modeling_fastesm.py

import torch
import torch.nn as nn
from torch.nn import functional as F
from typing import Optional, Tuple, Union
from einops import rearrange
from transformers.modeling_outputs import (
    MaskedLMOutput,
    BaseModelOutputWithPastAndCrossAttentions,
    BaseModelOutputWithPoolingAndCrossAttentions,
    SequenceClassifierOutput,
    TokenClassifierOutput
)
from transformers.models.esm.modeling_esm import (
    RotaryEmbedding,
    EsmContactPredictionHead,
    EsmIntermediate,
    EsmOutput,
    EsmPooler,
    EsmLMHead,
    EsmSelfOutput,
    EsmClassificationHead,
    EsmPreTrainedModel,
    create_position_ids_from_input_ids,
    gelu
)


class EsmEmbeddings(nn.Module):
    """
    Same as BertEmbeddings with a tiny tweak for positional embeddings indexing.
    """

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        if config.emb_layer_norm_before:
            self.layer_norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        else:
            self.layer_norm = None
        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
        self.register_buffer(
            "position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)), persistent=False
        )

        self.padding_idx = config.pad_token_id
        self.position_embeddings = nn.Embedding(
            config.max_position_embeddings, config.hidden_size, padding_idx=self.padding_idx
        )
        # Token dropout does not work correctly so we disable it
        # self.token_dropout = config.token_dropout
        self.mask_token_id = config.mask_token_id

    def forward(
        self, input_ids=None, attention_mask=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
    ):
        if position_ids is None:
            if input_ids is not None:
                # Create the position ids from the input token ids. Any padded tokens remain padded.
                position_ids = create_position_ids_from_input_ids(input_ids, self.padding_idx, past_key_values_length)
            else:
                position_ids = self.create_position_ids_from_inputs_embeds(inputs_embeds)

        if inputs_embeds is None:
            inputs_embeds = self.word_embeddings(input_ids)

        embeddings = inputs_embeds

        if self.position_embedding_type == "absolute":
            position_embeddings = self.position_embeddings(position_ids)
            embeddings = embeddings + position_embeddings

        if self.layer_norm is not None:
            embeddings = self.layer_norm(embeddings)
        if attention_mask is not None:
            embeddings = (embeddings * attention_mask.unsqueeze(-1)).to(embeddings.dtype)
        return embeddings

    def create_position_ids_from_inputs_embeds(self, inputs_embeds):
        """
        We are provided embeddings directly. We cannot infer which are padded so just generate sequential position ids.

        Args:
            inputs_embeds: torch.Tensor

        Returns: torch.Tensor
        """
        input_shape = inputs_embeds.size()[:-1]
        sequence_length = input_shape[1]

        position_ids = torch.arange(
            self.padding_idx + 1, sequence_length + self.padding_idx + 1, dtype=torch.long, device=inputs_embeds.device
        )
        return position_ids.unsqueeze(0).expand(input_shape)


class EsmSelfAttention(nn.Module):
    def __init__(self, config, position_embedding_type=None):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
                f"heads ({config.num_attention_heads})"
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        self.key = nn.Linear(config.hidden_size, self.all_head_size)
        self.value = nn.Linear(config.hidden_size, self.all_head_size)
        self.scale = self.attention_head_size**-0.5

        self.dropout_prob = config.attention_probs_dropout_prob
        self.position_embedding_type = position_embedding_type or getattr(
            config, "position_embedding_type", "absolute"
        )
        self.rotary_embeddings = None
        if self.position_embedding_type == "rotary":
            self.rotary_embeddings = RotaryEmbedding(dim=self.attention_head_size)

    def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor:
        return rearrange(x, 'b s (h d) -> b h s d', h=self.num_attention_heads)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
    ) -> Tuple[torch.Tensor]:
        query_layer = self.transpose_for_scores(self.query(hidden_states)) * self.scale
        key_layer = self.transpose_for_scores(self.key(hidden_states))
        value_layer = self.transpose_for_scores(self.value(hidden_states))

        if self.position_embedding_type == "rotary":
            query_layer, key_layer = self.rotary_embeddings(query_layer, key_layer)

        context_layer = F.scaled_dot_product_attention(
            query_layer,
            key_layer,
            value_layer,
            attn_mask=attention_mask,
            dropout_p=self.dropout_prob,
            scale=1.0
        )
        return rearrange(context_layer, 'b h s d -> b s (h d)')
        

class EsmAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.self = EsmSelfAttention(config)
        self.output = EsmSelfOutput(config)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
    ):
        hidden_states_ln = self.LayerNorm(hidden_states)
        attention_output = self.self(
            hidden_states_ln,
            attention_mask,
        )
        return self.output(attention_output, hidden_states)


class EsmLayer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = EsmAttention(config)
        self.intermediate = EsmIntermediate(config)
        self.output = EsmOutput(config)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
    ):
        attention_output = self.attention(
            hidden_states,
            attention_mask,
        )
        layer_output = self.feed_forward_chunk(attention_output)
        return layer_output

    def feed_forward_chunk(self, attention_output):
        attention_output_ln = self.LayerNorm(attention_output)
        intermediate_output = self.intermediate(attention_output_ln)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output


class EsmEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.layer = nn.ModuleList([EsmLayer(config) for _ in range(config.num_hidden_layers)])
        self.emb_layer_norm_after = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.gradient_checkpointing = False

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        output_hidden_states=False,
    ):
        all_hidden_states = () if output_hidden_states else None
        for layer_module in self.layer:
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            if self.gradient_checkpointing and self.training:
                hidden_states = self._gradient_checkpointing_func(
                    layer_module.__call__,
                    hidden_states,
                    attention_mask,
                )
            else:
                hidden_states = layer_module(
                    hidden_states,
                    attention_mask,
                )

        if self.emb_layer_norm_after:
            hidden_states = self.emb_layer_norm_after(hidden_states)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            hidden_states=all_hidden_states,
        )


class FastEsmModel(EsmPreTrainedModel):
    def __init__(self, config, add_pooling_layer=True):
        super().__init__(config)
        self.config = config
        self.embeddings = EsmEmbeddings(config)
        self.encoder = EsmEncoder(config)
        self.pooler = EsmPooler(config) if add_pooling_layer else None
        self.contact_head = EsmContactPredictionHead(
            in_features=config.num_hidden_layers * config.num_attention_heads, bias=True
        )
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embeddings.word_embeddings

    def set_input_embeddings(self, value):
        self.embeddings.word_embeddings = value

    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        output_hidden_states: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
    ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
        if output_attentions is not None:
            raise ValueError("output_attentions is not supported by F.scaled_dot_product_attention")
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
            input_shape = input_ids.size()
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        batch_size, seq_length = input_shape
        embedding_output = self.embeddings(
            input_ids=input_ids,
            position_ids=position_ids,
            attention_mask=attention_mask,
            inputs_embeds=inputs_embeds,
        )
        # Prepare attention mask
        if attention_mask is not None:
            # attention_mask shape should be (batch_size, 1, 1, seq_length)
            # Expand to (batch_size, 1, seq_length, seq_length)
            extended_attention_mask = attention_mask[:, None, None, :].expand(
                batch_size, 1, seq_length, seq_length
            )
            # Convert mask to float with 0.0 for positions to keep and -inf for masked positions
            attention_mask = attention_mask.to(dtype=embedding_output.dtype)  # fp16 compatibility
            attention_mask = (1.0 - attention_mask) * torch.finfo(embedding_output.dtype).min
        else:
            extended_attention_mask = None

        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask=extended_attention_mask,
            output_hidden_states=output_hidden_states,
        )
        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,
            hidden_states=encoder_outputs.hidden_states,
        )


class FastEsmForMaskedLM(EsmPreTrainedModel):
    _tied_weights_keys = ["lm_head.decoder.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.esm = FastEsmModel(config, add_pooling_layer=False)
        self.lm_head = EsmLMHead(config)
        self.loss_fct = nn.CrossEntropyLoss()
        self.init_weights()

    def get_output_embeddings(self):
        return self.lm_head.decoder

    def set_output_embeddings(self, new_embeddings):
        self.lm_head.decoder = new_embeddings

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
    ) -> Union[Tuple, MaskedLMOutput]:
        outputs = self.esm(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_hidden_states=output_hidden_states,
            output_attentions=output_attentions,
        )
        sequence_output = outputs.last_hidden_state
        prediction_scores = self.lm_head(sequence_output)

        loss = None
        if labels is not None:
            labels = labels.to(prediction_scores.device)
            loss = self.loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))

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

    def predict_contacts(self, tokens, attention_mask):
        raise NotImplementedError("predict_contacts is not supported by F.scaled_dot_product_attention")


class FastEsmForSequenceClassification(EsmPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.config = config
        self.esm = FastEsmModel(config, add_pooling_layer=False)
        self.classifier = EsmClassificationHead(config)
        self.mse = nn.MSELoss()
        self.ce = nn.CrossEntropyLoss()
        self.bce = nn.BCEWithLogitsLoss()
        self.init_weights()

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
    ) -> Union[Tuple, SequenceClassifierOutput]:
        outputs = self.esm(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )
        sequence_output = outputs.last_hidden_state
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            labels = labels.to(logits.device)
            if self.config.problem_type is None:
                if self.num_labels == 1:
                    self.config.problem_type = "regression"
                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
                    self.config.problem_type = "single_label_classification"
                else:
                    self.config.problem_type = "multi_label_classification"

            if self.config.problem_type == "regression":
                if self.num_labels == 1:
                    loss = self.mse(logits.squeeze(), labels.squeeze())
                else:
                    loss = self.mse(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss = self.ce(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss = self.bce(logits, labels)

        return SequenceClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
        )


class FastEsmForTokenClassification(EsmPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.esm = FastEsmModel(config, add_pooling_layer=False)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
        self.loss_fct = nn.CrossEntropyLoss()
        self.init_weights()

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
    ) -> Union[Tuple, TokenClassifierOutput]:
        outputs = self.esm(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )
        sequence_output = outputs.last_hidden_state
        sequence_output = self.dropout(sequence_output)
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            labels = labels.to(logits.device)
            loss = self.loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

        return TokenClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
        )


if __name__ == "__main__":
    """
    Test the hidden state differences between the FastEsmModel and the HF EsmModel.
    In full precision, the differences are very small, but nonzero due to floating point issues with F.scaled_dot_product_attention.
    In Pytorch 2.5+ (and linux kernel), this implementation is very fast and uses less memory than the HF implementation.
    """
    import random
    from transformers import EsmModel as TransformersEsmModel, EsmTokenizer

    model_paths = [
        "facebook/esm2_t6_8M_UR50D",
        "facebook/esm2_t12_35M_UR50D",
        "facebook/esm2_t30_150M_UR50D",
        "facebook/esm2_t33_650M_UR50D",
    ]
    canonical_amino_acids = "ACDEFGHIKLMNPQRSTVWY"
    length = 64
    seq_count = 100
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    tolerances = [1e-3, 1e-4, 1e-5, 1e-6, 1e-7, 1e-8]

    def generate_random_sequence(length: int) -> str:
        return 'M' + "".join(random.choices(canonical_amino_acids, k=length))

    print("Percentage of hidden states that are within the tolerance:")
    for model_path in model_paths:
        print(f"Testing {model_path}...")
        tokenizer = EsmTokenizer.from_pretrained(model_path)
        fast_model = FastEsmModel.from_pretrained(model_path, token_dropout=False).to(device)
        model = TransformersEsmModel.from_pretrained(model_path, token_dropout=False).to(device)

        counts = [0] * len(tolerances)
        for _ in range(seq_count):
            example_seq = generate_random_sequence(length)
            fast_tokens = tokenizer(example_seq, return_tensors="pt").input_ids.to(device)
            fast_output = fast_model(fast_tokens).last_hidden_state.detach().cpu()

            model_tokens = tokenizer(example_seq, return_tensors="pt").input_ids.to(device)
            model_output = model(model_tokens).last_hidden_state.detach().cpu()

            for i, atol in enumerate(tolerances):
                if torch.allclose(fast_output, model_output, atol=atol):
                    counts[i] += 1

        print(f"{model_path}:")
        for i, atol in enumerate(tolerances):
            print(f"    tolerance={atol}: {counts[i] / seq_count * 100}%")
    
        model.cpu()
        fast_model.cpu()
        del model
        del fast_model
        torch.cuda.empty_cache()