-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_multitask.py
197 lines (157 loc) · 6.64 KB
/
run_multitask.py
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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
import os
import argparse
import warnings
import logging
import wandb
import pytorch_lightning as pl
from transformers import AutoTokenizer
from saint.utils import py_io
from saint.utils.utils import set_seed
from saint.models.T5.t5_finetune import T5FineTuner, MetricsCallback
from saint.models.T5.t5_generate import T5AlignmentGenerator
from saint.dataset.multi_task_dataset import MultiTaskDataset
from saint.dataset.dataset import AlignmentGenerationDataset
from pytorch_lightning.loggers import WandbLogger
from pytorch_lightning.callbacks import RichProgressBar
wandb.init(
project="causal_scaffold_modeling",
name="casul_rewrite_1"
)
wandb_logger = WandbLogger(
project="causal_scaffold_modeling",
name="casul_rewrite_1"
)
warnings.filterwarnings('ignore')
logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S',
level=logging.INFO)
logger = logging.getLogger(__name__)
set_seed(42)
task_splits = {
"snli": ["snli"],
"anli": ["anli"],
"nlsat": ["nlsat"],
"logiqa": ["logiqa"],
"control": ["control"],
"analytic": ["analytic"],
"crowd_sourced": ["snli", "anli"],
"logic": ["nlsat", "logiqa"],
"complex": ["analytic", "control"]
}
def build_datasets(args):
tokenizer = AutoTokenizer.from_pretrained(args.model)
tasks = task_splits[args.task_split]
train_multi_data = MultiTaskDataset(
logger, data_path=args.data_dir,
tasks=tasks, data_split="train",
max_len_input=args.max_len_input,
max_len_output=args.max_len_output,
train_batch_size=args.train_batch_size,
val_batch_size=args.val_batch_size,
is_training=True
)
dev_multi_data = MultiTaskDataset(
logger, data_path=args.data_dir,
tasks=tasks, data_split="dev",
max_len_input=args.max_len_input,
max_len_output=args.max_len_output,
train_batch_size=args.train_batch_size,
val_batch_size=args.val_batch_size,
is_training=False
)
train_multi_data.load_dataset(tokenizer)
train_multi_data.load_dataloader()
dev_multi_data.load_dataset(tokenizer)
dev_multi_data.load_dataloader()
return train_multi_data, dev_multi_data
def train(model, train_params, checkpoint_pth):
trainer = pl.Trainer(**train_params)
logger.info("Training model ...")
trainer.fit(model)
logger.info("Saving model ...")
model.save_core_model()
logger.info(f"Model Saved at {checkpoint_pth}")
def evaluate(task_split, model, predict_dir):
t5_saint_generator = T5AlignmentGenerator(model)
datasets = task_splits[task_split]
for dataset in datasets:
nli_test_data = py_io.read_jsonl(f"data/{dataset}/dev.jsonl")
nli_output = t5_saint_generator.generate(nli_test_data)
evid_test_data = py_io.read_jsonl(f"data/{dataset}/test.jsonl")
evid_output = t5_saint_generator.generate(evid_test_data)
py_io.write_jsonl(nli_output, os.path.join(
predict_dir, "nli_pred.jsonl"))
py_io.write_jsonl(evid_output, os.path.join(
predict_dir, "evid_pred.jsonl"))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--data_dir", default="data")
parser.add_argument("--model", default="t5-base")
parser.add_argument("--task_split", default="causal_snli/causal_nli_entail")
parser.add_argument("--output_dir", default="runs")
parser.add_argument("--do_train", action='store_true')
parser.add_argument("--do_eval", action='store_true')
# Model parameters
parser.add_argument("--checkpoint", action='store_true')
parser.add_argument('--max_len_input', type=int, default=128)
parser.add_argument('--max_len_output', type=int, default=64)
# Training-related parameters
parser.add_argument("--train_batch_size", default=4, type=int,
help="Batch size per GPU/CPU for training.")
parser.add_argument("--val_batch_size", default=2, type=int,
help="Batch size per GPU/CPU for evaluation.")
parser.add_argument("--lr", default=3e-4, type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--warmup_proportion", default=0.01, type=float,
help="Weight decay if we apply some.")
parser.add_argument("--weight_decay", default=0.01, type=float,
help="Weight deay if we apply some.")
parser.add_argument("--adam_epsilon", default=1e-8, type=float,
help="Epsilon for Adam optimizer.")
parser.add_argument("--gradient_accumulation_steps", default=1, type=int,
help="gradient will be accumulated over this many steps.")
parser.add_argument("--max_epochs", default=10, type=float,
help="Total number of training epochs to perform.")
parser.add_argument("--warmup_steps", default=100, type=int,
help="Linear warmup over warmup_steps.")
parser.add_argument("--fp_16", action='store_true')
args = parser.parse_args()
logger.info(args)
pred_dir = os.path.join(args.output_dir, f"{args.task_split}-{args.model}")
checkpoint_callback = pl.callbacks.ModelCheckpoint(
dirpath=pred_dir,
filename="best-model",
monitor="val_loss",
mode="min",
save_top_k=1
)
metrics_callback = MetricsCallback()
train_params = dict(
gpus=1,
logger=wandb_logger,
log_every_n_steps=50,
max_epochs=20,
progress_bar_refresh_rate=10,
precision= 16 if args.fp_16 else 32,
# gradient_clip_val=args.max_grad_norm,
accumulate_grad_batches=args.gradient_accumulation_steps,
checkpoint_callback=True,
callbacks=[metrics_callback, checkpoint_callback, RichProgressBar()]
)
train_data = py_io.read_jsonl(
"./data/causal_snli/causal_snli_entail/train.jsonl")
val_data = py_io.read_jsonl("./data/causal_snli/causal_snli_entail/dev.jsonl")
tokenizer = AutoTokenizer.from_pretrained(args.model)
train_multi_data = AlignmentGenerationDataset(tokenizer, train_data)
dev_multi_data = AlignmentGenerationDataset(tokenizer, val_data)
train_multi_data.load_dataloader(train_multi_data, True, 4, 2)
dev_multi_data.load_dataloader(dev_multi_data, False, 4, 2)
#train_multi_data, dev_multi_data = build_datasets(args)
model = T5FineTuner(
args, train_dataset=train_multi_data,
val_dataset=dev_multi_data
)
if args.do_train:
train(model, train_params, pred_dir)
if args.do_eval:
evaluate(args.task_split, model, pred_dir)