tensor, Tensor], ...] | None = None,
use_cache: bool | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> Tuple[Tensor, ...] | BaseModelOutputWithPoolingAndCrossAttentions:
r"""
Args:
encoder_hidden_states:
Shape: `(batch_size, sequence_length, hidden_size)`
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
the model is configured as a decoder.
encoder_attention_mask:
Shape: `(batch_size, sequence_length)`
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used
in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
- 1 for tokens that are **not masked**,
- 0 for tokens that are **masked**.
past_key_values:
Tuple of length `config.n_layers` with each tuple having 4 tensors of shape
`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up
decoding.
If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
use_cache:
If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
(see `past_key_values`).
"""
if kwargs:
warn(
f"Additional keyword arguments `{', '.join(kwargs)}` are detected in "
f"`{self.__class__.__name__}.forward`, they will be ignored.\n"
"This is provided for backward compatibility and may lead to unexpected behavior."
)
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
if self.config.is_decoder:
use_cache = use_cache if use_cache is not None else self.config.use_cache
else:
use_cache = False
if isinstance(input_ids, NestedTensor):
input_ids, attention_mask = input_ids.tensor, input_ids.mask
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
if input_ids is not None:
self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
input_shape = input_ids.size()
elif inputs_embeds is not None:
input_shape = inputs_embeds.size()[:-1]
else:
raise ValueError("You have to specify either input_ids or inputs_embeds")
batch_size, seq_length = input_shape
device = input_ids.device if input_ids is not None else inputs_embeds.device # type: ignore[union-attr]
# past_key_values_length
past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
if attention_mask is None:
attention_mask = (
input_ids.ne(self.pad_token_id)
if self.pad_token_id is not None
else torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
)
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
# ourselves in which case we just need to make it broadcastable to all heads.
extended_attention_mask: Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
# If a 2D or 3D attention mask is provided for the cross-attention
# we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
if self.config.is_decoder and encoder_hidden_states is not None:
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
if encoder_attention_mask is None:
encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
else:
encoder_extended_attention_mask = None
# Prepare head mask if needed
# 1.0 in head_mask indicate we keep the head
# attention_probs has shape bsz x n_heads x N x N
# input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
# and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
embedding_output = self.embeddings(
input_ids=input_ids,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
past_key_values_length=past_key_values_length,
)
encoder_outputs = self.encoder(
embedding_output,
attention_mask=extended_attention_mask,
head_mask=head_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_extended_attention_mask,
past_key_values=past_key_values,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
sequence_output = encoder_outputs[0]
pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
if not return_dict:
return (sequence_output, pooled_output) + encoder_outputs[1:]
return BaseModelOutputWithPoolingAndCrossAttentions(
last_hidden_state=sequence_output,
pooler_output=pooled_output,
past_key_values=encoder_outputs.past_key_values,
hidden_states=encoder_outputs.hidden_states,
attentions=encoder_outputs.attentions,
cross_attentions=encoder_outputs.cross_attentions,
)
|
| Python |
|---|
| forward(input_ids: Tensor | NestedTensor, attention_mask: Tensor | None = None, position_ids: Tensor | None = None, head_mask: Tensor | None = None, inputs_embeds: Tensor | NestedTensor | None = None, encoder_hidden_states: Tensor | None = None, encoder_attention_mask: Tensor | None = None, past_key_values: Tuple[Tuple[Tensor, Tensor, Tensor, Tensor], ...] | None = None, use_cache: bool | None = None, output_attentions: bool | None = None, output_hidden_states: bool | None = None, return_dict: bool | None = None, **kwargs) -> Tuple[Tensor, ...] | BaseModelOutputWithPoolingAndCrossAttentions
|
Parameters:
| Name |
Type |
Description |
Default |
encoder_hidden_states
|
Tensor | None
|
Shape: (batch_size, sequence_length, hidden_size)
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
the model is configured as a decoder.
|
None
|
encoder_attention_mask
|
Tensor | None
|
Shape: (batch_size, sequence_length)
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used
in the cross-attention if the model is configured as a decoder. Mask values selected in [0, 1]:
- 1 for tokens that are not masked,
- 0 for tokens that are masked.
|
None
|
past_key_values
|
Tuple[Tuple[Tensor, Tensor, Tensor, Tensor], ...] | None
|
Tuple of length config.n_layers with each tuple having 4 tensors of shape
`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up
decoding.
If past_key_values are used, the user can optionally input only the last decoder_input_ids (those
that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of
all decoder_input_ids of shape (batch_size, sequence_length).
|
None
|
use_cache
|
bool | None
|
If set to True, past_key_values key value states are returned and can be used to speed up decoding
(see past_key_values).
|
None
|
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| def forward(
self,
input_ids: Tensor | NestedTensor,
attention_mask: Tensor | None = None,
position_ids: Tensor | None = None,
head_mask: Tensor | None = None,
inputs_embeds: Tensor | NestedTensor | None = None,
encoder_hidden_states: Tensor | None = None,
encoder_attention_mask: Tensor | None = None,
past_key_values: Tuple[Tuple[Tensor, Tensor, Tensor, Tensor], ...] | None = None,
use_cache: bool | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
return_dict: bool | None = None,
**kwargs,
) -> Tuple[Tensor, ...] | BaseModelOutputWithPoolingAndCrossAttentions:
r"""
Args:
encoder_hidden_states:
Shape: `(batch_size, sequence_length, hidden_size)`
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
the model is configured as a decoder.
encoder_attention_mask:
Shape: `(batch_size, sequence_length)`
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used
in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
- 1 for tokens that are **not masked**,
- 0 for tokens that are **masked**.
past_key_values:
Tuple of length `config.n_layers` with each tuple having 4 tensors of shape
`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up
decoding.
If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
use_cache:
If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
(see `past_key_values`).
"""
if kwargs:
warn(
f"Additional keyword arguments `{', '.join(kwargs)}` are detected in "
f"`{self.__class__.__name__}.forward`, they will be ignored.\n"
"This is provided for backward compatibility and may lead to unexpected behavior."
)
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
if self.config.is_decoder:
use_cache = use_cache if use_cache is not None else self.config.use_cache
else:
use_cache = False
if isinstance(input_ids, NestedTensor):
input_ids, attention_mask = input_ids.tensor, input_ids.mask
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
if input_ids is not None:
self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
input_shape = input_ids.size()
elif inputs_embeds is not None:
input_shape = inputs_embeds.size()[:-1]
else:
raise ValueError("You have to specify either input_ids or inputs_embeds")
batch_size, seq_length = input_shape
device = input_ids.device if input_ids is not None else inputs_embeds.device # type: ignore[union-attr]
# past_key_values_length
past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
if attention_mask is None:
attention_mask = (
input_ids.ne(self.pad_token_id)
if self.pad_token_id is not None
else torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
)
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
# ourselves in which case we just need to make it broadcastable to all heads.
extended_attention_mask: Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
# If a 2D or 3D attention mask is provided for the cross-attention
# we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
if self.config.is_decoder and encoder_hidden_states is not None:
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
if encoder_attention_mask is None:
encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
else:
encoder_extended_attention_mask = None
# Prepare head mask if needed
# 1.0 in head_mask indicate we keep the head
# attention_probs has shape bsz x n_heads x N x N
# input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
# and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
embedding_output = self.embeddings(
input_ids=input_ids,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
past_key_values_length=past_key_values_length,
)
encoder_outputs = self.encoder(
embedding_output,
attention_mask=extended_attention_mask,
head_mask=head_mask,
encoder_hidden_states=encoder_hidden_states,
encoder_attention_mask=encoder_extended_attention_mask,
past_key_values=past_key_values,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
sequence_output = encoder_outputs[0]
pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
if not return_dict:
return (sequence_output, pooled_output) + encoder_outputs[1:]
return BaseModelOutputWithPoolingAndCrossAttentions(
last_hidden_state=sequence_output,
pooler_output=pooled_output,
past_key_values=encoder_outputs.past_key_values,
hidden_states=encoder_outputs.hidden_states,
attentions=encoder_outputs.attentions,
cross_attentions=encoder_outputs.cross_attentions,
)
|
Bases: PreTrainedModel
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
models.
Source code in multimolecule/models/rnaernie/modeling_rnaernie.py
| Python |
|---|
| class RnaErniePreTrainedModel(PreTrainedModel):
"""
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
models.
"""
config_class = RnaErnieConfig
base_model_prefix = "rnaernie"
supports_gradient_checkpointing = True
_no_split_modules = ["RnaErnieLayer", "RnaErnieEmbeddings"]
# Copied from transformers.models.bert.modeling_bert.BertPreTrainedModel._init_weights
def _init_weights(self, module: nn.Module):
"""Initialize the weights"""
if isinstance(module, nn.Linear):
# Slightly different from the TF version which uses truncated_normal for initialization
# cf https://github.com/pytorch/pytorch/pull/5617
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
if module.bias is not None:
module.bias.data.zero_()
elif isinstance(module, nn.Embedding):
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
if module.padding_idx is not None:
module.weight.data[module.padding_idx].zero_()
elif isinstance(module, nn.LayerNorm):
module.bias.data.zero_()
module.weight.data.fill_(1.0)
def _set_gradient_checkpointing(self, module, value=False):
if isinstance(module, RnaErnieEncoder):
module.gradient_checkpointing = value
|