Datasets:

ArXiv:
File size: 2,960 Bytes
249a1e4
 
71c3f75
 
249a1e4
 
71c3f75
43bb101
249a1e4
 
71c3f75
 
 
 
 
43bb101
71c3f75
 
 
 
 
 
 
 
43bb101
71c3f75
 
 
 
 
 
 
 
 
 
 
 
 
43bb101
 
 
 
71c3f75
 
 
 
 
c640e90
71c3f75
 
 
 
 
 
 
 
 
 
43bb101
 
71c3f75
 
 
 
 
 
 
 
 
43bb101
 
71c3f75
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5a92e5e
71c3f75
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
from typing import Union

from .card import TaskCard
from .collections import ItemPicker, RandomPicker
from .operator import SourceOperator
from .recipe import Recipe, SequentialRecipe
from .schema import ToUnitxtGroup
from .splitters import RandomSampler, SeparateSplit, SliceSplit, SpreadSplit
from .stream import MultiStream
from .templates import RenderTemplatedICL


class CommonRecipe(Recipe, SourceOperator):
    card: TaskCard
    demos_pool_name: str = "demos_pool"
    demos_taken_from: str = "train"
    demos_pool_size: int = None
    demos_field: str = "demos"
    num_demos: int = None
    sampler_type: str = "random"
    instruction_item: Union[str, int] = None
    template_item: Union[str, int] = None

    def verify(self):
        assert self.sampler_type in ["random"], f"Uknown sampler type {self.sampler_type}"

    def prepare(self):
        steps = [
            self.card.loader,
        ]

        if self.card.preprocess_steps is not None:
            steps.extend(self.card.preprocess_steps)

        steps.append(self.card.task)

        if self.demos_pool_size is not None:
            steps.append(
                SeparateSplit(
                    from_split=self.demos_taken_from,
                    to_split_names=[self.demos_pool_name, self.demos_taken_from],
                    to_split_sizes=[int(self.demos_pool_size)],
                )
            )

        if self.num_demos is not None:
            if self.sampler_type == "random":
                sampler = RandomSampler(sample_size=int(self.num_demos))

            steps.append(
                SpreadSplit(
                    source_stream=self.demos_pool_name,
                    target_field=self.demos_field,
                    sampler=sampler,
                )
            )

        if self.card.instructions is not None:
            if not self.instruction_item is None:
                picker = ItemPicker(int(self.instruction_item))
            else:
                picker = RandomPicker()
            instruction = picker(self.card.instructions)
        else:
            instruction = None

        if self.card.templates is not None:
            if self.template_item is None:
                picker = RandomPicker()
            else:
                picker = ItemPicker(self.template_item)
            template = picker(self.card.templates)
        else:
            template = None

        render = RenderTemplatedICL(
            instruction=instruction,
            template=template,
            demos_field=self.demos_field,
        )

        steps.append(render)

        postprocessors = render.get_postprocessors()

        steps.append(
            ToUnitxtGroup(
                group="unitxt",
                metrics=self.card.task.metrics,
                postprocessors=postprocessors,
            )
        )

        self.recipe = SequentialRecipe(steps)

    def process(self) -> MultiStream:
        return self.recipe()