wehub-resource-sync
208 lines
8.5 KiB
Python
208 lines
8.5 KiB
Python
import logging
|
|
from typing import List
|
|
|
|
import torch
|
|
from transformers import AutoModel, PreTrainedModel, PreTrainedTokenizer
|
|
|
|
from FlagEmbedding.abc.finetune.embedder import AbsEmbedderModel
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class BiDecoderOnlyEmbedderModel(AbsEmbedderModel):
|
|
"""Embedder model class for decoder only model.
|
|
|
|
Args:
|
|
base_model (PreTrainedModel): The base model to train on.
|
|
tokenizer (PreTrainedTokenizer, optional): The tokenizer to use. Defaults to ``None``.
|
|
negatives_cross_device (bool, optional): If True, will compute cross devices negative loss. Defaults to ``False``.
|
|
temperature (float, optional): Temperature to control the scale of scores. Defaults to ``1.0``.
|
|
sub_batch_size (int, optional): Sub-batch size during encoding. If negative, will not split to sub-batch.
|
|
Defaults to ``-1``.
|
|
kd_loss_type (str, optional): Type of knowledge distillation loss. Defaults to ``'kl_div'``.
|
|
use_mrl (bool, optional): Whether to use MRL for training. Defaults to ``False``.
|
|
mrl_dims (List[int], optional): The dimensions of MRL layers. Defaults to ``[]``.
|
|
sentence_pooling_method (str, optional): Pooling method to get sentence embedding. Defaults to ``'last_token'``.
|
|
normalize_embeddings (bool, optional): If True, normalize the embedding vector. Defaults to ``False``.
|
|
"""
|
|
TRANSFORMER_CLS = AutoModel
|
|
|
|
def __init__(
|
|
self,
|
|
base_model: PreTrainedModel,
|
|
tokenizer: PreTrainedTokenizer = None,
|
|
negatives_cross_device: bool = False,
|
|
temperature: float = 1.0,
|
|
sub_batch_size: int = -1,
|
|
kd_loss_type: str = 'kl_div',
|
|
use_mrl: bool = False,
|
|
mrl_dims: List[int] = [],
|
|
sentence_pooling_method: str = 'last_token',
|
|
normalize_embeddings: bool = False,
|
|
):
|
|
super().__init__(
|
|
base_model,
|
|
tokenizer=tokenizer,
|
|
negatives_cross_device=negatives_cross_device,
|
|
temperature=temperature,
|
|
sub_batch_size=sub_batch_size,
|
|
kd_loss_type=kd_loss_type,
|
|
use_mrl=use_mrl,
|
|
mrl_dims=mrl_dims,
|
|
)
|
|
self.sentence_pooling_method = sentence_pooling_method
|
|
self.normalize_embeddings = normalize_embeddings
|
|
self.cross_entropy = torch.nn.CrossEntropyLoss(reduction='mean')
|
|
|
|
def encode(self, features):
|
|
"""
|
|
Encode and get the embedding.
|
|
|
|
Args:
|
|
features (Union[list, dict]): Features feed to the model.
|
|
|
|
Returns:
|
|
Union[torch.Tensor, List[torch.Tensor]]: The embedding vectors or a list of embedding vectors if MRL is used.
|
|
"""
|
|
if features is None:
|
|
return None
|
|
if not isinstance(features, list):
|
|
if self.sub_batch_size is not None and self.sub_batch_size > 0:
|
|
all_p_reps = []
|
|
for i in range(0, len(features['attention_mask']), self.sub_batch_size):
|
|
end_inx = min(i + self.sub_batch_size, len(features['attention_mask']))
|
|
sub_features = {}
|
|
for k, v in features.items():
|
|
sub_features[k] = v[i:end_inx]
|
|
last_hidden_state = self.model(**sub_features, return_dict=True).last_hidden_state
|
|
p_reps = self._sentence_embedding(last_hidden_state, sub_features['attention_mask'])
|
|
all_p_reps.append(p_reps)
|
|
all_p_reps = torch.cat(all_p_reps, 0).contiguous()
|
|
else:
|
|
last_hidden_state = self.model(**features, return_dict=True).last_hidden_state
|
|
all_p_reps = self._sentence_embedding(last_hidden_state, features['attention_mask'])
|
|
else:
|
|
all_p_reps = []
|
|
for sub_features in features:
|
|
last_hidden_state = self.model(**sub_features, return_dict=True).last_hidden_state
|
|
p_reps = self._sentence_embedding(last_hidden_state, sub_features['attention_mask'])
|
|
all_p_reps.append(p_reps)
|
|
all_p_reps = torch.cat(all_p_reps, 0).contiguous()
|
|
|
|
if self.use_mrl:
|
|
p_reps_list = []
|
|
ori_dim = all_p_reps.size(-1)
|
|
for dim in self.mrl_dims:
|
|
if dim > ori_dim:
|
|
logger.warning(f"MRL dim {dim} is larger than original dimension {ori_dim}, using original dimension instead.")
|
|
dim = min(dim, ori_dim)
|
|
dim_p_reps = all_p_reps[:, :dim]
|
|
if self.normalize_embeddings:
|
|
dim_p_reps = torch.nn.functional.normalize(dim_p_reps, dim=-1)
|
|
p_reps_list.append(dim_p_reps.contiguous())
|
|
return p_reps_list
|
|
else:
|
|
if self.normalize_embeddings:
|
|
all_p_reps = torch.nn.functional.normalize(all_p_reps, dim=-1)
|
|
return all_p_reps.contiguous()
|
|
|
|
def _sentence_embedding(self, last_hidden_state, attention_mask):
|
|
"""Use the pooling method to get the sentence embedding.
|
|
|
|
Args:
|
|
last_hidden_state (torch.Tensor): The model output's last hidden state.
|
|
attention_mask (torch.Tensor): Mask out padding tokens during pooling.
|
|
|
|
Raises:
|
|
NotImplementedError: Specified pooling method not implemented.
|
|
|
|
Returns:
|
|
torch.Tensor: The sentence embeddings.
|
|
"""
|
|
if self.sentence_pooling_method == "cls":
|
|
return last_hidden_state[:, 0]
|
|
elif self.sentence_pooling_method == "mean":
|
|
s = torch.sum(
|
|
last_hidden_state * attention_mask.unsqueeze(-1).float(), dim=1
|
|
)
|
|
d = attention_mask.sum(dim=1, keepdim=True).float()
|
|
return s / d
|
|
elif self.sentence_pooling_method == "last_token":
|
|
left_padding = attention_mask[:, -1].sum() == attention_mask.shape[0]
|
|
if left_padding:
|
|
return last_hidden_state[:, -1]
|
|
else:
|
|
sequence_lengths = attention_mask.sum(dim=1) - 1
|
|
batch_size = last_hidden_state.shape[0]
|
|
return last_hidden_state[
|
|
torch.arange(batch_size, device=last_hidden_state.device),
|
|
sequence_lengths,
|
|
]
|
|
else:
|
|
raise NotImplementedError(f"pooling method {self.sentence_pooling_method} not implemented")
|
|
|
|
def compute_score(self, q_reps, p_reps):
|
|
"""Computes the scores between query and passage representations.
|
|
|
|
Args:
|
|
q_reps (torch.Tensor): Query representations.
|
|
p_reps (torch.Tensor): Passage representations.
|
|
|
|
Returns:
|
|
torch.Tensor: The computed scores, adjusted by temperature.
|
|
"""
|
|
scores = self._compute_similarity(q_reps, p_reps) / self.temperature
|
|
scores = scores.view(q_reps.size(0), -1)
|
|
return scores
|
|
|
|
def _compute_similarity(self, q_reps, p_reps):
|
|
"""Computes the similarity between query and passage representations using inner product.
|
|
|
|
Args:
|
|
q_reps (torch.Tensor): Query representations.
|
|
p_reps (torch.Tensor): Passage representations.
|
|
|
|
Returns:
|
|
torch.Tensor: The computed similarity matrix.
|
|
"""
|
|
if len(p_reps.size()) == 2:
|
|
return torch.matmul(q_reps, p_reps.transpose(0, 1))
|
|
return torch.matmul(q_reps, p_reps.transpose(-2, -1))
|
|
|
|
def compute_loss(self, scores, target):
|
|
"""Compute the loss using cross entropy.
|
|
|
|
Args:
|
|
scores (torch.Tensor): Computed score.
|
|
target (torch.Tensor): The target value.
|
|
|
|
Returns:
|
|
torch.Tensor: The computed cross entropy loss.
|
|
"""
|
|
return self.cross_entropy(scores, target)
|
|
|
|
def gradient_checkpointing_enable(self, **kwargs):
|
|
"""
|
|
Activates gradient checkpointing for the current model.
|
|
"""
|
|
self.model.gradient_checkpointing_enable(**kwargs)
|
|
|
|
def enable_input_require_grads(self, **kwargs):
|
|
"""
|
|
Enables the gradients for the input embeddings.
|
|
"""
|
|
self.model.enable_input_require_grads(**kwargs)
|
|
|
|
def save(self, output_dir: str):
|
|
"""Save the model to the directory.
|
|
|
|
Args:
|
|
output_dir (str): Directory for saving the model.
|
|
"""
|
|
state_dict = self.model.state_dict()
|
|
state_dict = type(state_dict)(
|
|
{k: v.clone().cpu()
|
|
for k,
|
|
v in state_dict.items()})
|
|
self.model.save_pretrained(output_dir, state_dict=state_dict)
|