CaLM¶

Pre-trained model on protein-coding DNA (cDNA) using a masked language modeling (MLM) objective.

Statement¶

Codon language embeddings provide strong signals for use in protein engineering 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 Codon language embeddings provide strong signals for use in protein engineering by Carlos Outeiral and Charlotte M. Deane.

The OFFICIAL repository of CaLM is at oxpig/CaLM.

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 CaLM did not write this model card for this model so this model card has been written by the MultiMolecule team.

Model Details¶

CaLM is a bert-style model pre-trained on a large corpus of protein-coding DNA sequences in a self-supervised fashion. This means that the model was trained on the raw nucleotides of DNA 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.

Model Specification¶

Num Layers	Hidden Size	Num Heads	Intermediate Size	Num Parameters (M)	FLOPs (G)	MACs (G)	Max Num Tokens
12	768	12	3072	85.75	22.36	11.17	1024

Links¶

Code: multimolecule.calm
Weights: multimolecule/calm
Data: European Nucleotide Archive
Paper: Codon language embeddings provide strong signals for use in protein engineering
Developed by: Carlos Outeiral, Charlotte M. Deane
Model type: BERT - ESM
Original Repository: oxpig/CaLM

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
import multimolecule  # you must import multimolecule to register models
from transformers import pipeline

predictor = pipeline("fill-mask", model="multimolecule/calm")
output = predictor("agc<mask>cattatggcgaaccttggctgctg")

Downstream Use¶

Extract Features¶

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

Python
from multimolecule import RnaTokenizer, CaLmModel


tokenizer = RnaTokenizer.from_pretrained("multimolecule/calm")
model = CaLmModel.from_pretrained("multimolecule/calm")

text = "GCCAGTCGCTGACAGCCGCGG"
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, CaLmForSequencePrediction


tokenizer = RnaTokenizer.from_pretrained("multimolecule/calm")
model = CaLmForSequencePrediction.from_pretrained("multimolecule/calm")

text = "GCCAGTCGCTGACAGCCGCGG"
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
import torch
from multimolecule import RnaTokenizer, CaLmForTokenPrediction


tokenizer = RnaTokenizer.from_pretrained("multimolecule/calm")
model = CaLmForTokenPrediction.from_pretrained("multimolecule/calm")

text = "GCCAGTCGCTGACAGCCGCGG"
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
import torch
from multimolecule import RnaTokenizer, CaLmForContactPrediction


tokenizer = RnaTokenizer.from_pretrained("multimolecule/calm")
model = CaLmForContactPrediction.from_pretrained("multimolecule/calm")

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

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

Training Details¶

CaLM used Masked Language Modeling (MLM) as the pre-training objective: taking a sequence, the model randomly masks 25% 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 CaLM model was pre-trained coding sequences of all organisms available on the European Nucleotide Archive (ENA). European Nucleotide Archive provides a comprehensive record of the world’s nucleotide sequencing information, covering raw sequencing data, sequence assembly information and functional annotation.

CaLM collected coding sequences of all organisms from ENA on April 2022, including 114,214,475 sequences. Only high level assembly information (dataclass CON) were used. Sequences matching the following criteria were filtered out:

with unknown nucleotides (N, Y, R)
start codon is not ATG
contains interstitial stop codons
number of nucleotides is not a multiple of three

To reduce redundancy, CaLM grouped the entries by organism, and apply CD-HIT (CD-HIT-EST) with a cut-off at 40% sequence identity to the translated protein sequences.

The final dataset contains 9,858,385 cDNA sequences.

Note that the alphabet in the original implementation is RNA instead of DNA, therefore, we use RnaTokenizer to tokenize the sequences. RnaTokenizer of multimolecule will convert “U”s to “T”s for you, you may disable this behaviour by passing replace_T_with_U=False.

Training Procedure¶

Preprocessing¶

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

25% 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.

Pre-training¶

The model was trained on 4 NVIDIA Quadro RTX4000 GPUs with 8GiB memories.

Batch Size: 1,000
Epochs: 14
Optimizer: AdamW
Learning rate: 1e-4
Learning rate scheduler: Cosine
Learning rate warm-up: 1,000 steps

Citation¶

BibTeX
@article {outeiral2022coodn,
    author = {Outeiral, Carlos and Deane, Charlotte M.},
    title = {Codon language embeddings provide strong signals for protein engineering},
    elocation-id = {2022.12.15.519894},
    year = {2022},
    doi = {10.1101/2022.12.15.519894},
    publisher = {Cold Spring Harbor Laboratory},
    abstract = {Protein representations from deep language models have yielded state-of-the-art performance across many tasks in computational protein engineering. In recent years, progress has primarily focused on parameter count, with recent models{\textquoteright} capacities surpassing the size of the very datasets they were trained on. Here, we propose an alternative direction. We show that large language models trained on codons, instead of amino acid sequences, provide high-quality representations that outperform comparable state-of-the-art models across a variety of tasks. In some tasks, like species recognition, prediction of protein and transcript abundance, or melting point estimation, we show that a language model trained on codons outperforms every other published protein language model, including some that contain over 50 times more parameters. These results suggest that, in addition to commonly studied scale and model complexity, the information content of biological data provides an orthogonal direction to improve the power of machine learning in biology.Competing Interest StatementThe authors have declared no competing interest.},
    URL = {https://www.biorxiv.org/content/early/2022/12/19/2022.12.15.519894},
    eprint = {https://www.biorxiv.org/content/early/2022/12/19/2022.12.15.519894.full.pdf},
    journal = {bioRxiv}
}

Note

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

BibTeX
@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 CaLM paper for questions or comments on the paper/model.

License¶

This model 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`

multimolecule.models.calm ¶

CaLmConfig ¶

Bases: PreTrainedConfig

This is the configuration class to store the configuration of a CaLmModel. It is used to instantiate a CaLM 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 CaLM oxpig/CaLM 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 CaLM model. Defines the number of different tokens that can be represented by the `input_ids` passed when calling [`CaLmModel`]. Defaults to 131 if `codon=True` else 26.	`None`
`codon` ¶	`bool`	Whether to use codon tokenization.	`True`
`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`
`intermediate_size` ¶	`int`	Dimensionality of the “intermediate” (often named feed-forward) layer in the Transformer encoder.	`3072`
`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-12`
`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.).	`'rotary'`
`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`
`emb_layer_norm_before` ¶	`bool`	Whether to apply layer normalization after embeddings but before the main stem of the network.	`False`
`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 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
>>> from multimolecule import CaLmConfig, CaLmModel
>>> # Initializing a CaLM multimolecule/calm style configuration
>>> configuration = CaLmConfig()
>>> # Initializing a model (with random weights) from the multimolecule/calm style configuration
>>> model = CaLmModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config

Source code in multimolecule/models/calm/configuration_calm.py

Python
class CaLmConfig(PreTrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`CaLmModel`][multimolecule.models.CaLmModel]. It
    is used to instantiate a CaLM 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 CaLM
    [oxpig/CaLM](https://github.com/oxpig/CaLM) 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 CaLM model. Defines the number of different tokens that can be represented by the
            `input_ids` passed when calling [`CaLmModel`].
            Defaults to 131 if `codon=True` else 26.
        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`.
        emb_layer_norm_before:
            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 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 CaLmConfig, CaLmModel
        >>> # Initializing a CaLM multimolecule/calm style configuration
        >>> configuration = CaLmConfig()
        >>> # Initializing a model (with random weights) from the multimolecule/calm style configuration
        >>> model = CaLmModel(configuration)
        >>> # Accessing the model configuration
        >>> configuration = model.config
    """

    model_type = "calm"

    def __init__(
        self,
        vocab_size: int | None = None,
        codon: bool = True,
        hidden_size: int = 768,
        num_hidden_layers: int = 12,
        num_attention_heads: int = 12,
        intermediate_size: int = 3072,
        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-12,
        position_embedding_type: str = "rotary",
        is_decoder: bool = False,
        use_cache: bool = True,
        emb_layer_norm_before: bool = False,
        token_dropout: bool = False,
        head: HeadConfig | None = None,
        lm_head: MaskedLMHeadConfig | None = None,
        add_cross_attention: bool = False,
        **kwargs,
    ):
        super().__init__(**kwargs)
        if vocab_size is None:
            vocab_size = 131 if codon else 26
        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.emb_layer_norm_before = emb_layer_norm_before
        self.token_dropout = token_dropout
        self.head = HeadConfig(**head) if head is not None else None
        self.lm_head = MaskedLMHeadConfig(**lm_head) if lm_head is not None else None
        self.add_cross_attention = add_cross_attention

CaLmForContactPrediction ¶

Bases: CaLmPreTrainedModel

Examples:

Python Console Session
>>> import torch
>>> from multimolecule import CaLmConfig, CaLmForContactPrediction, RnaTokenizer
>>> config = CaLmConfig()
>>> model = CaLmForContactPrediction(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/calm/modeling_calm.py

Python
class CaLmForContactPrediction(CaLmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import CaLmConfig, CaLmForContactPrediction, RnaTokenizer
        >>> config = CaLmConfig()
        >>> model = CaLmForContactPrediction(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: CaLmConfig):
        super().__init__(config)
        self.model = CaLmModel(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,
        )

CaLmForMaskedLM ¶

Bases: CaLmPreTrainedModel

Examples:

Python Console Session
>>> import torch
>>> from multimolecule import CaLmConfig, CaLmForMaskedLM, RnaTokenizer
>>> config = CaLmConfig()
>>> model = CaLmForMaskedLM(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, 131])
>>> output["loss"]
tensor(..., grad_fn=<NllLossBackward0>)

Source code in multimolecule/models/calm/modeling_calm.py

Python
class CaLmForMaskedLM(CaLmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import CaLmConfig, CaLmForMaskedLM, RnaTokenizer
        >>> config = CaLmConfig()
        >>> model = CaLmForMaskedLM(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, 131])
        >>> 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: CaLmConfig):
        super().__init__(config)
        if config.is_decoder:
            warn(
                "If you want to use `CaLmForMaskedLM` make sure `config.is_decoder=False` for "
                "bi-directional self-attention."
            )
        self.model = CaLmModel(config, add_pooling_layer=False)
        self.lm_head = MaskedLMHead(config, self.model.embeddings.word_embeddings.weight)

        # 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,
        )

CaLmForSequencePrediction ¶

Bases: CaLmPreTrainedModel

Examples:

Python Console Session
>>> import torch
>>> from multimolecule import CaLmConfig, CaLmForSequencePrediction, RnaTokenizer
>>> config = CaLmConfig()
>>> model = CaLmForSequencePrediction(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/calm/modeling_calm.py

Python
class CaLmForSequencePrediction(CaLmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import CaLmConfig, CaLmForSequencePrediction, RnaTokenizer
        >>> config = CaLmConfig()
        >>> model = CaLmForSequencePrediction(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: CaLmConfig):
        super().__init__(config)
        self.model = CaLmModel(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,
        )

CaLmForTokenPrediction ¶

Bases: CaLmPreTrainedModel

Examples:

Python Console Session
>>> import torch
>>> from multimolecule import CaLmConfig, CaLmForTokenPrediction, RnaTokenizer
>>> config = CaLmConfig()
>>> model = CaLmForTokenPrediction(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/calm/modeling_calm.py

Python
class CaLmForTokenPrediction(CaLmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import CaLmConfig, CaLmForTokenPrediction, RnaTokenizer
        >>> config = CaLmConfig()
        >>> model = CaLmForTokenPrediction(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: CaLmConfig):
        super().__init__(config)
        self.model = CaLmModel(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,
        )

CaLmModel ¶

Bases: CaLmPreTrainedModel

Examples:

Python Console Session
>>> import torch
>>> from multimolecule import CaLmConfig, CaLmModel, RnaTokenizer
>>> config = CaLmConfig()
>>> model = CaLmModel(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/calm/modeling_calm.py

Python
class CaLmModel(CaLmPreTrainedModel):
    """
    Examples:
        >>> import torch
        >>> from multimolecule import CaLmConfig, CaLmModel, RnaTokenizer
        >>> config = CaLmConfig()
        >>> model = CaLmModel(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: CaLmConfig, add_pooling_layer: bool = True):
        super().__init__(config)
        self.pad_token_id = config.pad_token_id
        self.gradient_checkpointing = False
        self.embeddings = CaLmEmbeddings(config)
        self.encoder = CaLmEncoder(config)
        self.pooler = CaLmPooler(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

    @check_model_inputs
    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.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 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,
                input_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, input_embeds=embedding_output, attention_mask=attention_mask
            )

        if encoder_attention_mask is not None:
            encoder_attention_mask = create_bidirectional_mask(
                config=self.config,
                input_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.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/calm/modeling_calm.py

Python
@check_model_inputs
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.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 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,
    )

CaLmPreTrainedModel ¶

Bases: PreTrainedModel

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

Source code in multimolecule/models/calm/modeling_calm.py

Python
class CaLmPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = CaLmConfig
    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 = ["CaLmLayer", "CaLmEmbeddings"]

    @torch.no_grad()
    def _init_weights(self, module: nn.Module):
        super()._init_weights(module)
        if isinstance(module, CaLmEmbeddings):
            init.copy_(module.position_ids, torch.arange(module.position_ids.shape[-1]).expand((1, -1)))

CaLM¶

CaLM¶

Statement¶

Disclaimer¶

Model Details¶

Model Specification¶

Links¶

Usage¶

Direct Use¶

Masked Language Modeling¶

Downstream Use¶

Extract Features¶

Sequence Classification / Regression¶

Token Classification / Regression¶

Contact Classification / Regression¶

Training Details¶

Training Data¶

Training Procedure¶

Preprocessing¶

Pre-training¶

Citation¶

Contact¶

License¶

multimolecule.models.calm ¶

CaLmConfig ¶

vocab_size ¶

codon ¶

hidden_size ¶

num_hidden_layers ¶

num_attention_heads ¶

intermediate_size ¶

hidden_act ¶

hidden_dropout ¶

attention_dropout ¶

max_position_embeddings ¶

initializer_range ¶

layer_norm_eps ¶

position_embedding_type ¶

is_decoder ¶

use_cache ¶

emb_layer_norm_before ¶

token_dropout ¶

head ¶

lm_head ¶

add_cross_attention ¶

CaLmForContactPrediction ¶

CaLmForMaskedLM ¶

CaLmForSequencePrediction ¶

CaLmForTokenPrediction ¶

CaLmModel ¶

forward ¶

encoder_hidden_states ¶

encoder_attention_mask ¶

past_key_values ¶

use_cache ¶

CaLmPreTrainedModel ¶

`vocab_size` ¶

`codon` ¶

`hidden_size` ¶

`num_hidden_layers` ¶

`num_attention_heads` ¶

`intermediate_size` ¶

`hidden_act` ¶

`hidden_dropout` ¶

`attention_dropout` ¶

`max_position_embeddings` ¶

`initializer_range` ¶

`layer_norm_eps` ¶

`position_embedding_type` ¶

`is_decoder` ¶

`use_cache` ¶

`emb_layer_norm_before` ¶

`token_dropout` ¶

`head` ¶

`lm_head` ¶

`add_cross_attention` ¶

`encoder_hidden_states` ¶

`encoder_attention_mask` ¶

`past_key_values` ¶

`use_cache` ¶