File size: 4,306 Bytes
07423df
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
import json
import os
import sys

import numpy as np
import pandas as pd
import pytest
import yaml
from transformers.testing_utils import execute_subprocess_async

from llm_studio.app_utils.default_datasets import (
    prepare_default_dataset_causal_language_modeling,
)


def get_experiment_status(path: str) -> str:
    """Get status information from experiment.

    Args:
        path: path to experiment folder
    Returns:
        experiment status
    """

    try:
        flag_json_path = os.path.join(path, "flags.json")
        if not os.path.exists(flag_json_path):
            return "none"
        with open(flag_json_path) as file:
            flags = json.load(file)
            status = flags.get("status", "none")
        return status
    except Exception:
        return "none"


@pytest.mark.parametrize(
    "config_name",
    [
        "test_causal_language_modeling_oasst_cfg",
        "test_sequence_to_sequence_modeling_oasst_cfg",
    ],
)
@pytest.mark.parametrize(
    "metric",
    [
        "Perplexity",
        "BLEU",
    ],
)
def test_oasst_training_gpu(tmp_path, config_name, metric):
    run_oasst(tmp_path, config_name, metric)


@pytest.mark.parametrize(
    "settings",
    [
        ["AUC", "test_causal_binary_classification_modeling_cfg"],
        ["LogLoss", "test_causal_multiclass_classification_modeling_cfg"],
    ],
)
def test_oasst_classification_training_gpu(tmp_path, settings):
    metric, config_name = settings
    run_oasst(
        tmp_path,
        config_name=config_name,
        metric=metric,
    )


@pytest.mark.parametrize(
    "settings",
    [
        ["AUC", "test_causal_binary_classification_modeling_cpu_cfg"],
        ["LogLoss", "test_causal_multiclass_classification_modeling_cpu_cfg"],
    ],
)
def test_oasst_classification_training_cpu(tmp_path, settings):
    metric, config_name = settings
    run_oasst(
        tmp_path,
        config_name=config_name,
        metric=metric,
    )


@pytest.mark.parametrize(
    "config_name",
    [
        "test_causal_language_modeling_oasst_cpu_cfg",
        "test_sequence_to_sequence_modeling_oasst_cpu_cfg",
    ],
)
@pytest.mark.parametrize(
    "metric",
    [
        "Perplexity",
        "BLEU",
    ],
)
def test_oasst_training_cpu(tmp_path, config_name, metric):
    run_oasst(tmp_path, config_name, metric)


def run_oasst(tmp_path, config_name, metric):
    """
    Test training on OASST dataset.

    Pytest keeps around the last 3 test runs in the tmp_path fixture.
    """
    prepare_default_dataset_causal_language_modeling(tmp_path)
    train_path = os.path.join(tmp_path, "train_full.pq")
    # create dummy labels for classification problem type,
    # unused for other problem types
    df = pd.read_parquet(train_path)
    df["multiclass_label"] = np.random.choice(["0", "1", "2"], size=len(df))
    df["binary_label"] = np.random.choice(["0", "1"], size=len(df))
    df.to_parquet(train_path)

    with open(
        os.path.join(
            os.path.dirname(os.path.realpath(__file__)), f"{config_name}.yaml"
        ),
        "r",
    ) as fp:
        cfg = yaml.load(fp, Loader=yaml.FullLoader)
    # set paths and save in tmp folder
    cfg["dataset"]["train_dataframe"] = train_path
    cfg["output_directory"] = os.path.join(tmp_path, "output")
    # set metric
    cfg["prediction"]["metric"] = metric
    modifed_config_path = os.path.join(tmp_path, "cfg.yaml")
    with open(modifed_config_path, "w") as fp:
        yaml.dump(cfg, fp)

    # llm studio root directory.
    root_dir = os.path.abspath(
        os.path.join(os.path.dirname(os.path.realpath(__file__)), "../../")
    )
    cmd = [
        f"{sys.executable}",
        os.path.join(root_dir, "train.py"),
        "-Y",
        f"{modifed_config_path}",
    ]
    execute_subprocess_async(cmd)
    assert os.path.exists(cfg["output_directory"])
    status = get_experiment_status(path=cfg["output_directory"])
    assert status == "finished"
    assert os.path.exists(os.path.join(cfg["output_directory"], "charts.db"))
    assert os.path.exists(os.path.join(cfg["output_directory"], "checkpoint.pth"))
    assert os.path.exists(os.path.join(cfg["output_directory"], "logs.log"))
    assert os.path.exists(
        os.path.join(cfg["output_directory"], "validation_predictions.csv")
    )