FTFT/run.py at main · nlpsoc/FTFT · GitHub

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
157
158
159
160
161
162
# Description: This file constructs the command line scripts for the project.
import os
import argparse
import yaml


def read_yaml(config_path):
    with open(config_path, "r") as file:
        config = yaml.safe_load(file)
    return config


def get_hf_name(model, size):
    if model == "debertav3":
        return f"microsoft/deberta-v3-{size}"
    elif model == "electra":
        return f"google/electra-{size}-discriminator"
    elif model == "roberta":
        return f"roberta-{size}"
    elif model == "bert":
        if size == "tiny":
            return "prajjwal1/bert-tiny"
        else:
            return f"bert-{size}-uncased"
    else:
        raise ValueError(f"Unsupported model: {model}")


def construct_erm_commands(config, rng_seed):
    script_command = f"accelerate launch --config_file {config['environment']['accelerate_config']} "\
        f"train.py --rng_seed {rng_seed}"

    data_args = {**config['data']['main']}
    data_command = ' '.join([f"--{k} {v}" for k, v in data_args.items() if v is not None])

    hf_name = get_hf_name(config['model']['main']['model'], config['model']['main']['size'])
    model_command = f"--model_name {hf_name}"

    optim_args = {'mode': 'erm', **config['optim']['main']}
    optim_command = ' '.join([f"--{k} {v}" for k, v in optim_args.items() if v is not None])

    save_dir = os.path.join(
        "ckpt", config['data']['main']['train_dataset_name'],
        config['log']['main']['project_name'],
        f"{config['model']['main']['model']}_{config['model']['main']['size']}",
        f"seed_{rng_seed}_steps_{config['optim']['main']['max_train_steps']}"
    )
    wandb_run_name = f"data_{config['data']['main']['train_dataset_name']}_"\
        f"model_{config['model']['main']['model']}_{config['model']['main']['size']}_"\
        f"seed_{rng_seed}_steps_{config['optim']['main']['max_train_steps']}"
    log_args = {'save_dir': save_dir, **{k: v for k, v in config['log']['main'].items() if k != 'project_name'}}
    log_command = ' '.join([f"--{k} {v}" for k, v in log_args.items() if k != 'use_wandb' and v is not None])
    if log_args['use_wandb']:
        log_command += " --use_wandb"
        log_command += f' --wandb_project_name {config["log"]["main"]["project_name"]}'
        log_command += f" --wandb_run_name {wandb_run_name}"


    command = f"{script_command} {data_command} {model_command} {optim_command} {log_command}"

    return command


def construct_dm_commands(config, rng_seed):
    script_command = f"accelerate launch --config_file {config['environment']['accelerate_config']} "\
        f"train.py --rng_seed {rng_seed}"

    data_args = {**config['data']['main']}
    data_command = ' '.join([f"--{k} {v}" for k, v in data_args.items() if v is not None])

    hf_name = get_hf_name(config['model']['main']['model'], config['model']['main']['size'])
    model_command = f"--model_name {hf_name}"

    optim_args = {'mode': 'dm', **config['optim']['main']}
    optim_command = ' '.join([f"--{k} {v}" for k, v in optim_args.items() if v is not None])
    if config['optim']['main']['dm_filter_type'] != 'random':
        ref_dir = os.path.join(
            "ckpt", config['data']['main']['train_dataset_name'],
            config['log']['ref']['project_name'],
            f"{config['model']['ref']['model']}_{config['model']['ref']['size']}",
            f"seed_{rng_seed}_steps_{config['optim']['ref']['max_train_steps']}"
        )
        optim_command += f" --reference_run_dir {ref_dir}"


    if config['optim']['main']['dm_filter_type'] == 'random':
        save_dir = os.path.join(
            "ckpt", config['data']['main']['train_dataset_name'],
            config['log']['main']['project_name'],
            f"{config['model']['main']['model']}_{config['model']['main']['size']}",
            f"type_{config['optim']['main']['dm_filter_type']}_"\
            f"seed_{rng_seed}_steps_{config['optim']['main']['max_train_steps']}"
        )
        wandb_run_name = f"data_{config['data']['main']['train_dataset_name']}_"\
            f"model_{config['model']['main']['model']}_{config['model']['main']['size']}_"\
            f"type_{config['optim']['main']['dm_filter_type']}_"\
            f"seed_{rng_seed}_steps_{config['optim']['main']['max_train_steps']}"
    else:
        save_dir = os.path.join(
            "ckpt", config['data']['main']['train_dataset_name'],
            config['log']['main']['project_name'],
            f"{config['model']['main']['model']}_{config['model']['main']['size']}",
            f"ref_{config['model']['ref']['model']}_{config['model']['ref']['size']}_"\
            f"steps_{config['optim']['ref']['max_train_steps']}_"\
            f"type_{config['optim']['main']['dm_filter_type']}_"\
            f"seed_{rng_seed}_steps_{config['optim']['main']['max_train_steps']}"
        )
        wandb_run_name = f"data_{config['data']['main']['train_dataset_name']}_"\
            f"model_{config['model']['main']['model']}_{config['model']['main']['size']}_"\
            f"ref_{config['model']['ref']['model']}_{config['model']['ref']['size']}_"\
            f"steps_{config['optim']['ref']['max_train_steps']}_"\
            f"type_{config['optim']['main']['dm_filter_type']}_"\
            f"seed_{rng_seed}_steps_{config['optim']['main']['max_train_steps']}"

    log_args = {'save_dir': save_dir, **{k: v for k, v in config['log']['main'].items() if k != 'project_name'}}
    log_command = ' '.join([f"--{k} {v}" for k, v in log_args.items() if k != 'use_wandb' and v is not None])
    if log_args['use_wandb']:
        log_command += " --use_wandb"
        log_command += f' --wandb_project_name {config["log"]["main"]["project_name"]}'
        log_command += f" --wandb_run_name {wandb_run_name}"

    log_command += ' --save_last'

    command = f"{script_command} {data_command} {model_command} {optim_command} {log_command}"
    return command


def construct_train_commands(config):
    if config["mode"] == "dm":
        train_command_func = construct_dm_commands
    elif config["mode"] == "erm":
        train_command_func = construct_erm_commands
    else:
        raise ValueError(f"Unsupported mode: {config['mode']}")

    train_commands = ['#!/bin/bash -x\n']
    for rng_seed in config["rng_seeds"]:
        train_command = train_command_func(config, rng_seed)
        train_commands.append(train_command)
    return "\n".join(train_commands)


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--config_path", type=str)
    args = parser.parse_args()

    config_path = args.config_path
    # read the YAML file
    config = read_yaml(config_path)
    # Control training duration by setting `max_train_steps`
    assert config['optim']['main']['max_train_steps'] is not None

    commands = construct_train_commands(config)
    output_path = config_path.replace("yaml", "sh")
    with open(output_path, "w") as file:
        file.write(commands)
    print(f"Output bash file: {output_path}")


if __name__ == "__main__":
    main()