|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
""" PyTorch ESM model.""" |
|
|
|
import math |
|
from dataclasses import dataclass |
|
from typing import List, Optional, Tuple, Union |
|
|
|
import torch |
|
import torch.utils.checkpoint |
|
from torch import nn |
|
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss, SiLU |
|
from transformers.file_utils import ( |
|
add_code_sample_docstrings, |
|
add_start_docstrings, |
|
add_start_docstrings_to_model_forward, |
|
) |
|
from transformers.modeling_outputs import ( |
|
BaseModelOutputWithPastAndCrossAttentions, |
|
BaseModelOutputWithPoolingAndCrossAttentions, |
|
MaskedLMOutput, |
|
SequenceClassifierOutput, |
|
TokenClassifierOutput, |
|
) |
|
from transformers.modeling_utils import ( |
|
PreTrainedModel, |
|
find_pruneable_heads_and_indices, |
|
prune_linear_layer, |
|
) |
|
from transformers.utils import logging |
|
|
|
from .segment_nt_config import SegmentNTConfig |
|
|
|
logger = logging.get_logger(__name__) |
|
|
|
_CHECKPOINT_FOR_DOC = "facebook/esm2_t6_8M_UR50D" |
|
_CONFIG_FOR_DOC = "SegmentNTConfig" |
|
|
|
ESM_PRETRAINED_MODEL_ARCHIVE_LIST = [ |
|
"facebook/esm2_t6_8M_UR50D", |
|
"facebook/esm2_t12_35M_UR50D", |
|
|
|
|
|
] |
|
|
|
|
|
def rotate_half(x): |
|
x1, x2 = x.chunk(2, dim=-1) |
|
return torch.cat((-x2, x1), dim=-1) |
|
|
|
|
|
def apply_rotary_pos_emb(x, cos, sin): |
|
cos = cos[:, :, : x.shape[-2], :] |
|
sin = sin[:, :, : x.shape[-2], :] |
|
|
|
return (x * cos) + (rotate_half(x) * sin) |
|
|
|
|
|
def gelu(x): |
|
""" |
|
This is the gelu implementation from the original ESM repo. Using F.gelu yields subtly wrong results. |
|
""" |
|
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0))) |
|
|
|
|
|
def symmetrize(x): |
|
"Make layer symmetric in final two dimensions, used for contact prediction." |
|
return x + x.transpose(-1, -2) |
|
|
|
|
|
def average_product_correct(x): |
|
"Perform average product correct, used for contact prediction." |
|
a1 = x.sum(-1, keepdims=True) |
|
a2 = x.sum(-2, keepdims=True) |
|
a12 = x.sum((-1, -2), keepdims=True) |
|
|
|
avg = a1 * a2 |
|
avg.div_(a12) |
|
normalized = x - avg |
|
return normalized |
|
|
|
@dataclass |
|
class RotaryEmbeddingConfig: |
|
""" |
|
Parameters to initialize the RotaryEmbedding layer. The rescaling factor allows |
|
to adapt the rotary embeddings to larger lengths than what was used for training. |
|
One of this strategy is presented in the Yarn paper: https://arxiv.org/pdf/2309.00071.pdf. # noqa |
|
|
|
Args: |
|
|
|
""" |
|
|
|
rescaling_factor: Optional[float] |
|
|
|
class RotaryEmbedding(torch.nn.Module): |
|
""" |
|
Rotary position embeddings based on those in |
|
[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer). Query and keys are transformed by rotation |
|
matrices which depend on their relative positions. |
|
""" |
|
|
|
def __init__(self, dim: int, rotary_embedding_config: RotaryEmbeddingConfig): |
|
super().__init__() |
|
|
|
|
|
rescaling_factor = rotary_embedding_config.rescaling_factor |
|
upper_freq = 10000 |
|
|
|
if rescaling_factor is None: |
|
inv_freq = 1.0 / (upper_freq ** (torch.arange(0, dim, 2).float() / dim)) |
|
else: |
|
updated_base = upper_freq * ( |
|
rescaling_factor ** (dim / (dim - 2)) |
|
) |
|
inv_freq = 1.0 / ( |
|
updated_base ** (torch.arange(0, dim, 2).float() / dim) |
|
) |
|
|
|
self.register_buffer("inv_freq", inv_freq) |
|
|
|
self._seq_len_cached = None |
|
self._cos_cached = None |
|
self._sin_cached = None |
|
|
|
def _update_cos_sin_tables(self, x, seq_dimension=2): |
|
seq_len = x.shape[seq_dimension] |
|
|
|
|
|
|
|
if seq_len != self._seq_len_cached or self._cos_cached.device != x.device: |
|
self._seq_len_cached = seq_len |
|
t = torch.arange(x.shape[seq_dimension], device=x.device).type_as( |
|
self.inv_freq |
|
) |
|
freqs = torch.outer(t, self.inv_freq) |
|
emb = torch.cat((freqs, freqs), dim=-1).to(x.device) |
|
|
|
self._cos_cached = emb.cos()[None, None, :, :] |
|
self._sin_cached = emb.sin()[None, None, :, :] |
|
|
|
return self._cos_cached, self._sin_cached |
|
|
|
def forward( |
|
self, q: torch.Tensor, k: torch.Tensor |
|
) -> Tuple[torch.Tensor, torch.Tensor]: |
|
self._cos_cached, self._sin_cached = self._update_cos_sin_tables( |
|
k, seq_dimension=-2 |
|
) |
|
|
|
return ( |
|
apply_rotary_pos_emb(q, self._cos_cached, self._sin_cached), |
|
apply_rotary_pos_emb(k, self._cos_cached, self._sin_cached), |
|
) |
|
|
|
|
|
class EsmContactPredictionHead(nn.Module): |
|
"""Performs symmetrization, apc, and computes a logistic regression on the output features""" |
|
|
|
def __init__( |
|
self, |
|
in_features: int, |
|
bias=True, |
|
eos_idx: int = 2, |
|
): |
|
super().__init__() |
|
self.in_features = in_features |
|
self.eos_idx = eos_idx |
|
self.regression = nn.Linear(in_features, 1, bias) |
|
self.activation = nn.Sigmoid() |
|
|
|
def forward(self, tokens, attentions): |
|
|
|
eos_mask = tokens.ne(self.eos_idx).to(attentions) |
|
eos_mask = eos_mask.unsqueeze(1) * eos_mask.unsqueeze(2) |
|
attentions = attentions * eos_mask[:, None, None, :, :] |
|
attentions = attentions[..., :-1, :-1] |
|
|
|
attentions = attentions[..., 1:, 1:] |
|
batch_size, layers, heads, seqlen, _ = attentions.size() |
|
attentions = attentions.view(batch_size, layers * heads, seqlen, seqlen) |
|
|
|
|
|
attentions = attentions.to( |
|
self.regression.weight.device |
|
) |
|
attentions = average_product_correct(symmetrize(attentions)) |
|
attentions = attentions.permute(0, 2, 3, 1) |
|
return self.activation(self.regression(attentions).squeeze(3)) |
|
|
|
|
|
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.dropout = nn.Dropout(config.hidden_dropout_prob) |
|
|
|
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, |
|
) |
|
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: |
|
|
|
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.token_dropout: |
|
embeddings.masked_fill_( |
|
(input_ids == self.mask_token_id).unsqueeze(-1), 0.0 |
|
) |
|
mask_ratio_train = ( |
|
0.15 * 0.8 |
|
) |
|
src_lengths = attention_mask.sum(-1) |
|
mask_ratio_observed = (input_ids == self.mask_token_id).sum( |
|
-1 |
|
).float() / src_lengths |
|
embeddings = ( |
|
embeddings |
|
* (1 - mask_ratio_train) |
|
/ (1 - mask_ratio_observed)[:, None, None] |
|
).to(embeddings.dtype) |
|
|
|
if self.position_embedding_type == "absolute": |
|
position_embeddings = self.position_embeddings(position_ids) |
|
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.dropout = nn.Dropout(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 == "relative_key" |
|
or self.position_embedding_type == "relative_key_query" |
|
): |
|
self.max_position_embeddings = config.max_position_embeddings |
|
self.distance_embedding = nn.Embedding( |
|
2 * config.max_position_embeddings - 1, self.attention_head_size |
|
) |
|
elif self.position_embedding_type == "rotary": |
|
|
|
rescaling_factor = config.rescaling_factor |
|
rotary_embedding_config = RotaryEmbeddingConfig(rescaling_factor=rescaling_factor) |
|
|
|
self.rotary_embeddings = RotaryEmbedding(dim=self.attention_head_size, rotary_embedding_config=rotary_embedding_config) |
|
|
|
self.is_decoder = config.is_decoder |
|
|
|
def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor: |
|
new_x_shape = x.size()[:-1] + ( |
|
self.num_attention_heads, |
|
self.attention_head_size, |
|
) |
|
x = x.view(new_x_shape) |
|
return x.permute(0, 2, 1, 3) |
|
|
|
def forward( |
|
self, |
|
hidden_states: torch.Tensor, |
|
attention_mask: Optional[torch.FloatTensor] = None, |
|
head_mask: Optional[torch.FloatTensor] = None, |
|
encoder_hidden_states: Optional[torch.FloatTensor] = None, |
|
encoder_attention_mask: Optional[torch.FloatTensor] = None, |
|
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None, |
|
output_attentions: Optional[bool] = False, |
|
) -> Tuple[torch.Tensor]: |
|
mixed_query_layer = self.query(hidden_states) |
|
|
|
|
|
|
|
|
|
is_cross_attention = encoder_hidden_states is not None |
|
|
|
if is_cross_attention and past_key_value is not None: |
|
|
|
key_layer = past_key_value[0] |
|
value_layer = past_key_value[1] |
|
attention_mask = encoder_attention_mask |
|
elif is_cross_attention: |
|
key_layer = self.transpose_for_scores(self.key(encoder_hidden_states)) |
|
value_layer = self.transpose_for_scores(self.value(encoder_hidden_states)) |
|
attention_mask = encoder_attention_mask |
|
elif past_key_value is not None: |
|
key_layer = self.transpose_for_scores(self.key(hidden_states)) |
|
value_layer = self.transpose_for_scores(self.value(hidden_states)) |
|
key_layer = torch.cat([past_key_value[0], key_layer], dim=2) |
|
value_layer = torch.cat([past_key_value[1], value_layer], dim=2) |
|
else: |
|
key_layer = self.transpose_for_scores(self.key(hidden_states)) |
|
value_layer = self.transpose_for_scores(self.value(hidden_states)) |
|
|
|
query_layer = self.transpose_for_scores(mixed_query_layer) |
|
|
|
|
|
|
|
|
|
|
|
query_layer = query_layer * self.attention_head_size**-0.5 |
|
|
|
if self.is_decoder: |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
past_key_value = (key_layer, value_layer) |
|
|
|
if self.position_embedding_type == "rotary": |
|
query_layer, key_layer = self.rotary_embeddings(query_layer, key_layer) |
|
|
|
|
|
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2)) |
|
|
|
if ( |
|
self.position_embedding_type == "relative_key" |
|
or self.position_embedding_type == "relative_key_query" |
|
): |
|
seq_length = hidden_states.size()[1] |
|
position_ids_l = torch.arange( |
|
seq_length, dtype=torch.long, device=hidden_states.device |
|
).view(-1, 1) |
|
position_ids_r = torch.arange( |
|
seq_length, dtype=torch.long, device=hidden_states.device |
|
).view(1, -1) |
|
distance = position_ids_l - position_ids_r |
|
positional_embedding = self.distance_embedding( |
|
distance + self.max_position_embeddings - 1 |
|
) |
|
positional_embedding = positional_embedding.to( |
|
dtype=query_layer.dtype |
|
) |
|
|
|
if self.position_embedding_type == "relative_key": |
|
relative_position_scores = torch.einsum( |
|
"bhld,lrd->bhlr", query_layer, positional_embedding |
|
) |
|
attention_scores = attention_scores + relative_position_scores |
|
elif self.position_embedding_type == "relative_key_query": |
|
relative_position_scores_query = torch.einsum( |
|
"bhld,lrd->bhlr", query_layer, positional_embedding |
|
) |
|
relative_position_scores_key = torch.einsum( |
|
"bhrd,lrd->bhlr", key_layer, positional_embedding |
|
) |
|
attention_scores = ( |
|
attention_scores |
|
+ relative_position_scores_query |
|
+ relative_position_scores_key |
|
) |
|
|
|
if attention_mask is not None: |
|
|
|
attention_scores = attention_scores + attention_mask |
|
|
|
|
|
attention_probs = nn.functional.softmax(attention_scores, dim=-1) |
|
|
|
|
|
|
|
attention_probs = self.dropout(attention_probs) |
|
|
|
|
|
if head_mask is not None: |
|
attention_probs = attention_probs * head_mask |
|
|
|
context_layer = torch.matmul(attention_probs, value_layer) |
|
|
|
context_layer = context_layer.permute(0, 2, 1, 3).contiguous() |
|
new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,) |
|
context_layer = context_layer.view(new_context_layer_shape) |
|
|
|
outputs = ( |
|
(context_layer, attention_probs) if output_attentions else (context_layer,) |
|
) |
|
|
|
if self.is_decoder: |
|
outputs = outputs + (past_key_value,) |
|
return outputs |
|
|
|
|
|
class EsmSelfOutput(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
self.dense = nn.Linear(config.hidden_size, config.hidden_size) |
|
self.dropout = nn.Dropout(config.hidden_dropout_prob) |
|
|
|
def forward(self, hidden_states, input_tensor): |
|
hidden_states = self.dense(hidden_states) |
|
hidden_states = self.dropout(hidden_states) |
|
hidden_states += input_tensor |
|
return hidden_states |
|
|
|
|
|
class EsmAttention(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
self.self = EsmSelfAttention(config) |
|
self.output = EsmSelfOutput(config) |
|
self.pruned_heads = set() |
|
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) |
|
|
|
def prune_heads(self, heads): |
|
if len(heads) == 0: |
|
return |
|
heads, index = find_pruneable_heads_and_indices( |
|
heads, |
|
self.self.num_attention_heads, |
|
self.self.attention_head_size, |
|
self.pruned_heads, |
|
) |
|
|
|
|
|
self.self.query = prune_linear_layer(self.self.query, index) |
|
self.self.key = prune_linear_layer(self.self.key, index) |
|
self.self.value = prune_linear_layer(self.self.value, index) |
|
self.output.dense = prune_linear_layer(self.output.dense, index, dim=1) |
|
|
|
|
|
self.self.num_attention_heads = self.self.num_attention_heads - len(heads) |
|
self.self.all_head_size = ( |
|
self.self.attention_head_size * self.self.num_attention_heads |
|
) |
|
self.pruned_heads = self.pruned_heads.union(heads) |
|
|
|
def forward( |
|
self, |
|
hidden_states, |
|
attention_mask=None, |
|
head_mask=None, |
|
encoder_hidden_states=None, |
|
encoder_attention_mask=None, |
|
past_key_value=None, |
|
output_attentions=False, |
|
): |
|
hidden_states_ln = self.LayerNorm(hidden_states) |
|
self_outputs = self.self( |
|
hidden_states_ln, |
|
attention_mask, |
|
head_mask, |
|
encoder_hidden_states, |
|
encoder_attention_mask, |
|
past_key_value, |
|
output_attentions, |
|
) |
|
attention_output = self.output(self_outputs[0], hidden_states) |
|
outputs = (attention_output,) + self_outputs[ |
|
1: |
|
] |
|
return outputs |
|
|
|
|
|
class EsmIntermediate(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
|
|
self.dense = nn.Linear( |
|
config.hidden_size, |
|
int(config.intermediate_size * 2), |
|
bias=config.add_bias_fnn, |
|
) |
|
self.activation_fn = SiLU() |
|
|
|
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: |
|
hidden_states = self.dense(hidden_states) |
|
|
|
|
|
x1, x2 = hidden_states.split(int(hidden_states.size(-1) / 2), -1) |
|
hidden_states = self.activation_fn(x1) * x2 |
|
|
|
return hidden_states |
|
|
|
|
|
class EsmOutput(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
self.dense = nn.Linear( |
|
config.intermediate_size, config.hidden_size, bias=config.add_bias_fnn |
|
) |
|
self.dropout = nn.Dropout(config.hidden_dropout_prob) |
|
|
|
def forward(self, hidden_states, input_tensor): |
|
hidden_states = self.dense(hidden_states) |
|
hidden_states = self.dropout(hidden_states) |
|
hidden_states += input_tensor |
|
return 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.is_decoder = config.is_decoder |
|
self.add_cross_attention = config.add_cross_attention |
|
if self.add_cross_attention: |
|
if not self.is_decoder: |
|
raise RuntimeError( |
|
f"{self} should be used as a decoder model if cross attention is added" |
|
) |
|
self.crossattention = 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, |
|
head_mask=None, |
|
encoder_hidden_states=None, |
|
encoder_attention_mask=None, |
|
past_key_value=None, |
|
output_attentions=False, |
|
): |
|
|
|
self_attn_past_key_value = ( |
|
past_key_value[:2] if past_key_value is not None else None |
|
) |
|
self_attention_outputs = self.attention( |
|
hidden_states, |
|
attention_mask, |
|
head_mask, |
|
output_attentions=output_attentions, |
|
past_key_value=self_attn_past_key_value, |
|
) |
|
attention_output = self_attention_outputs[0] |
|
|
|
|
|
if self.is_decoder: |
|
outputs = self_attention_outputs[1:-1] |
|
present_key_value = self_attention_outputs[-1] |
|
else: |
|
outputs = self_attention_outputs[ |
|
1: |
|
] |
|
|
|
cross_attn_present_key_value = None |
|
if self.is_decoder and encoder_hidden_states is not None: |
|
if not hasattr(self, "crossattention"): |
|
raise AttributeError( |
|
f"If `encoder_hidden_states` are passed, {self} has to be instantiated" |
|
" with cross-attention layers by setting `config.add_cross_attention=True`" |
|
) |
|
|
|
|
|
cross_attn_past_key_value = ( |
|
past_key_value[-2:] if past_key_value is not None else None |
|
) |
|
cross_attention_outputs = self.crossattention( |
|
attention_output, |
|
attention_mask, |
|
head_mask, |
|
encoder_hidden_states, |
|
encoder_attention_mask, |
|
cross_attn_past_key_value, |
|
output_attentions, |
|
) |
|
attention_output = cross_attention_outputs[0] |
|
outputs = ( |
|
outputs + cross_attention_outputs[1:-1] |
|
) |
|
|
|
|
|
cross_attn_present_key_value = cross_attention_outputs[-1] |
|
present_key_value = present_key_value + cross_attn_present_key_value |
|
|
|
layer_output = self.feed_forward_chunk(attention_output) |
|
|
|
outputs = (layer_output,) + outputs |
|
|
|
|
|
if self.is_decoder: |
|
outputs = outputs + (present_key_value,) |
|
return outputs |
|
|
|
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, |
|
head_mask=None, |
|
encoder_hidden_states=None, |
|
encoder_attention_mask=None, |
|
past_key_values=None, |
|
use_cache=None, |
|
output_attentions=False, |
|
output_hidden_states=False, |
|
return_dict=True, |
|
): |
|
if self.gradient_checkpointing and self.training: |
|
if use_cache: |
|
logger.warning_once( |
|
"`use_cache=True` is incompatible with `config.gradient_checkpointing=True`. Setting " |
|
"`use_cache=False`..." |
|
) |
|
use_cache = False |
|
all_hidden_states = () if output_hidden_states else None |
|
all_self_attentions = () if output_attentions else None |
|
all_cross_attentions = ( |
|
() if output_attentions and self.config.add_cross_attention else None |
|
) |
|
|
|
next_decoder_cache = () if use_cache else None |
|
for i, layer_module in enumerate(self.layer): |
|
if output_hidden_states: |
|
all_hidden_states = all_hidden_states + (hidden_states,) |
|
|
|
layer_head_mask = head_mask[i] if head_mask is not None else None |
|
past_key_value = past_key_values[i] if past_key_values is not None else None |
|
|
|
if self.gradient_checkpointing and self.training: |
|
|
|
def create_custom_forward(module): |
|
def custom_forward(*inputs): |
|
return module(*inputs, past_key_value, output_attentions) |
|
|
|
return custom_forward |
|
|
|
layer_outputs = torch.utils.checkpoint.checkpoint( |
|
create_custom_forward(layer_module), |
|
hidden_states, |
|
attention_mask, |
|
layer_head_mask, |
|
encoder_hidden_states, |
|
encoder_attention_mask, |
|
) |
|
else: |
|
layer_outputs = layer_module( |
|
hidden_states, |
|
attention_mask, |
|
layer_head_mask, |
|
encoder_hidden_states, |
|
encoder_attention_mask, |
|
past_key_value, |
|
output_attentions, |
|
) |
|
|
|
hidden_states = layer_outputs[0] |
|
if use_cache: |
|
next_decoder_cache += (layer_outputs[-1],) |
|
if output_attentions: |
|
all_self_attentions = all_self_attentions + (layer_outputs[1],) |
|
if self.config.add_cross_attention: |
|
all_cross_attentions = all_cross_attentions + (layer_outputs[2],) |
|
|
|
|
|
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,) |
|
|
|
if not return_dict: |
|
return tuple( |
|
v |
|
for v in [ |
|
hidden_states, |
|
next_decoder_cache, |
|
all_hidden_states, |
|
all_self_attentions, |
|
all_cross_attentions, |
|
] |
|
if v is not None |
|
) |
|
return BaseModelOutputWithPastAndCrossAttentions( |
|
last_hidden_state=hidden_states, |
|
past_key_values=next_decoder_cache, |
|
hidden_states=all_hidden_states, |
|
attentions=all_self_attentions, |
|
cross_attentions=all_cross_attentions, |
|
) |
|
|
|
|
|
|
|
class EsmPooler(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
self.dense = nn.Linear(config.hidden_size, config.hidden_size) |
|
self.activation = nn.Tanh() |
|
|
|
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: |
|
|
|
|
|
first_token_tensor = hidden_states[:, 0] |
|
pooled_output = self.dense(first_token_tensor) |
|
pooled_output = self.activation(pooled_output) |
|
return pooled_output |
|
|
|
|
|
class EsmPreTrainedModel(PreTrainedModel): |
|
""" |
|
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained |
|
models. |
|
""" |
|
|
|
config_class = SegmentNTConfig |
|
base_model_prefix = "esm" |
|
_no_split_modules = ["EsmLayer", "EsmFoldTriangularSelfAttentionBlock"] |
|
|
|
|
|
def _init_weights(self, module): |
|
"""Initialize the weights""" |
|
if isinstance(module, nn.Linear): |
|
|
|
|
|
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) |
|
if module.bias is not None: |
|
module.bias.data.zero_() |
|
elif isinstance(module, nn.Embedding): |
|
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) |
|
if module.padding_idx is not None: |
|
module.weight.data[module.padding_idx].zero_() |
|
elif isinstance(module, nn.LayerNorm): |
|
module.bias.data.zero_() |
|
module.weight.data.fill_(1.0) |
|
|
|
|
|
ESM_START_DOCSTRING = r""" |
|
|
|
This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the |
|
library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads |
|
etc.) |
|
|
|
This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. |
|
Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage |
|
and behavior. |
|
|
|
Parameters: |
|
config ([`EsmConfig`]): Model configuration class with all the parameters of the |
|
model. Initializing with a config file does not load the weights associated with the model, only the |
|
configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights. |
|
""" |
|
|
|
ESM_INPUTS_DOCSTRING = r""" |
|
Args: |
|
input_ids (`torch.LongTensor` of shape `({0})`): |
|
Indices of input sequence tokens in the vocabulary. |
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and |
|
[`PreTrainedTokenizer.__call__`] for details. |
|
|
|
[What are input IDs?](../glossary#input-ids) |
|
attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*): |
|
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: |
|
|
|
- 1 for tokens that are **not masked**, |
|
- 0 for tokens that are **masked**. |
|
|
|
[What are attention masks?](../glossary#attention-mask) |
|
position_ids (`torch.LongTensor` of shape `({0})`, *optional*): |
|
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, |
|
config.max_position_embeddings - 1]`. |
|
|
|
[What are position IDs?](../glossary#position-ids) |
|
head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*): |
|
Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`: |
|
|
|
- 1 indicates the head is **not masked**, |
|
- 0 indicates the head is **masked**. |
|
|
|
inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*): |
|
Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This |
|
is useful if you want more control over how to convert `input_ids` indices into associated vectors than the |
|
model's internal embedding lookup matrix. |
|
output_attentions (`bool`, *optional*): |
|
Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned |
|
tensors for more detail. |
|
output_hidden_states (`bool`, *optional*): |
|
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for |
|
more detail. |
|
return_dict (`bool`, *optional*): |
|
Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple. |
|
""" |
|
|
|
|
|
@add_start_docstrings( |
|
"The bare ESM Model transformer outputting raw hidden-states without any specific head on top.", |
|
ESM_START_DOCSTRING, |
|
) |
|
class EsmModel(EsmPreTrainedModel): |
|
""" |
|
|
|
The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of |
|
cross-attention is added between the self-attention layers, following the architecture described in [Attention is |
|
all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, |
|
Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin. |
|
|
|
To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set |
|
to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder` argument and |
|
`add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass. |
|
""" |
|
|
|
supports_gradient_checkpointing = False |
|
|
|
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 |
|
) |
|
|
|
|
|
self.post_init() |
|
|
|
def _set_gradient_checkpointing(self, module, value=False): |
|
if isinstance(module, EsmEncoder): |
|
module.gradient_checkpointing = value |
|
|
|
def get_input_embeddings(self): |
|
return self.embeddings.word_embeddings |
|
|
|
def set_input_embeddings(self, value): |
|
self.embeddings.word_embeddings = value |
|
|
|
def _prune_heads(self, heads_to_prune): |
|
""" |
|
Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base |
|
class PreTrainedModel |
|
""" |
|
for layer, heads in heads_to_prune.items(): |
|
self.encoder.layer[layer].attention.prune_heads(heads) |
|
|
|
@add_start_docstrings_to_model_forward( |
|
ESM_INPUTS_DOCSTRING.format("(batch_size, sequence_length)") |
|
) |
|
@add_code_sample_docstrings( |
|
checkpoint=_CHECKPOINT_FOR_DOC, |
|
output_type=BaseModelOutputWithPoolingAndCrossAttentions, |
|
config_class=_CONFIG_FOR_DOC, |
|
) |
|
def forward( |
|
self, |
|
input_ids: Optional[torch.Tensor] = None, |
|
attention_mask: Optional[torch.Tensor] = None, |
|
position_ids: Optional[torch.Tensor] = None, |
|
head_mask: Optional[torch.Tensor] = None, |
|
inputs_embeds: Optional[torch.Tensor] = None, |
|
encoder_hidden_states: Optional[torch.Tensor] = None, |
|
encoder_attention_mask: Optional[torch.Tensor] = None, |
|
past_key_values: Optional[List[torch.FloatTensor]] = None, |
|
use_cache: Optional[bool] = None, |
|
output_attentions: Optional[bool] = None, |
|
output_hidden_states: Optional[bool] = None, |
|
return_dict: Optional[bool] = None, |
|
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]: |
|
r""" |
|
encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): |
|
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 (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): |
|
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(tuple(torch.FloatTensor))` 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 (`bool`, *optional*): |
|
If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see |
|
`past_key_values`). |
|
""" |
|
output_attentions = ( |
|
output_attentions |
|
if output_attentions is not None |
|
else self.config.output_attentions |
|
) |
|
output_hidden_states = ( |
|
output_hidden_states |
|
if output_hidden_states is not None |
|
else self.config.output_hidden_states |
|
) |
|
return_dict = ( |
|
return_dict if return_dict is not None else self.config.use_return_dict |
|
) |
|
|
|
if self.config.is_decoder: |
|
use_cache = use_cache if use_cache is not None else self.config.use_cache |
|
else: |
|
use_cache = False |
|
|
|
if 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: |
|
input_shape = input_ids.size() |
|
elif inputs_embeds is not None: |
|
input_shape = inputs_embeds.size()[:-1] |
|
else: |
|
raise ValueError("You have to specify either input_ids or inputs_embeds") |
|
|
|
batch_size, seq_length = input_shape |
|
device = input_ids.device if input_ids is not None else inputs_embeds.device |
|
|
|
|
|
past_key_values_length = ( |
|
past_key_values[0][0].shape[2] if past_key_values is not None else 0 |
|
) |
|
|
|
if attention_mask is None: |
|
attention_mask = torch.ones( |
|
((batch_size, seq_length + past_key_values_length)), device=device |
|
) |
|
|
|
|
|
|
|
extended_attention_mask: torch.Tensor = self.get_extended_attention_mask( |
|
attention_mask, input_shape |
|
) |
|
|
|
|
|
|
|
if self.config.is_decoder and encoder_hidden_states is not None: |
|
( |
|
encoder_batch_size, |
|
encoder_sequence_length, |
|
_, |
|
) = encoder_hidden_states.size() |
|
encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length) |
|
if encoder_attention_mask is None: |
|
encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device) |
|
encoder_extended_attention_mask = self.invert_attention_mask( |
|
encoder_attention_mask |
|
) |
|
else: |
|
encoder_extended_attention_mask = None |
|
|
|
|
|
|
|
|
|
|
|
|
|
head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers) |
|
|
|
embedding_output = self.embeddings( |
|
input_ids=input_ids, |
|
position_ids=position_ids, |
|
attention_mask=attention_mask, |
|
inputs_embeds=inputs_embeds, |
|
past_key_values_length=past_key_values_length, |
|
) |
|
encoder_outputs = self.encoder( |
|
embedding_output, |
|
attention_mask=extended_attention_mask, |
|
head_mask=head_mask, |
|
encoder_hidden_states=encoder_hidden_states, |
|
encoder_attention_mask=encoder_extended_attention_mask, |
|
past_key_values=past_key_values, |
|
use_cache=use_cache, |
|
output_attentions=output_attentions, |
|
output_hidden_states=output_hidden_states, |
|
return_dict=return_dict, |
|
) |
|
sequence_output = encoder_outputs[0] |
|
pooled_output = ( |
|
self.pooler(sequence_output) if self.pooler is not None else None |
|
) |
|
|
|
if not return_dict: |
|
return (sequence_output, pooled_output) + encoder_outputs[1:] |
|
|
|
return BaseModelOutputWithPoolingAndCrossAttentions( |
|
last_hidden_state=sequence_output, |
|
pooler_output=pooled_output, |
|
past_key_values=encoder_outputs.past_key_values, |
|
hidden_states=encoder_outputs.hidden_states, |
|
attentions=encoder_outputs.attentions, |
|
cross_attentions=encoder_outputs.cross_attentions, |
|
) |
|
|
|
def predict_contacts(self, tokens, attention_mask): |
|
attns = self( |
|
tokens, |
|
attention_mask=attention_mask, |
|
return_dict=True, |
|
output_attentions=True, |
|
).attentions |
|
attns = torch.stack(attns, dim=1) |
|
|
|
|
|
|
|
|
|
attns *= attention_mask.unsqueeze(1).unsqueeze(2).unsqueeze(3) |
|
attns *= attention_mask.unsqueeze(1).unsqueeze(2).unsqueeze(4) |
|
return self.contact_head(tokens, attns) |
|
|
|
def create_position_ids_from_input_ids( |
|
input_ids, padding_idx, past_key_values_length=0 |
|
): |
|
""" |
|
Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols |
|
are ignored. This is modified from fairseq's `utils.make_positions`. |
|
|
|
Args: |
|
x: torch.Tensor x: |
|
|
|
Returns: torch.Tensor |
|
""" |
|
|
|
mask = input_ids.ne(padding_idx).int() |
|
incremental_indices = ( |
|
torch.cumsum(mask, dim=1).type_as(mask) + past_key_values_length |
|
) * mask |
|
return incremental_indices.long() + padding_idx |
|
|
|
|
|
|
|
|
|
class SegmentNT(EsmPreTrainedModel): |
|
def __init__(self, config): |
|
super().__init__(config) |
|
self.num_labels = config.num_labels |
|
self.config = config |
|
self.num_features = len(config.features) |
|
|
|
self.esm = EsmModel(config, add_pooling_layer=False) |
|
|
|
embed_dim = config.hidden_size |
|
num_layers = config.num_layers_head |
|
self.unet = UNET1DSegmentationHead( |
|
embed_dim=embed_dim, |
|
num_classes=embed_dim // 2, |
|
output_channels_list=tuple( |
|
embed_dim * (2**i) for i in range(num_layers) |
|
), |
|
) |
|
self.fc = nn.Linear(in_features=embed_dim, out_features=6 * 2 * self.num_features) |
|
self.activation_fn = nn.SiLU() |
|
|
|
self.init_weights() |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def forward( |
|
self, |
|
input_ids: Optional[torch.LongTensor] = None, |
|
attention_mask: Optional[torch.Tensor] = None, |
|
position_ids: Optional[torch.LongTensor] = None, |
|
head_mask: Optional[torch.Tensor] = None, |
|
inputs_embeds: Optional[torch.FloatTensor] = None, |
|
labels: Optional[torch.LongTensor] = None, |
|
output_attentions: Optional[bool] = None, |
|
output_hidden_states: Optional[bool] = None, |
|
return_dict: Optional[bool] = None, |
|
) -> Union[Tuple, SequenceClassifierOutput]: |
|
r""" |
|
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): |
|
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., |
|
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If |
|
`config.num_labels > 1` a classification loss is computed (Cross-Entropy). |
|
""" |
|
return_dict = ( |
|
return_dict if return_dict is not None else self.config.use_return_dict |
|
) |
|
|
|
outputs = self.esm( |
|
input_ids, |
|
attention_mask=attention_mask, |
|
position_ids=position_ids, |
|
head_mask=head_mask, |
|
inputs_embeds=inputs_embeds, |
|
output_attentions=output_attentions, |
|
output_hidden_states=output_hidden_states, |
|
return_dict=return_dict, |
|
) |
|
sequence_output = outputs[0] |
|
|
|
sequence_output = sequence_output[:,1:,:] |
|
|
|
|
|
|
|
sequence_output = torch.transpose(sequence_output, 2,1) |
|
|
|
x = self.activation_fn(self.unet(sequence_output)) |
|
|
|
|
|
x = torch.transpose(x, 2,1) |
|
|
|
logits = self.fc(x) |
|
|
|
|
|
logits = torch.reshape(logits, (x.shape[0], x.shape[1] * 6, self.num_features, 2)) |
|
|
|
|
|
outputs["logits"] = logits |
|
|
|
return outputs |
|
|
|
|
|
class DownSample1D(nn.Module): |
|
""" |
|
1D-UNET downsampling block. |
|
""" |
|
|
|
def __init__( |
|
self, |
|
input_channels: int, |
|
output_channels: int, |
|
num_layers: int = 2, |
|
): |
|
""" |
|
Args: |
|
output_channels: number of output channels. |
|
activation_fn: name of the activation function to use. |
|
Should be one of "gelu", |
|
"gelu-no-approx", "relu", "swish", "silu", "sin". |
|
num_layers: number of convolution layers. |
|
name: module name. |
|
""" |
|
|
|
super().__init__() |
|
self.first_layer = [nn.Conv1d( |
|
in_channels=input_channels, |
|
out_channels=output_channels, |
|
kernel_size=3, |
|
stride=1, |
|
dilation=1, |
|
padding="same", |
|
)] |
|
|
|
|
|
self.next_layers = [ |
|
nn.Conv1d( |
|
in_channels=output_channels, |
|
out_channels=output_channels, |
|
kernel_size=3, |
|
stride=1, |
|
dilation=1, |
|
padding="same", |
|
) |
|
for _ in range(num_layers-1) |
|
] |
|
self.conv_layers = nn.ModuleList(self.first_layer + self.next_layers) |
|
|
|
self.avg_pool = nn.AvgPool1d( |
|
kernel_size=2, |
|
stride=2, |
|
padding=0, |
|
) |
|
self.activation_fn = nn.SiLU() |
|
|
|
|
|
def forward(self, x: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: |
|
for i, conv_layer in enumerate(self.conv_layers): |
|
x = self.activation_fn(conv_layer(x)) |
|
|
|
hidden = x |
|
x = self.avg_pool(hidden) |
|
return x, hidden |
|
|
|
|
|
|
|
class UpSample1D(nn.Module): |
|
""" |
|
1D-UNET upsampling block. |
|
""" |
|
|
|
def __init__( |
|
self, |
|
input_channels: int, |
|
output_channels: int, |
|
num_layers: int = 2, |
|
): |
|
""" |
|
Args: |
|
output_channels: number of output channels. |
|
activation_fn: name of the activation function to use. |
|
Should be one of "gelu", |
|
"gelu-no-approx", "relu", "swish", "silu", "sin". |
|
interpolation_method: Method to be used for upsampling interpolation. |
|
Should be one of "nearest", "linear", "cubic", "lanczos3", "lanczos5". |
|
num_layers: number of convolution layers. |
|
name: module name. |
|
""" |
|
super().__init__() |
|
|
|
self._first_layer = [nn.ConvTranspose1d( |
|
in_channels=input_channels, |
|
out_channels=output_channels, |
|
kernel_size=3, |
|
stride=1, |
|
padding=1, |
|
)] |
|
|
|
|
|
self._next_layers = [ |
|
nn.ConvTranspose1d( |
|
in_channels=output_channels, |
|
out_channels=output_channels, |
|
kernel_size=3, |
|
stride=1, |
|
padding=1, |
|
) |
|
for _ in range(num_layers-1) |
|
] |
|
|
|
self.conv_layers = nn.ModuleList(self._first_layer + self._next_layers) |
|
|
|
self._activation_fn = nn.SiLU() |
|
|
|
|
|
|
|
def forward(self, x: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: |
|
for i, conv_layer in enumerate(self.conv_layers): |
|
x = self._activation_fn(conv_layer(x)) |
|
|
|
|
|
|
|
x = nn.functional.interpolate(x, size=2 * x.shape[2], mode="nearest") |
|
|
|
|
|
return x |
|
|
|
|
|
|
|
class FinalConv1D(nn.Module): |
|
""" |
|
Final output block of the 1D-UNET. |
|
""" |
|
|
|
def __init__( |
|
self, |
|
input_channels: int, |
|
output_channels: int, |
|
num_layers: int = 2, |
|
): |
|
""" |
|
Args: |
|
output_channels: number of output channels. |
|
activation_fn: name of the activation function to use. |
|
Should be one of "gelu", |
|
"gelu-no-approx", "relu", "swish", "silu", "sin". |
|
num_layers: number of convolution layers. |
|
name: module name. |
|
""" |
|
super().__init__() |
|
|
|
self._first_layer = [nn.Conv1d( |
|
in_channels=input_channels, |
|
out_channels=output_channels, |
|
kernel_size=3, |
|
stride=1, |
|
dilation=1, |
|
padding="same", |
|
)] |
|
|
|
self._next_layers = [ |
|
nn.Conv1d( |
|
in_channels=output_channels, |
|
out_channels=output_channels, |
|
kernel_size=3, |
|
stride=1, |
|
dilation=1, |
|
padding="same", |
|
) |
|
for _ in range(num_layers-1) |
|
] |
|
self.conv_layers = nn.ModuleList(self._first_layer + self._next_layers) |
|
|
|
self._activation_fn = nn.SiLU() |
|
|
|
|
|
|
|
def forward(self, x: torch.Tensor) -> torch.Tensor: |
|
for i, conv_layer in enumerate(self.conv_layers): |
|
x = conv_layer(x) |
|
if i < len(self.conv_layers) - 1: |
|
x = self._activation_fn(x) |
|
return x |
|
|
|
|
|
class UNET1DSegmentationHead(nn.Module): |
|
""" |
|
1D-UNET based head to be plugged on top of a pretrained model to perform |
|
semantic segmentation. |
|
""" |
|
|
|
def __init__( |
|
self, |
|
embed_dim: int, |
|
num_classes: int, |
|
output_channels_list: Tuple[int, ...] = (64, 128, 256), |
|
num_conv_layers_per_block: int = 2, |
|
): |
|
""" |
|
Args: |
|
num_classes: number of classes to segment |
|
output_channels_list: list of the number of output channel at each level of |
|
the UNET |
|
num_conv_layers_per_block: number of convolution layers per block. |
|
""" |
|
super().__init__() |
|
self._num_pooling_layers = len(output_channels_list) |
|
|
|
|
|
downsample_input_channels_list = (embed_dim, ) + output_channels_list[:-1] |
|
|
|
output_channels_list_reversed = tuple(reversed(output_channels_list)) |
|
upsample_input_channels_list = (output_channels_list[-1],) + output_channels_list_reversed |
|
upsample_output_channels_list = output_channels_list_reversed |
|
|
|
self._downsample_blocks = nn.ModuleList([ |
|
DownSample1D( |
|
input_channels= input_channels, |
|
output_channels=output_channels, |
|
num_layers=num_conv_layers_per_block, |
|
) |
|
for input_channels, output_channels in zip(downsample_input_channels_list, output_channels_list) |
|
]) |
|
|
|
self._upsample_blocks = nn.ModuleList([ |
|
UpSample1D( |
|
input_channels = input_channels, |
|
output_channels=output_channels, |
|
num_layers=num_conv_layers_per_block, |
|
) |
|
for input_channels, output_channels in zip(upsample_input_channels_list, upsample_output_channels_list) |
|
]) |
|
|
|
self.final_block = FinalConv1D( |
|
input_channels=output_channels_list[0], |
|
output_channels=num_classes * 2, |
|
num_layers=num_conv_layers_per_block, |
|
) |
|
|
|
def forward(self, x: torch.Tensor) -> torch.Tensor: |
|
|
|
if x.shape[2] % 2**self._num_pooling_layers: |
|
raise ValueError( |
|
"Input length must be divisible by the 2 to the power of" |
|
" number of poolign layers." |
|
) |
|
|
|
hiddens = [] |
|
for downsample_block in self._downsample_blocks: |
|
x, hidden = downsample_block(x) |
|
hiddens.append(hidden) |
|
|
|
|
|
|
|
for i, (upsample_block, hidden) in enumerate(zip(self._upsample_blocks, reversed(hiddens))): |
|
x = upsample_block(x) + hidden |
|
x = self.final_block(x) |
|
return x |
|
|
|
|