Datasets:

ArXiv:
Elron commited on
Commit
0eaf17a
1 Parent(s): b3f7bba

Upload fusion.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. fusion.py +11 -13
fusion.py CHANGED
@@ -1,19 +1,15 @@
1
  import copy
2
  from abc import abstractmethod
3
- from dataclasses import asdict
4
  from typing import Generator, List, Optional
5
 
6
- from .card import ICLCard, TaskCard
7
- from .common import CommonRecipe
8
  from .dataclass import NonPositionalField
9
  from .operator import SourceOperator, StreamSource
10
- from .random_utils import random
11
  from .stream import MultiStream, Stream
12
 
13
 
14
  class BaseFusion(SourceOperator):
15
- """
16
- BaseFusion operator that combines multiple streams into one.
17
 
18
  Args:
19
  include_splits: List of splits to include. If None, all splits are included.
@@ -45,8 +41,7 @@ class BaseFusion(SourceOperator):
45
 
46
 
47
  class FixedFusion(BaseFusion):
48
- """
49
- FixedFusion operator that combines multiple streams into one based on a fixed number of examples per task.
50
 
51
  Args:
52
  orgins: List of StreamSource objects.
@@ -70,8 +65,7 @@ class FixedFusion(BaseFusion):
70
 
71
 
72
  class WeightedFusion(BaseFusion):
73
- """
74
- Fusion operator that combines multiple streams based
75
 
76
  Args:
77
  orgins: List of StreamSource objects.
@@ -87,14 +81,18 @@ class WeightedFusion(BaseFusion):
87
  super().verify()
88
  assert self.origins is not None, "origins must be specified"
89
  assert self.weights is not None, "weights must be specified"
90
- assert len(self.origins) == len(self.weights), "origins and weights must have the same length"
 
 
91
 
92
  def fusion_generator(self, split) -> Generator:
93
  weights = copy.deepcopy(self.weights)
94
  iterators = [iter(origin()[split]) for origin in self.origins]
95
  total_examples = 0
96
- while (self.max_total_examples is None or total_examples <= self.max_total_examples) and len(iterators) > 0:
97
- iterator = random.choices(population=iterators, weights=weights)[0]
 
 
98
  try:
99
  yield next(iterator)
100
  total_examples += 1
 
1
  import copy
2
  from abc import abstractmethod
 
3
  from typing import Generator, List, Optional
4
 
 
 
5
  from .dataclass import NonPositionalField
6
  from .operator import SourceOperator, StreamSource
7
+ from .random_utils import get_random
8
  from .stream import MultiStream, Stream
9
 
10
 
11
  class BaseFusion(SourceOperator):
12
+ """BaseFusion operator that combines multiple streams into one.
 
13
 
14
  Args:
15
  include_splits: List of splits to include. If None, all splits are included.
 
41
 
42
 
43
  class FixedFusion(BaseFusion):
44
+ """FixedFusion operator that combines multiple streams into one based on a fixed number of examples per task.
 
45
 
46
  Args:
47
  orgins: List of StreamSource objects.
 
65
 
66
 
67
  class WeightedFusion(BaseFusion):
68
+ """Fusion operator that combines multiple streams based.
 
69
 
70
  Args:
71
  orgins: List of StreamSource objects.
 
81
  super().verify()
82
  assert self.origins is not None, "origins must be specified"
83
  assert self.weights is not None, "weights must be specified"
84
+ assert len(self.origins) == len(
85
+ self.weights
86
+ ), "origins and weights must have the same length"
87
 
88
  def fusion_generator(self, split) -> Generator:
89
  weights = copy.deepcopy(self.weights)
90
  iterators = [iter(origin()[split]) for origin in self.origins]
91
  total_examples = 0
92
+ while (
93
+ self.max_total_examples is None or total_examples <= self.max_total_examples
94
+ ) and len(iterators) > 0:
95
+ iterator = get_random().choices(population=iterators, weights=weights)[0]
96
  try:
97
  yield next(iterator)
98
  total_examples += 1