|
from dataclasses import dataclass |
|
from typing import Any, Dict, List |
|
|
|
import torch |
|
|
|
|
|
@dataclass |
|
class MyCollator: |
|
pad_token_id: int |
|
attention_pad_value: int = 0 |
|
label_pad_value: int = -100 |
|
|
|
def __call__(self, features: List[Dict[str, Any]]) -> Dict[str, Any]: |
|
|
|
input_ids = [f["input_ids"] for f in features] |
|
placeholder_indices = [f["placeholder_indices"] for f in features] |
|
super_input_ids = [f["super_input_ids"] for f in features] |
|
super_token_indices = [f["super_token_indices"] for f in features] |
|
labels = [f["labels"] for f in features] if "labels" in features[0] else None |
|
|
|
|
|
( |
|
input_ids, |
|
attention_mask, |
|
placeholder_indices, |
|
labels, |
|
) = self.process_model_inputs( |
|
input_ids, |
|
placeholder_indices, |
|
labels, |
|
) |
|
|
|
|
|
( |
|
super_input_ids, |
|
super_attention_mask, |
|
super_token_indices, |
|
) = self.process_super_tokenizer_inputs( |
|
super_input_ids, |
|
super_token_indices, |
|
) |
|
|
|
|
|
input_ids = torch.tensor(input_ids) |
|
attention_mask = torch.tensor(attention_mask) |
|
super_input_ids = torch.tensor(super_input_ids) |
|
super_attention_mask = torch.tensor(super_attention_mask) |
|
labels = torch.tensor(labels) if labels else None |
|
|
|
|
|
res = { |
|
"input_ids": input_ids, |
|
"attention_mask": attention_mask, |
|
"super_input_ids": super_input_ids, |
|
"super_attention_mask": super_attention_mask, |
|
"placeholder_indices": placeholder_indices, |
|
"super_token_indices": super_token_indices, |
|
"labels": labels, |
|
} |
|
|
|
return res |
|
|
|
def process_model_inputs(self, input_ids, placeholder_indices, labels): |
|
|
|
max_len = get_max_length_in_nested_lists(input_ids) |
|
attention_mask = get_attention_mask_from_nested_lists(input_ids) |
|
|
|
|
|
placeholder_indices = [ |
|
[idx + max_len - len(input_ids[i]) for idx in placeholder_indices[i]] |
|
for i in range(len(placeholder_indices)) |
|
] |
|
|
|
|
|
input_ids = pad_nested_lists(input_ids, max_len, self.pad_token_id, "left") |
|
attention_mask = pad_nested_lists(attention_mask, max_len, self.attention_pad_value, "left") |
|
if labels: |
|
labels = pad_nested_lists(labels, max_len, self.label_pad_value, "left") |
|
|
|
return input_ids, attention_mask, placeholder_indices, labels |
|
|
|
def process_super_tokenizer_inputs(self, input_ids, super_token_indices): |
|
|
|
input_ids = sum(input_ids, []) |
|
super_token_indices = sum(super_token_indices, []) |
|
|
|
|
|
new_input_ids = [] |
|
new_super_token_indices = [] |
|
for i in range(len(input_ids)): |
|
if len(super_token_indices[i]) != 0: |
|
new_input_ids.append(input_ids[i]) |
|
new_super_token_indices.append(super_token_indices[i]) |
|
input_ids = new_input_ids |
|
super_token_indices = new_super_token_indices |
|
|
|
if len(input_ids) == 0: |
|
return [], [], [] |
|
|
|
|
|
max_len = get_max_length_in_nested_lists(input_ids) |
|
attention_mask = get_attention_mask_from_nested_lists(input_ids) |
|
|
|
input_ids = pad_nested_lists(input_ids, max_len, self.pad_token_id) |
|
attention_mask = pad_nested_lists(attention_mask, max_len, self.attention_pad_value) |
|
|
|
return input_ids, attention_mask, super_token_indices |
|
|
|
def get_max_length_in_nested_lists(lst): |
|
if isinstance(lst[0], list): |
|
lengths = [] |
|
for elem in lst: |
|
length = get_max_length_in_nested_lists(elem) |
|
lengths.append(length) |
|
max_length = max(lengths) |
|
return max_length |
|
else: |
|
return len(lst) |
|
|
|
|
|
def get_attention_mask_from_nested_lists(lst): |
|
if isinstance(lst[0], list): |
|
attention_mask = [] |
|
for elem in lst: |
|
mask = get_attention_mask_from_nested_lists(elem) |
|
attention_mask.append(mask) |
|
return attention_mask |
|
else: |
|
return [1] * len(lst) |
|
|
|
|
|
def pad_nested_lists(lst, max_length, padding_value, padding_side="right"): |
|
if isinstance(lst, list) and len(lst) and isinstance(lst[0], list): |
|
for i, elem in enumerate(lst): |
|
lst[i] = pad_nested_lists(elem, max_length, padding_value, padding_side) |
|
return lst |
|
elif isinstance(lst, list): |
|
if padding_side == "right": |
|
return lst + [padding_value for _ in range(max_length - len(lst))] |
|
else: |
|
return [padding_value for _ in range(max_length - len(lst))] + lst |
|
else: |
|
raise NotImplementedError(f"Unrecognized type {lst}") |
|
|