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
|
import os, sys
from argparse import ArgumentParser
from functools import partial
import torcharrow.pytorch as tap
import torchtext.functional as F
from benchmark.utils import Timer
from torchtext.datasets import DATASETS
sys.path.append(os.path.join(os.path.dirname(os.path.realpath(__file__)), "../examples"))
from data_pipeline.roberta_dataframe import RobertaTransformDataFrameNativeOps, RobertaTransformDataFrameUDF
from data_pipeline.roberta_datapipe import RobertaTransformDataPipe
def benchmark_roberta_datapipe(args):
print("********Running Benchmark using DataPipes**************\n")
batch_size = args.batch_size
dataset_name = args.dataset_name
columns = args.columns
# Instantiate transform
with Timer("Initialize Roberta Transform (for datapipe)"):
transform = RobertaTransformDataPipe()
with Timer("Initialize Pipeline"):
# Create SST2 datapipe and apply pre-processing
train_dp = DATASETS[dataset_name](split="train")
train_dp = train_dp.batch(batch_size).rows2columnar(columns)
# Apply text pre-processing
train_dp = train_dp.map(transform)
# convert to Tensor
train_dp = train_dp.map(partial(F.to_tensor, padding_value=1), input_col="tokens")
train_dp = train_dp.map(F.to_tensor, input_col="label")
with Timer("Execute Pipeline"):
list(train_dp)
def benchmark_roberta_dataframe(args, native_ops):
print("****************Running Benchmark using TorchArrow Dataframes*********************\n")
batch_size = args.batch_size
dataset_name = args.dataset_name
columns = args.columns
if native_ops:
append_text = "as native ops"
else:
append_text = "as UDF"
# Instantiate transform
with Timer("Initialize Roberta Transform (for DataFrame) {}".format(append_text)):
if native_ops:
transform = RobertaTransformDataFrameNativeOps()
else:
transform = RobertaTransformDataFrameUDF()
with Timer("Initialize Pipeline"):
# Create SST2 datapipe and apply pre-processing
train_dp = DATASETS[dataset_name](split="train")
# convert to DataFrame of size batches
# TODO: Figure out how to create DataFrame of larger size and create smaller batches
train_dp = train_dp.dataframe(columns=columns, dataframe_size=batch_size)
# Apply transformation on DataFrame
train_dp = train_dp.map(transform)
# Remove not required columns
train_dp = train_dp.map(lambda x: x.drop(["text"]))
# convert DataFrame to tensor (This will yield named tuple)
train_dp = train_dp.map(lambda x: x.to_tensor({"tokens": tap.PadSequence(padding_value=1)}))
with Timer("Execute Pipeline"):
list(train_dp)
if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument("--batch-size", default=32, type=int)
parser.add_argument("--dataset-name", default="SST2", type=str)
parser.add_argument("--columns", default=["text", "label"], nargs="+")
benchmark_roberta_datapipe(parser.parse_args())
benchmark_roberta_dataframe(parser.parse_args(), native_ops=False)
benchmark_roberta_dataframe(parser.parse_args(), native_ops=True)
|
