Skip to content

HumanEval

get_input_data_model()

Returns HumanEval input data model.

Source code in synalinks/src/datasets/built_in/humaneval.py 
27
28
29
30
@synalinks_export("synalinks.datasets.humaneval.get_input_data_model")
def get_input_data_model():
    """Returns HumanEval input data model."""
    return HumanEvalQuestion

get_output_data_model()

Returns HumanEval output data model.

Source code in synalinks/src/datasets/built_in/humaneval.py 
33
34
35
36
@synalinks_export("synalinks.datasets.humaneval.get_output_data_model")
def get_output_data_model():
    """Returns HumanEval output data model."""
    return HumanEvalAnswer

iterable_dataset(repeat=1, batch_size=1, limit=None, split='test')

Streaming dataset for RL-style training.

Returns:

Type Description
HuggingFaceDataset

A streaming, iterable dataset.
Source code in synalinks/src/datasets/built_in/humaneval.py 
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
@synalinks_export("synalinks.datasets.humaneval.iterable_dataset")
def iterable_dataset(repeat=1, batch_size=1, limit=None, split="test"):
    """
    Streaming dataset for RL-style training.

    Returns:
        (HuggingFaceDataset): A streaming, iterable dataset.
    """
    return HuggingFaceDataset(
        path="openai/openai_humaneval",
        split=split,
        streaming=True,
        input_data_model=HumanEvalQuestion,
        input_template=_INPUT_TEMPLATE,
        output_data_model=HumanEvalAnswer,
        output_template=_OUTPUT_TEMPLATE,
        batch_size=batch_size,
        limit=limit,
        repeat=repeat,
    )

load_data(validation_split=0.2)

Load HumanEval.

HF ships only a test split (164 problems), so we split it deterministically into train / test. Real scoring requires running unit tests against the completion — exact match against the canonical solution is a placeholder reward.

Parameters:

Name Type Description Default
validation_split float

Fraction held out for evaluation (default 0.2).

 0.2

Returns:

Type Description
tuple

(x_train, y_train), (x_test, y_test).
Source code in synalinks/src/datasets/built_in/humaneval.py 
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
@synalinks_export("synalinks.datasets.humaneval.load_data")
def load_data(validation_split=0.2):
    """
    Load HumanEval.

    HF ships only a ``test`` split (164 problems), so we split it
    deterministically into train / test. Real scoring requires running
    unit tests against the completion — exact match against the
    canonical solution is a placeholder reward.

    Args:
        validation_split (float): Fraction held out for evaluation
            (default ``0.2``).

    Returns:
        (tuple): ``(x_train, y_train), (x_test, y_test)``.
    """
    x, y = load_split(
        path="openai/openai_humaneval",
        split="test",
        input_data_model=HumanEvalQuestion,
        input_template=_INPUT_TEMPLATE,
        output_data_model=HumanEvalAnswer,
        output_template=_OUTPUT_TEMPLATE,
    )
    return split_train_test(x, y, validation_split=validation_split)