跳转至

RNA-FM

Pre-trained model on non-coding RNA (ncRNA) using a masked language modeling (MLM) objective.

Disclaimer

This is an UNOFFICIAL implementation of the Interpretable RNA Foundation Model from Unannotated Data for Highly Accurate RNA Structure and Function Predictions by Jiayang Chen, Zhihang Hu, Siqi Sun, et al.

The OFFICIAL repository of RNA-FM is at ml4bio/RNA-FM.

Tip

The MultiMolecule team has confirmed that the provided model and checkpoints are producing the same intermediate representations as the original implementation.

The team releasing RNA-FM did not write this model card for this model so this model card has been written by the MultiMolecule team.

Model Details

RNA-FM 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.

Variants

Model Specification

Variants Num Layers Hidden Size Num Heads Intermediate Size Num Parameters (M) FLOPs (G) MACs (G) Max Num Tokens
mRNA-FM 12 1280 20 5120 239.26 258.08 128.85 1024
RNA-FM 640 99.52 109.02 54.36

Usage

The model file depends on the multimolecule library. You can install it using pip:

Bash
1
pip install multimolecule

Direct Use

Masked Language Modeling

You can use this model directly with a pipeline for masked language modeling:

Python
1
2
3
4
5
import multimolecule  # you must import multimolecule to register models
from transformers import pipeline

predictor = pipeline("fill-mask", model="multimolecule/rnafm")
output = predictor("gguc<mask>cucugguuagaccagaucugagccu")

Downstream Use

Extract Features

Here is how to use this model to get the features of a given sequence in PyTorch:

Python
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
from multimolecule import RnaTokenizer, RnaFmModel


tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnafm")
model = RnaFmModel.from_pretrained("multimolecule/rnafm")

text = "UAGCUUAUCAGACUGAUGUUG"
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
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
import torch
from multimolecule import RnaTokenizer, RnaFmForSequencePrediction


tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnafm")
model = RnaFmForSequencePrediction.from_pretrained("multimolecule/rnafm")

text = "UAGCUUAUCAGACUGAUGUUG"
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 token classification or regression.

Here is how to use this model as backbone to fine-tune for a nucleotide-level task in PyTorch:

Python
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
import torch
from multimolecule import RnaTokenizer, RnaFmForTokenPrediction


tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnafm")
model = RnaFmForTokenPrediction.from_pretrained("multimolecule/rnafm")

text = "UAGCUUAUCAGACUGAUGUUG"
input = tokenizer(text, return_tensors="pt")
label = torch.randint(2, (len(text), ))

output = model(**input, labels=label)

Contact 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 contact classification or regression.

Here is how to use this model as backbone to fine-tune for a contact-level task in PyTorch:

Python
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
import torch
from multimolecule import RnaTokenizer, RnaFmForContactPrediction


tokenizer = RnaTokenizer.from_pretrained("multimolecule/rnafm")
model = RnaFmForContactPrediction.from_pretrained("multimolecule/rnafm")

text = "UAGCUUAUCAGACUGAUGUUG"
input = tokenizer(text, return_tensors="pt")
label = torch.randint(2, (len(text), len(text)))

output = model(**input, labels=label)

Training Details

RNA-FM 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 RNA-FM 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.

RNA-FM applied CD-HIT (CD-HIT-EST) with a cut-off at 100% sequence identity to remove redundancy from the RNAcentral. The final dataset contains 23.7 million non-redundant RNA sequences.

RNA-FM preprocessed all tokens by replacing “U”s with “T”s.

Note that during model conversions, “T” is replaced with “U”. 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

RNA-FM used masked language modeling (MLM) as the pre-training objective. The masking procedure is similar to the one used in BERT:

  • Mask rate: 15%
  • Replacement: <mask> for 80% of masked tokens
  • Replacement: random token for 10% of masked tokens
  • Replacement: unchanged token for 10% of masked tokens

Pre-training

The model was trained on 8 NVIDIA A100 GPUs with 80GiB memories.

  • Learning rate: 1e-4
  • Learning rate scheduler: Inverse square root
  • Learning rate warm-up: 10,000 steps
  • Weight decay: 0.01

Citation

BibTeX
1
2
3
4
5
6
@article{chen2022interpretable,
  title={Interpretable rna foundation model from unannotated data for highly accurate rna structure and function predictions},
  author={Chen, Jiayang and Hu, Zhihang and Sun, Siqi and Tan, Qingxiong and Wang, Yixuan and Yu, Qinze and Zong, Licheng and Hong, Liang and Xiao, Jin and King, Irwin and others},
  journal={arXiv preprint arXiv:2204.00300},
  year={2022}
}

Note

The artifacts distributed in this repository are part of the MultiMolecule project. If MultiMolecule supports your research, please cite the MultiMolecule project as follows:

BibTeX
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
@software{chen_2024_12638419,
  author    = {Chen, Zhiyuan and Zhu, Sophia Y.},
  title     = {MultiMolecule},
  doi       = {10.5281/zenodo.12638419},
  publisher = {Zenodo},
  url       = {https://doi.org/10.5281/zenodo.12638419},
  year      = 2024,
  month     = may,
  day       = 4
}

Contact

Please use GitHub issues of MultiMolecule for any questions or comments on the model card.

Please contact the authors of the RNA-FM paper for questions or comments on the paper/model.

License

This model implementation is licensed under the GNU Affero General Public License.

For additional terms and clarifications, please refer to our License FAQ.

Text Only
1
SPDX-License-Identifier: AGPL-3.0-or-later

API Reference

RnaFmConfig

Bases: PreTrainedConfig

This is the configuration class to store the configuration of a RnaFmModel. It is used to instantiate a RNA-FM 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 RNA-FM ml4bio/RNA-FM 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 | None

Vocabulary size of the RNA-FM model. Defines the number of different tokens that can be represented by the input_ids passed when calling [RnaFmModel]. Defaults to 26 if codon=False else 131.

 None

codon

bool

Whether to use codon tokenization.

 False

hidden_size

int

Dimensionality of the encoder layers and the pooler layer.

 640

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.

 20

intermediate_size

int

Dimensionality of the “intermediate” (often named feed-forward) layer in the Transformer encoder.

 5120

hidden_act

str

The non-linear activation function (function or string) in the encoder and pooler. If string, "gelu", "relu", "silu" and "gelu_new" are supported.

 'gelu'

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).

 1026

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-05

position_embedding_type

str

Type of position embedding. Choose one of "absolute", "relative_key", "relative_key_query", "rotary". For positional embeddings use "absolute". For more information on "relative_key", please refer to Self-Attention with Relative Position Representations (Shaw et al.). For more information on "relative_key_query", please refer to Method 4 in Improve Transformer Models with Better Relative Position Embeddings (Huang et al.).

 'absolute'

is_decoder

bool

Whether the model is used as a decoder or not. If False, the model is used as an encoder.

 False

use_cache

bool

Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if config.is_decoder=True.

 True

embed_norm

bool

Whether to apply layer normalization after embeddings but before the main stem of the network.

 True

token_dropout

bool

When this is enabled, masked tokens are treated as if they had been dropped out by input dropout.

 False

head

HeadConfig | None

The configuration of the prediction head.

 None

lm_head

MaskedLMHeadConfig | None

The configuration of the masked language model head.

 None

add_cross_attention

bool

Whether to add cross-attention layers when the model is used as a decoder.

 False

Examples:

Python Console Session
1
2
3
4
5
6
7
>>> from multimolecule import RnaFmConfig, RnaFmModel
>>> # Initializing a RNA-FM multimolecule/rnafm style configuration
>>> configuration = RnaFmConfig()
>>> # Initializing a model (with random weights) from the multimolecule/rnafm style configuration
>>> model = RnaFmModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
Source code in multimolecule/models/rnafm/configuration_rnafm.py
Python
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
class RnaFmConfig(PreTrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`RnaFmModel`][multimolecule.models.RnaFmModel].
    It is used to instantiate a RNA-FM 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 RNA-FM
    [ml4bio/RNA-FM](https://github.com/ml4bio/RNA-FM) 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 RNA-FM model. Defines the number of different tokens that can be represented by the
            `input_ids` passed when calling [`RnaFmModel`].
            Defaults to 26 if `codon=False` else 131.
        codon:
            Whether to use codon tokenization.
        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_act:
            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
            `"relu"`, `"silu"` and `"gelu_new"` are supported.
        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.
        position_embedding_type:
            Type of position embedding. Choose one of `"absolute"`, `"relative_key"`, `"relative_key_query"`,
            `"rotary"`.
            For positional embeddings use `"absolute"`. For more information on `"relative_key"`, please refer to
            [Self-Attention with Relative Position Representations (Shaw et al.)](https://arxiv.org/abs/1803.02155).
            For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models
            with Better Relative Position Embeddings (Huang et al.)](https://arxiv.org/abs/2009.13658).
        is_decoder:
            Whether the model is used as a decoder or not. If `False`, the model is used as an encoder.
        use_cache:
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if `config.is_decoder=True`.
        embed_norm:
            Whether to apply layer normalization after embeddings but before the main stem of the network.
        token_dropout:
            When this is enabled, masked tokens are treated as if they had been dropped out by input dropout.
        head:
            The configuration of the prediction head.
        lm_head:
            The configuration of the masked language model head.
        add_cross_attention:
            Whether to add cross-attention layers when the model is used as a decoder.

    Examples:
        >>> from multimolecule import RnaFmConfig, RnaFmModel
        >>> # Initializing a RNA-FM multimolecule/rnafm style configuration
        >>> configuration = RnaFmConfig()
        >>> # Initializing a model (with random weights) from the multimolecule/rnafm style configuration
        >>> model = RnaFmModel(configuration)
        >>> # Accessing the model configuration
        >>> configuration = model.config
    """

    model_type = "rnafm"

    def __init__(
        self,
        vocab_size: int | None = None,
        codon: bool = False,
        hidden_size: int = 640,
        num_hidden_layers: int = 12,
        num_attention_heads: int = 20,
        intermediate_size: int = 5120,
        hidden_act: str = "gelu",
        hidden_dropout: float = 0.1,
        attention_dropout: float = 0.1,
        max_position_embeddings: int = 1026,
        initializer_range: float = 0.02,
        layer_norm_eps: float = 1e-5,
        position_embedding_type: str = "absolute",
        is_decoder: bool = False,
        use_cache: bool = True,
        embed_norm: bool = True,
        token_dropout: bool = False,
        head: HeadConfig | None = None,
        lm_head: MaskedLMHeadConfig | None = None,
        add_cross_attention: bool = False,
        **kwargs,
    ):
        super().__init__(**kwargs)
        expected_vocab_size = 131 if codon else 28
        if vocab_size is None:
            vocab_size = expected_vocab_size
        elif vocab_size != expected_vocab_size:
            raise ValueError(f"vocab_size ({vocab_size}) must be {expected_vocab_size} when codon={codon}.")
        validate_attention_dimensions(hidden_size, num_attention_heads)
        self.vocab_size = vocab_size
        self.codon = codon
        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.embed_norm = embed_norm
        self.token_dropout = token_dropout
        self.head = HeadConfig(**head) if head is not None else None
        lm_head_kwargs = dict(lm_head or {})
        lm_head_kwargs.setdefault("layer_norm_eps", layer_norm_eps)
        self.lm_head = MaskedLMHeadConfig(**lm_head_kwargs)
        self.add_cross_attention = add_cross_attention

RnaFmSecondaryStructureHeadConfig

Bases: BaseHeadConfig

Configuration class for a prediction head.

Parameters:

Name Type Description Default

num_labels

Number of labels to use in the last layer added to the model, typically for a classification task.

Head should look for Config.num_labels if is None.

 required

problem_type

Problem type for XxxForYyyPrediction models. Can be one of "binary", "regression", "multiclass" or "multilabel".

Head should look for Config.problem_type if is None.

 required

dropout

The dropout ratio for the hidden states.

 required

kernel_size

Kernel size of the 1D convolutional layers in the head.

 required

num_layers

Number of convolutional layers in the head.

 required

num_channels

Number of channels in the convolutional layers.

 required

bias

Whether to include bias terms in the convolutional layers.

 required

activation

Activation function used in the head.

 required

hidden_size

Hidden size used by the head.

 required

intermediate_size

Intermediate size used by the head.

 required
Source code in multimolecule/models/rnafm/configuration_rnafm.py
Python
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
class RnaFmSecondaryStructureHeadConfig(BaseHeadConfig):
    r"""
    Configuration class for a prediction head.

    Args:
        num_labels:
            Number of labels to use in the last layer added to the model, typically for a classification task.

            Head should look for [`Config.num_labels`][multimolecule.PreTrainedConfig] if is `None`.
        problem_type:
            Problem type for `XxxForYyyPrediction` models. Can be one of `"binary"`, `"regression"`,
            `"multiclass"` or `"multilabel"`.

            Head should look for [`Config.problem_type`][multimolecule.PreTrainedConfig] if is `None`.
        dropout:
            The dropout ratio for the hidden states.
        kernel_size:
            Kernel size of the 1D convolutional layers in the head.
        num_layers:
            Number of convolutional layers in the head.
        num_channels:
            Number of channels in the convolutional layers.
        bias:
            Whether to include bias terms in the convolutional layers.
        activation:
            Activation function used in the head.
        hidden_size:
            Hidden size used by the head.
        intermediate_size:
            Intermediate size used by the head.
    """

    num_labels: int = 1
    problem_type: str | None = None
    dropout: float = 0.3
    kernel_size: int = 7
    num_layers: int = 32
    num_channels: int = 64
    bias: bool = False
    activation: str = "relu"
    hidden_size: int = 128
    intermediate_size: int = 256

RnaFmForContactPrediction

Bases: RnaFmPreTrainedModel

Examples:

Python Console Session
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmForContactPrediction, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmForContactPrediction(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, 1])
>>> output["loss"]
tensor(..., grad_fn=<BinaryCrossEntropyWithLogitsBackward0>)
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
class RnaFmForContactPrediction(RnaFmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmForContactPrediction, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmForContactPrediction(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, 1])
        >>> output["loss"]  # doctest:+ELLIPSIS
        tensor(..., grad_fn=<BinaryCrossEntropyWithLogitsBackward0>)
    """

    def __init__(self, config: RnaFmConfig):
        super().__init__(config)
        self.model = RnaFmModel(config, add_pooling_layer=False)
        self.contact_head = ContactPredictionHead(config)
        self.head_config = self.contact_head.config
        self.require_attentions = self.contact_head.require_attentions

        # Initialize weights and apply final processing
        self.post_init()

    @can_return_tuple
    def forward(
        self,
        input_ids: Tensor | NestedTensor | None = None,
        attention_mask: Tensor | None = None,
        position_ids: Tensor | None = None,
        inputs_embeds: Tensor | NestedTensor | None = None,
        labels: Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[Tensor, ...] | ContactPredictorOutput:
        if self.require_attentions:
            output_attentions = kwargs.get("output_attentions", self.config.output_attentions)
            if output_attentions is False:
                warn("output_attentions must be True since prediction head requires attentions.")
            kwargs["output_attentions"] = True
        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            return_dict=True,
            **kwargs,
        )

        output = self.contact_head(outputs, attention_mask, input_ids, labels)
        logits, loss = output.logits, output.loss

        return ContactPredictorOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

RnaFmForMaskedLM

Bases: RnaFmPreTrainedModel

Examples:

Python Console Session
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmForMaskedLM, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmForMaskedLM(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, 28])
>>> output["loss"]
tensor(..., grad_fn=<NllLossBackward0>)
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
class RnaFmForMaskedLM(RnaFmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmForMaskedLM, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmForMaskedLM(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, 28])
        >>> output["loss"]  # doctest:+ELLIPSIS
        tensor(..., grad_fn=<NllLossBackward0>)
    """

    _tied_weights_keys = {
        "lm_head.decoder.weight": "model.embeddings.word_embeddings.weight",
        "lm_head.decoder.bias": "lm_head.bias",
    }

    def __init__(self, config: RnaFmConfig):
        super().__init__(config)
        if config.is_decoder:
            warn(
                "If you want to use `RnaFmForMaskedLM` make sure `config.is_decoder=False` for "
                "bi-directional self-attention."
            )
        self.model = RnaFmModel(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, embeddings):
        self.lm_head.decoder = embeddings
        if hasattr(self.lm_head, "bias"):
            self.lm_head.bias = embeddings.bias

    @can_return_tuple
    def forward(
        self,
        input_ids: Tensor | NestedTensor | None = None,
        attention_mask: Tensor | None = None,
        position_ids: 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,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[Tensor, ...] | MaskedLMOutput:
        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            return_dict=True,
            **kwargs,
        )

        output = self.lm_head(outputs, labels)
        logits, loss = output.logits, output.loss

        return MaskedLMOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

RnaFmForPreTraining

Bases: RnaFmForMaskedLM

Examples:

Python Console Session
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmForPreTraining, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmForPreTraining(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input, labels_lm=input["input_ids"])
>>> output["loss"]
tensor(..., grad_fn=<MeanBackward0>)
>>> output["logits_lm"].shape
torch.Size([1, 7, 28])
>>> output["logits_ss"].shape
torch.Size([1, 5, 5, 1])
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
class RnaFmForPreTraining(RnaFmForMaskedLM):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmForPreTraining, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmForPreTraining(config)
        >>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
        >>> input = tokenizer("ACGUN", return_tensors="pt")
        >>> output = model(**input, labels_lm=input["input_ids"])
        >>> output["loss"]  # doctest:+ELLIPSIS
        tensor(..., grad_fn=<MeanBackward0>)
        >>> output["logits_lm"].shape
        torch.Size([1, 7, 28])
        >>> output["logits_ss"].shape
        torch.Size([1, 5, 5, 1])
    """

    def __init__(self, config: RnaFmConfig):
        super().__init__(config)
        self.ss_head = ContactAttentionHead(config)
        self.require_attentions = self.ss_head.require_attentions

        # Initialize weights and apply final processing
        self.post_init()

    @can_return_tuple
    def forward(  # type: ignore[override]
        self,
        input_ids: Tensor | NestedTensor | None = None,
        attention_mask: Tensor | None = None,
        position_ids: Tensor | None = None,
        inputs_embeds: Tensor | NestedTensor | None = None,
        encoder_hidden_states: Tensor | None = None,
        encoder_attention_mask: Tensor | None = None,
        labels_lm: Tensor | None = None,
        labels_ss: Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[Tensor, ...] | RnaFmForPreTrainingOutput:
        if self.require_attentions:
            output_attentions = kwargs.get("output_attentions", self.config.output_attentions)
            if output_attentions is False:
                warn("output_attentions must be True since prediction head requires attentions.")
            kwargs["output_attentions"] = True
        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            return_dict=True,
            **kwargs,
        )

        output_lm = self.lm_head(outputs, labels=labels_lm)
        logits_lm, loss_lm = output_lm.logits, output_lm.loss

        output_ss = self.ss_head(outputs, attention_mask, input_ids, labels=labels_ss)
        logits_ss, loss_ss = output_ss.logits, output_ss.loss

        losses = tuple(l for l in (loss_lm, loss_ss) if l is not None)  # noqa: E741
        loss = torch.mean(torch.stack(losses)) if losses else None

        return RnaFmForPreTrainingOutput(
            loss=loss,
            logits_lm=logits_lm,
            loss_lm=loss_lm,
            logits_ss=logits_ss,
            loss_ss=loss_ss,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

RnaFmForSecondaryStructurePrediction

Bases: RnaFmForPreTraining

Examples:

Python Console Session
1
2
3
4
5
6
7
8
9
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmForSecondaryStructurePrediction, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmForSecondaryStructurePrediction(config)
>>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
>>> input = tokenizer("ACGUN", return_tensors="pt")
>>> output = model(**input)
>>> output["logits"].shape
torch.Size([1, 5, 5, 1])
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
class RnaFmForSecondaryStructurePrediction(RnaFmForPreTraining):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmForSecondaryStructurePrediction, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmForSecondaryStructurePrediction(config)
        >>> tokenizer = RnaTokenizer.from_pretrained("multimolecule/rna")
        >>> input = tokenizer("ACGUN", return_tensors="pt")
        >>> output = model(**input)
        >>> output["logits"].shape
        torch.Size([1, 5, 5, 1])
    """

    # Note: inherits from RnaFmForPreTraining (rather than RnaFmPreTrainedModel directly) to retain
    # `lm_head` in the state dict. The published rnafm-ss checkpoint was saved with lm_head weights
    # present; changing the inheritance would break checkpoint loading.

    def __init__(self, config: RnaFmConfig):
        super().__init__(config)
        self.model = RnaFmModel(config, add_pooling_layer=False)
        self.ss_head = RnaFmSecondaryStructurePredictionHead(config)

        # Initialize weights and apply final processing
        self.post_init()

    @can_return_tuple
    def forward(  # type: ignore[override]
        self,
        input_ids: Tensor | NestedTensor,
        attention_mask: Tensor | None = None,
        position_ids: 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,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[Tensor, ...] | ContactPredictorOutput:
        if self.require_attentions:
            output_attentions = kwargs.get("output_attentions", self.config.output_attentions)
            if output_attentions is False:
                warn("output_attentions must be True since prediction head requires attentions.")
            kwargs["output_attentions"] = True
        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            return_dict=True,
            **kwargs,
        )

        output = self.ss_head(outputs, attention_mask, input_ids, labels=labels)
        logits, loss = output.logits, output.loss

        return ContactPredictorOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

RnaFmForSequencePrediction

Bases: RnaFmPreTrainedModel

Examples:

Python Console Session
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmForSequencePrediction, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmForSequencePrediction(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, 1])
>>> output["loss"]
tensor(..., grad_fn=<BinaryCrossEntropyWithLogitsBackward0>)
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
class RnaFmForSequencePrediction(RnaFmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmForSequencePrediction, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmForSequencePrediction(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, 1])
        >>> output["loss"]  # doctest:+ELLIPSIS
        tensor(..., grad_fn=<BinaryCrossEntropyWithLogitsBackward0>)
    """

    def __init__(self, config: RnaFmConfig):
        super().__init__(config)
        self.model = RnaFmModel(config)
        self.sequence_head = SequencePredictionHead(config)
        self.head_config = self.sequence_head.config

        # Initialize weights and apply final processing
        self.post_init()

    @can_return_tuple
    def forward(
        self,
        input_ids: Tensor | NestedTensor | None = None,
        attention_mask: Tensor | None = None,
        position_ids: Tensor | None = None,
        inputs_embeds: Tensor | NestedTensor | None = None,
        labels: Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[Tensor, ...] | SequencePredictorOutput:
        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            return_dict=True,
            **kwargs,
        )

        output = self.sequence_head(outputs, labels)
        logits, loss = output.logits, output.loss

        return SequencePredictorOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

RnaFmForTokenPrediction

Bases: RnaFmPreTrainedModel

Examples:

Python Console Session
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmForTokenPrediction, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmForTokenPrediction(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, 1])
>>> output["loss"]
tensor(..., grad_fn=<BinaryCrossEntropyWithLogitsBackward0>)
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
class RnaFmForTokenPrediction(RnaFmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmForTokenPrediction, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmForTokenPrediction(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, 1])
        >>> output["loss"]  # doctest:+ELLIPSIS
        tensor(..., grad_fn=<BinaryCrossEntropyWithLogitsBackward0>)
    """

    def __init__(self, config: RnaFmConfig):
        super().__init__(config)
        self.model = RnaFmModel(config, add_pooling_layer=False)
        self.token_head = TokenPredictionHead(config)
        self.head_config = self.token_head.config

        # Initialize weights and apply final processing
        self.post_init()

    @can_return_tuple
    def forward(
        self,
        input_ids: Tensor | NestedTensor | None = None,
        attention_mask: Tensor | None = None,
        position_ids: Tensor | None = None,
        inputs_embeds: Tensor | NestedTensor | None = None,
        labels: Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[Tensor, ...] | TokenPredictorOutput:
        outputs = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            return_dict=True,
            **kwargs,
        )

        output = self.token_head(outputs, attention_mask, input_ids, labels)
        logits, loss = output.logits, output.loss

        return TokenPredictorOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

RnaFmModel

Bases: RnaFmPreTrainedModel

Examples:

Python Console Session
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
>>> import torch
>>> from multimolecule import RnaFmConfig, RnaFmModel, RnaTokenizer
>>> config = RnaFmConfig()
>>> model = RnaFmModel(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, 640])
>>> output["pooler_output"].shape
torch.Size([1, 640])
Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
class RnaFmModel(RnaFmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import RnaFmConfig, RnaFmModel, RnaTokenizer
        >>> config = RnaFmConfig()
        >>> model = RnaFmModel(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, 640])
        >>> output["pooler_output"].shape
        torch.Size([1, 640])
    """

    def __init__(self, config: RnaFmConfig, add_pooling_layer: bool = True):
        super().__init__(config)
        self.pad_token_id = config.pad_token_id
        self.embeddings = RnaFmEmbeddings(config)
        self.encoder = RnaFmEncoder(config)
        self.pooler = RnaFmPooler(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

    @merge_with_config_defaults
    @capture_outputs
    def forward(
        self,
        input_ids: Tensor | NestedTensor | None = None,
        attention_mask: Tensor | None = None,
        position_ids: Tensor | None = None,
        inputs_embeds: Tensor | NestedTensor | None = None,
        encoder_hidden_states: Tensor | None = None,
        encoder_attention_mask: Tensor | None = None,
        past_key_values: Cache | None = None,
        use_cache: bool | None = None,
        cache_position: Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> 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.num_hidden_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 self.config.is_decoder:
            use_cache = use_cache if use_cache is not None else self.config.use_cache
        else:
            use_cache = False
        if use_cache and past_key_values is None:
            past_key_values = (
                EncoderDecoderCache(DynamicCache(config=self.config), DynamicCache(config=self.config))
                if encoder_hidden_states is not None or self.config.is_encoder_decoder
                else DynamicCache(config=self.config)
            )

        if isinstance(input_ids, NestedTensor) and attention_mask is None:
            attention_mask = input_ids.mask
        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
        if input_ids is not None:
            device = input_ids.device
            seq_length = input_ids.shape[1]
        else:
            device = inputs_embeds.device  # type: ignore[union-attr]
            seq_length = inputs_embeds.shape[1]  # type: ignore[union-attr]

        # past_key_values_length
        past_key_values_length = past_key_values.get_seq_length() if past_key_values is not None else 0
        if cache_position is None:
            cache_position = torch.arange(past_key_values_length, past_key_values_length + seq_length, device=device)

        if attention_mask is None and input_ids is not None and self.pad_token_id is not None:
            attention_mask = input_ids.ne(self.pad_token_id)

        embedding_output = self.embeddings(
            input_ids=input_ids,
            position_ids=position_ids,
            attention_mask=attention_mask,
            inputs_embeds=inputs_embeds,
            past_key_values_length=past_key_values_length,
        )
        attention_mask, encoder_attention_mask = self._create_attention_masks(
            attention_mask=attention_mask,
            encoder_attention_mask=encoder_attention_mask,
            embedding_output=embedding_output,
            encoder_hidden_states=encoder_hidden_states,
            cache_position=cache_position,
            past_key_values=past_key_values,
        )
        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask,
            encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            cache_position=cache_position,
            position_ids=position_ids,
            **kwargs,
        )
        sequence_output = encoder_outputs.last_hidden_state
        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None

        return BaseModelOutputWithPoolingAndCrossAttentions(
            last_hidden_state=sequence_output,
            pooler_output=pooled_output,
            past_key_values=encoder_outputs.past_key_values,
        )

    def _create_attention_masks(
        self,
        attention_mask,
        encoder_attention_mask,
        embedding_output,
        encoder_hidden_states,
        cache_position,
        past_key_values,
    ):
        if self.config.is_decoder:
            attention_mask = create_causal_mask(
                config=self.config,
                inputs_embeds=embedding_output,
                attention_mask=attention_mask,
                cache_position=cache_position,
                past_key_values=past_key_values,
            )
        else:
            attention_mask = create_bidirectional_mask(
                config=self.config, inputs_embeds=embedding_output, attention_mask=attention_mask
            )

        if encoder_attention_mask is not None:
            encoder_attention_mask = create_bidirectional_mask(
                config=self.config,
                inputs_embeds=embedding_output,
                attention_mask=encoder_attention_mask,
                encoder_hidden_states=encoder_hidden_states,
            )

        return attention_mask, encoder_attention_mask

forward

Python
forward(
    input_ids: Tensor | NestedTensor | None = None,
    attention_mask: Tensor | None = None,
    position_ids: Tensor | None = None,
    inputs_embeds: Tensor | NestedTensor | None = None,
    encoder_hidden_states: Tensor | None = None,
    encoder_attention_mask: Tensor | None = None,
    past_key_values: Cache | None = None,
    use_cache: bool | None = None,
    cache_position: Tensor | None = None,
    **kwargs: Unpack[TransformersKwargs]
) -> (
    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

Cache | None

tuple of length config.num_hidden_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/rnafm/modeling_rnafm.py
Python
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
@merge_with_config_defaults
@capture_outputs
def forward(
    self,
    input_ids: Tensor | NestedTensor | None = None,
    attention_mask: Tensor | None = None,
    position_ids: Tensor | None = None,
    inputs_embeds: Tensor | NestedTensor | None = None,
    encoder_hidden_states: Tensor | None = None,
    encoder_attention_mask: Tensor | None = None,
    past_key_values: Cache | None = None,
    use_cache: bool | None = None,
    cache_position: Tensor | None = None,
    **kwargs: Unpack[TransformersKwargs],
) -> 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.num_hidden_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 self.config.is_decoder:
        use_cache = use_cache if use_cache is not None else self.config.use_cache
    else:
        use_cache = False
    if use_cache and past_key_values is None:
        past_key_values = (
            EncoderDecoderCache(DynamicCache(config=self.config), DynamicCache(config=self.config))
            if encoder_hidden_states is not None or self.config.is_encoder_decoder
            else DynamicCache(config=self.config)
        )

    if isinstance(input_ids, NestedTensor) and attention_mask is None:
        attention_mask = input_ids.mask
    if (input_ids is None) ^ (inputs_embeds is not None):
        raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
    if input_ids is not None:
        device = input_ids.device
        seq_length = input_ids.shape[1]
    else:
        device = inputs_embeds.device  # type: ignore[union-attr]
        seq_length = inputs_embeds.shape[1]  # type: ignore[union-attr]

    # past_key_values_length
    past_key_values_length = past_key_values.get_seq_length() if past_key_values is not None else 0
    if cache_position is None:
        cache_position = torch.arange(past_key_values_length, past_key_values_length + seq_length, device=device)

    if attention_mask is None and input_ids is not None and self.pad_token_id is not None:
        attention_mask = input_ids.ne(self.pad_token_id)

    embedding_output = self.embeddings(
        input_ids=input_ids,
        position_ids=position_ids,
        attention_mask=attention_mask,
        inputs_embeds=inputs_embeds,
        past_key_values_length=past_key_values_length,
    )
    attention_mask, encoder_attention_mask = self._create_attention_masks(
        attention_mask=attention_mask,
        encoder_attention_mask=encoder_attention_mask,
        embedding_output=embedding_output,
        encoder_hidden_states=encoder_hidden_states,
        cache_position=cache_position,
        past_key_values=past_key_values,
    )
    encoder_outputs = self.encoder(
        embedding_output,
        attention_mask,
        encoder_hidden_states,
        encoder_attention_mask=encoder_attention_mask,
        past_key_values=past_key_values,
        use_cache=use_cache,
        cache_position=cache_position,
        position_ids=position_ids,
        **kwargs,
    )
    sequence_output = encoder_outputs.last_hidden_state
    pooled_output = self.pooler(sequence_output) if self.pooler is not None else None

    return BaseModelOutputWithPoolingAndCrossAttentions(
        last_hidden_state=sequence_output,
        pooler_output=pooled_output,
        past_key_values=encoder_outputs.past_key_values,
    )

RnaFmPreTrainedModel

Bases: PreTrainedModel

An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.

Source code in multimolecule/models/rnafm/modeling_rnafm.py
Python
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
class RnaFmPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = RnaFmConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _supports_flash_attn = True
    _supports_sdpa = True
    _supports_flex_attn = True
    _supports_attention_backend = True
    _can_record_outputs: dict[str, Any] | None = None
    _no_split_modules = ["RnaFmLayer", "RnaFmEmbeddings"]

    @torch.no_grad()
    def _init_weights(self, module: nn.Module):
        super()._init_weights(module)
        if isinstance(module, RnaFmEmbeddings):
            # `position_ids` is a non-persistent buffer; under transformers v5 meta-init it
            # stays on the meta device after `from_pretrained`, so populate it here.
            init.copy_(module.position_ids, torch.arange(module.position_ids.shape[-1]).expand((1, -1)))