Commit
•
0d03fdb
1
Parent(s):
f45cb5a
Support streaming XGLUE dataset (#4249)
Browse files* Support streaming XGLUE dataset
* Fix dataset card
* Fix dataset card by adding task ID for ntg config
Commit from https://github.com/huggingface/datasets/commit/e74d69c1d41dd320e77ca7244c624592f1a9fa3d
README.md
CHANGED
@@ -262,7 +262,8 @@ task_ids:
|
|
262 |
- topic-classification
|
263 |
ner:
|
264 |
- named-entity-recognition
|
265 |
-
ntg:
|
|
|
266 |
paws-x:
|
267 |
- text-classification-other-paraphrase-identification
|
268 |
pos:
|
@@ -284,6 +285,7 @@ pretty_name: XGLUE
|
|
284 |
# Dataset Card for XGLUE
|
285 |
|
286 |
## Table of Contents
|
|
|
287 |
- [Dataset Description](#dataset-description)
|
288 |
- [Dataset Summary](#dataset-summary)
|
289 |
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
@@ -323,11 +325,28 @@ The following table shows which languages are present as validation and test dat
|
|
323 |
|
324 |
Therefore, for each config, a cross-lingual pre-trained model should be fine-tuned on the English training data, and evaluated on for all languages.
|
325 |
|
326 |
-
### Leaderboards
|
327 |
|
328 |
The XGLUE leaderboard can be found on the [homepage](https://microsoft.github.io/XGLUE/) and
|
329 |
consits of a XGLUE-Understanding Score (the average of the tasks `ner`, `pos`, `mlqa`, `nc`, `xnli`, `paws-x`, `qadsm`, `wpr`, `qam`) and a XGLUE-Generation Score (the average of the tasks `qg`, `ntg`).
|
330 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
331 |
## Dataset Structure
|
332 |
|
333 |
### Data Instances
|
@@ -720,12 +739,6 @@ The following table shows the number of data samples/number of rows for each spl
|
|
720 |
|----|-----:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|
|
721 |
|xnli|392702| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010|
|
722 |
|
723 |
-
The following table shows the number of data samples/number of rows for each split in mlqa.
|
724 |
-
|
725 |
-
| |train|validation.en|validation.de|validation.ar|validation.es|validation.hi|validation.vi|validation.zh|test.en|test.de|test.ar|test.es|test.hi|test.vi|test.zh|
|
726 |
-
|----|----:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------:|------:|------:|------:|------:|------:|------:|
|
727 |
-
|mlqa|87599| 1148| 512| 517| 500| 507| 511| 504| 11590| 4517| 5335| 5253| 4918| 5495| 5137|
|
728 |
-
|
729 |
|
730 |
#### nc
|
731 |
|
|
|
262 |
- topic-classification
|
263 |
ner:
|
264 |
- named-entity-recognition
|
265 |
+
ntg:
|
266 |
+
- news-articles-headline-generation
|
267 |
paws-x:
|
268 |
- text-classification-other-paraphrase-identification
|
269 |
pos:
|
|
|
285 |
# Dataset Card for XGLUE
|
286 |
|
287 |
## Table of Contents
|
288 |
+
- [Table of Contents](#table-of-contents)
|
289 |
- [Dataset Description](#dataset-description)
|
290 |
- [Dataset Summary](#dataset-summary)
|
291 |
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
|
|
325 |
|
326 |
Therefore, for each config, a cross-lingual pre-trained model should be fine-tuned on the English training data, and evaluated on for all languages.
|
327 |
|
328 |
+
### Supported Tasks and Leaderboards
|
329 |
|
330 |
The XGLUE leaderboard can be found on the [homepage](https://microsoft.github.io/XGLUE/) and
|
331 |
consits of a XGLUE-Understanding Score (the average of the tasks `ner`, `pos`, `mlqa`, `nc`, `xnli`, `paws-x`, `qadsm`, `wpr`, `qam`) and a XGLUE-Generation Score (the average of the tasks `qg`, `ntg`).
|
332 |
|
333 |
+
### Languages
|
334 |
+
|
335 |
+
For all tasks (configurations), the "train" split is in English (`en`).
|
336 |
+
|
337 |
+
For each task, the "validation" and "test" splits are present in these languages:
|
338 |
+
- ner: `en`, `de`, `es`, `nl`
|
339 |
+
- pos: `en`, `de`, `es`, `nl`, `bg`, `el`, `fr`, `pl`, `tr`, `vi`, `zh`, `ur`, `hi`, `it`, `ar`, `ru`, `th`
|
340 |
+
- mlqa: `en`, `de`, `ar`, `es`, `hi`, `vi`, `zh`
|
341 |
+
- nc: `en`, `de`, `es`, `fr`, `ru`
|
342 |
+
- xnli: `en`, `ar`, `bg`, `de`, `el`, `es`, `fr`, `hi`, `ru`, `sw`, `th`, `tr`, `ur`, `vi`, `zh`
|
343 |
+
- paws-x: `en`, `de`, `es`, `fr`
|
344 |
+
- qadsm: `en`, `de`, `fr`
|
345 |
+
- wpr: `en`, `de`, `es`, `fr`, `it`, `pt`, `zh`
|
346 |
+
- qam: `en`, `de`, `fr`
|
347 |
+
- qg: `en`, `de`, `es`, `fr`, `it`, `pt`
|
348 |
+
- ntg: `en`, `de`, `es`, `fr`, `ru`
|
349 |
+
|
350 |
## Dataset Structure
|
351 |
|
352 |
### Data Instances
|
|
|
739 |
|----|-----:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|------:|
|
740 |
|xnli|392702| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 2490| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010| 5010|
|
741 |
|
|
|
|
|
|
|
|
|
|
|
|
|
742 |
|
743 |
#### nc
|
744 |
|
xglue.py
CHANGED
@@ -18,7 +18,6 @@
|
|
18 |
|
19 |
|
20 |
import json
|
21 |
-
import os
|
22 |
import textwrap
|
23 |
|
24 |
import datasets
|
@@ -75,15 +74,15 @@ _LANGUAGES = {
|
|
75 |
|
76 |
_PATHS = {
|
77 |
"mlqa": {
|
78 |
-
"train":
|
79 |
-
"dev":
|
80 |
-
"test":
|
81 |
},
|
82 |
"xnli": {"train": "multinli.train.en.tsv", "dev": "{}.dev", "test": "{}.test"},
|
83 |
"paws-x": {
|
84 |
-
"train":
|
85 |
-
"dev":
|
86 |
-
"test":
|
87 |
},
|
88 |
}
|
89 |
for name in ["ner", "pos"]:
|
@@ -473,8 +472,8 @@ Portuguese. BLEU-4 score should be used as the metric.
|
|
473 |
)
|
474 |
|
475 |
def _split_generators(self, dl_manager):
|
476 |
-
|
477 |
-
data_folder =
|
478 |
name = self.config.name
|
479 |
|
480 |
languages = _LANGUAGES[name]
|
@@ -482,14 +481,19 @@ Portuguese. BLEU-4 score should be used as the metric.
|
|
482 |
[
|
483 |
datasets.SplitGenerator(
|
484 |
name=datasets.Split.TRAIN,
|
485 |
-
gen_kwargs={
|
|
|
|
|
|
|
|
|
486 |
),
|
487 |
]
|
488 |
+ [
|
489 |
datasets.SplitGenerator(
|
490 |
name=datasets.Split(f"validation.{lang}"),
|
491 |
gen_kwargs={
|
492 |
-
"
|
|
|
493 |
"split": "dev",
|
494 |
},
|
495 |
)
|
@@ -499,7 +503,8 @@ Portuguese. BLEU-4 score should be used as the metric.
|
|
499 |
datasets.SplitGenerator(
|
500 |
name=datasets.Split(f"test.{lang}"),
|
501 |
gen_kwargs={
|
502 |
-
"
|
|
|
503 |
"split": "test",
|
504 |
},
|
505 |
)
|
@@ -507,68 +512,73 @@ Portuguese. BLEU-4 score should be used as the metric.
|
|
507 |
]
|
508 |
)
|
509 |
|
510 |
-
def _generate_examples(self,
|
511 |
keys = list(self._info().features.keys())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
512 |
|
513 |
-
|
514 |
-
|
515 |
-
|
516 |
-
|
517 |
-
for example in examples["paragraphs"]:
|
518 |
-
context = example["context"]
|
519 |
-
for qa in example["qas"]:
|
520 |
-
question = qa["question"]
|
521 |
-
id_ = qa["id"]
|
522 |
-
answers = qa["answers"]
|
523 |
-
answers_start = [answer["answer_start"] for answer in answers]
|
524 |
-
answers_text = [answer["text"] for answer in answers]
|
525 |
-
yield id_, {
|
526 |
-
"context": context,
|
527 |
-
"question": question,
|
528 |
-
"answers": {"answer_start": answers_start, "text": answers_text},
|
529 |
-
}
|
530 |
-
elif self.config.name in ["ner", "pos"]:
|
531 |
-
words = []
|
532 |
-
result = []
|
533 |
-
idx = -1
|
534 |
-
with open(data_file, encoding="utf-8") as f:
|
535 |
-
for line in f:
|
536 |
-
if line.strip() == "":
|
537 |
-
if len(words) > 0:
|
538 |
-
out_dict = {keys[0]: words, keys[1]: result}
|
539 |
-
words = []
|
540 |
-
result = []
|
541 |
-
idx += 1
|
542 |
-
yield idx, out_dict
|
543 |
-
else:
|
544 |
-
splits = line.strip().split(" ")
|
545 |
-
words.append(splits[0])
|
546 |
-
result.append(splits[1])
|
547 |
-
elif self.config.name in ["ntg", "qg"]:
|
548 |
-
with open(data_file + ".src." + split, encoding="utf-8") as src_f, open(
|
549 |
-
data_file + ".tgt." + split, encoding="utf-8"
|
550 |
-
) as tgt_f:
|
551 |
-
for idx, (src_line, tgt_line) in enumerate(zip(src_f, tgt_f)):
|
552 |
-
yield idx, {keys[0]: src_line.strip(), keys[1]: tgt_line.strip()}
|
553 |
-
else:
|
554 |
-
_process_dict = {
|
555 |
-
"paws-x": {"0": "different", "1": "same"},
|
556 |
-
"xnli": {"contradictory": "contradiction"},
|
557 |
-
"qam": {"0": "False", "1": "True"},
|
558 |
-
"wpr": {"0": "Bad", "1": "Fair", "2": "Good", "3": "Excellent", "4": "Perfect"},
|
559 |
-
}
|
560 |
-
|
561 |
-
def _process(value):
|
562 |
-
if self.config.name in _process_dict and value in _process_dict[self.config.name]:
|
563 |
-
return _process_dict[self.config.name][value]
|
564 |
-
return value
|
565 |
|
566 |
-
|
567 |
-
|
568 |
-
|
569 |
-
|
570 |
-
|
571 |
-
|
572 |
-
|
573 |
-
|
574 |
-
|
|
|
|
18 |
|
19 |
|
20 |
import json
|
|
|
21 |
import textwrap
|
22 |
|
23 |
import datasets
|
|
|
74 |
|
75 |
_PATHS = {
|
76 |
"mlqa": {
|
77 |
+
"train": "squad1.1/train-v1.1.json",
|
78 |
+
"dev": "MLQA_V1/dev/dev-context-{0}-question-{0}.json",
|
79 |
+
"test": "MLQA_V1/test/test-context-{0}-question-{0}.json",
|
80 |
},
|
81 |
"xnli": {"train": "multinli.train.en.tsv", "dev": "{}.dev", "test": "{}.test"},
|
82 |
"paws-x": {
|
83 |
+
"train": "en/train.tsv",
|
84 |
+
"dev": "{}/dev_2k.tsv",
|
85 |
+
"test": "{}/test_2k.tsv",
|
86 |
},
|
87 |
}
|
88 |
for name in ["ner", "pos"]:
|
|
|
472 |
)
|
473 |
|
474 |
def _split_generators(self, dl_manager):
|
475 |
+
archive = dl_manager.download(_XGLUE_ALL_DATA)
|
476 |
+
data_folder = f"xglue_full_dataset/{self.config.data_dir}"
|
477 |
name = self.config.name
|
478 |
|
479 |
languages = _LANGUAGES[name]
|
|
|
481 |
[
|
482 |
datasets.SplitGenerator(
|
483 |
name=datasets.Split.TRAIN,
|
484 |
+
gen_kwargs={
|
485 |
+
"archive": dl_manager.iter_archive(archive),
|
486 |
+
"data_path": f"{data_folder}/{_PATHS[name]['train']}",
|
487 |
+
"split": "train",
|
488 |
+
},
|
489 |
),
|
490 |
]
|
491 |
+ [
|
492 |
datasets.SplitGenerator(
|
493 |
name=datasets.Split(f"validation.{lang}"),
|
494 |
gen_kwargs={
|
495 |
+
"archive": dl_manager.iter_archive(archive),
|
496 |
+
"data_path": f"{data_folder}/{_PATHS[name]['dev'].format(lang)}",
|
497 |
"split": "dev",
|
498 |
},
|
499 |
)
|
|
|
503 |
datasets.SplitGenerator(
|
504 |
name=datasets.Split(f"test.{lang}"),
|
505 |
gen_kwargs={
|
506 |
+
"archive": dl_manager.iter_archive(archive),
|
507 |
+
"data_path": f"{data_folder}/{_PATHS[name]['test'].format(lang)}",
|
508 |
"split": "test",
|
509 |
},
|
510 |
)
|
|
|
512 |
]
|
513 |
)
|
514 |
|
515 |
+
def _generate_examples(self, archive, data_path, split=None):
|
516 |
keys = list(self._info().features.keys())
|
517 |
+
src_f = tgt_f = None
|
518 |
+
for path, file in archive:
|
519 |
+
if self.config.name == "mlqa":
|
520 |
+
if path == data_path:
|
521 |
+
data = json.load(file)
|
522 |
+
for examples in data["data"]:
|
523 |
+
for example in examples["paragraphs"]:
|
524 |
+
context = example["context"]
|
525 |
+
for qa in example["qas"]:
|
526 |
+
question = qa["question"]
|
527 |
+
id_ = qa["id"]
|
528 |
+
answers = qa["answers"]
|
529 |
+
answers_start = [answer["answer_start"] for answer in answers]
|
530 |
+
answers_text = [answer["text"] for answer in answers]
|
531 |
+
yield id_, {
|
532 |
+
"context": context,
|
533 |
+
"question": question,
|
534 |
+
"answers": {"answer_start": answers_start, "text": answers_text},
|
535 |
+
}
|
536 |
+
elif self.config.name in ["ner", "pos"]:
|
537 |
+
if path == data_path:
|
538 |
+
words = []
|
539 |
+
result = []
|
540 |
+
idx = -1
|
541 |
+
for line in file:
|
542 |
+
line = line.decode("utf-8")
|
543 |
+
if line.strip() == "":
|
544 |
+
if len(words) > 0:
|
545 |
+
out_dict = {keys[0]: words, keys[1]: result}
|
546 |
+
words = []
|
547 |
+
result = []
|
548 |
+
idx += 1
|
549 |
+
yield idx, out_dict
|
550 |
+
else:
|
551 |
+
splits = line.strip().split(" ")
|
552 |
+
words.append(splits[0])
|
553 |
+
result.append(splits[1])
|
554 |
+
elif self.config.name in ["ntg", "qg"]:
|
555 |
+
if path == data_path + ".src." + split:
|
556 |
+
src_f = [line.decode("utf-8") for line in file]
|
557 |
+
elif path == data_path + ".tgt." + split:
|
558 |
+
tgt_f = [line.decode("utf-8") for line in file]
|
559 |
+
if src_f and tgt_f:
|
560 |
+
for idx, (src_line, tgt_line) in enumerate(zip(src_f, tgt_f)):
|
561 |
+
yield idx, {keys[0]: src_line.strip(), keys[1]: tgt_line.strip()}
|
562 |
+
else:
|
563 |
+
_process_dict = {
|
564 |
+
"paws-x": {"0": "different", "1": "same"},
|
565 |
+
"xnli": {"contradictory": "contradiction"},
|
566 |
+
"qam": {"0": "False", "1": "True"},
|
567 |
+
"wpr": {"0": "Bad", "1": "Fair", "2": "Good", "3": "Excellent", "4": "Perfect"},
|
568 |
+
}
|
569 |
|
570 |
+
def _process(value):
|
571 |
+
if self.config.name in _process_dict and value in _process_dict[self.config.name]:
|
572 |
+
return _process_dict[self.config.name][value]
|
573 |
+
return value
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
574 |
|
575 |
+
if path == data_path:
|
576 |
+
for idx, line in enumerate(file):
|
577 |
+
line = line.decode("utf-8")
|
578 |
+
if data_path.split(".")[-1] == "tsv" and idx == 0:
|
579 |
+
continue
|
580 |
+
items = line.strip().split("\t")
|
581 |
+
yield idx, {
|
582 |
+
key: _process(value)
|
583 |
+
for key, value in zip(keys, items[1:] if self.config.name == "paws-x" else items)
|
584 |
+
}
|