Bokmaal version of xsum
Browse files- nob/dataset_dict.json +1 -0
- nob/nob_test.json.tar.gz +3 -0
- nob/nob_train.json.tar.gz +3 -0
- nob/nob_validation.json.tar.gz +3 -0
- nob/test/data-00000-of-00001.arrow +3 -0
- nob/test/dataset_info.json +65 -0
- nob/test/state.json +13 -0
- nob/train/data-00000-of-00001.arrow +3 -0
- nob/train/dataset_info.json +65 -0
- nob/train/state.json +13 -0
- nob/validation/data-00000-of-00001.arrow +3 -0
- nob/validation/dataset_info.json +65 -0
- nob/validation/state.json +13 -0
- translator.py +127 -0
nob/dataset_dict.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"splits": ["train", "validation", "test"]}
|
nob/nob_test.json.tar.gz
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e53b90d8579baebfb94b1cfea5695a75633d71c38bac43ad3d5c06586e2204e7
|
3 |
+
size 3363690
|
nob/nob_train.json.tar.gz
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c533b8c8596718c7f3954c0578562efa314e141b3880cb43145d6c7e3ef11053
|
3 |
+
size 60171284
|
nob/nob_validation.json.tar.gz
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:51ecfd130e61fb1860448427b83a18a64ca6f779f1785517366faeb8d5d790de
|
3 |
+
size 3280742
|
nob/test/data-00000-of-00001.arrow
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2b6b9eb616116b75a8c23eabde80b9ab21227c7f3f09738e9461548a13670c06
|
3 |
+
size 48076152
|
nob/test/dataset_info.json
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"builder_name": "xsum",
|
3 |
+
"citation": "\n@article{Narayan2018DontGM,\n title={Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization},\n author={Shashi Narayan and Shay B. Cohen and Mirella Lapata},\n journal={ArXiv},\n year={2018},\n volume={abs/1808.08745}\n}\n",
|
4 |
+
"config_name": "default",
|
5 |
+
"dataset_size": 532255381,
|
6 |
+
"description": "\nExtreme Summarization (XSum) Dataset.\n\nThere are three features:\n - document: Input news article.\n - summary: One sentence summary of the article.\n - id: BBC ID of the article.\n\n",
|
7 |
+
"download_checksums": {
|
8 |
+
"data/XSUM-EMNLP18-Summary-Data-Original.tar.gz": {
|
9 |
+
"num_bytes": 254582292,
|
10 |
+
"checksum": "10b48aa187fc9c904b30f76ca97e2da0de8d3a1238acc26acadef93e2001af90"
|
11 |
+
},
|
12 |
+
"https://raw.githubusercontent.com/EdinburghNLP/XSum/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json": {
|
13 |
+
"num_bytes": 2720574,
|
14 |
+
"checksum": "9c0c5d8f048a90bd68b19a34e4c30577ed270d3247b2119fa06a04ef46292068"
|
15 |
+
}
|
16 |
+
},
|
17 |
+
"download_size": 257302866,
|
18 |
+
"features": {
|
19 |
+
"document": {
|
20 |
+
"dtype": "string",
|
21 |
+
"_type": "Value"
|
22 |
+
},
|
23 |
+
"summary": {
|
24 |
+
"dtype": "string",
|
25 |
+
"_type": "Value"
|
26 |
+
},
|
27 |
+
"id": {
|
28 |
+
"dtype": "string",
|
29 |
+
"_type": "Value"
|
30 |
+
}
|
31 |
+
},
|
32 |
+
"homepage": "https://github.com/EdinburghNLP/XSum/tree/master/XSum-Dataset",
|
33 |
+
"license": "",
|
34 |
+
"size_in_bytes": 789558247,
|
35 |
+
"splits": {
|
36 |
+
"train": {
|
37 |
+
"name": "train",
|
38 |
+
"num_bytes": 479206363,
|
39 |
+
"num_examples": 204045,
|
40 |
+
"dataset_name": "xsum"
|
41 |
+
},
|
42 |
+
"validation": {
|
43 |
+
"name": "validation",
|
44 |
+
"num_bytes": 26292877,
|
45 |
+
"num_examples": 11332,
|
46 |
+
"dataset_name": "xsum"
|
47 |
+
},
|
48 |
+
"test": {
|
49 |
+
"name": "test",
|
50 |
+
"num_bytes": 26756141,
|
51 |
+
"num_examples": 11334,
|
52 |
+
"dataset_name": "xsum"
|
53 |
+
}
|
54 |
+
},
|
55 |
+
"supervised_keys": {
|
56 |
+
"input": "document",
|
57 |
+
"output": "summary"
|
58 |
+
},
|
59 |
+
"version": {
|
60 |
+
"version_str": "1.2.0",
|
61 |
+
"major": 1,
|
62 |
+
"minor": 2,
|
63 |
+
"patch": 0
|
64 |
+
}
|
65 |
+
}
|
nob/test/state.json
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_data_files": [
|
3 |
+
{
|
4 |
+
"filename": "data-00000-of-00001.arrow"
|
5 |
+
}
|
6 |
+
],
|
7 |
+
"_fingerprint": "0cd0deb949ec246b",
|
8 |
+
"_format_columns": null,
|
9 |
+
"_format_kwargs": {},
|
10 |
+
"_format_type": null,
|
11 |
+
"_output_all_columns": false,
|
12 |
+
"_split": "test"
|
13 |
+
}
|
nob/train/data-00000-of-00001.arrow
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2e799e9bdddcdf2997088f16298df4614f8adb6068f634d52b576b11a66d3e04
|
3 |
+
size 833308256
|
nob/train/dataset_info.json
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"builder_name": "xsum",
|
3 |
+
"citation": "\n@article{Narayan2018DontGM,\n title={Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization},\n author={Shashi Narayan and Shay B. Cohen and Mirella Lapata},\n journal={ArXiv},\n year={2018},\n volume={abs/1808.08745}\n}\n",
|
4 |
+
"config_name": "default",
|
5 |
+
"dataset_size": 532255381,
|
6 |
+
"description": "\nExtreme Summarization (XSum) Dataset.\n\nThere are three features:\n - document: Input news article.\n - summary: One sentence summary of the article.\n - id: BBC ID of the article.\n\n",
|
7 |
+
"download_checksums": {
|
8 |
+
"data/XSUM-EMNLP18-Summary-Data-Original.tar.gz": {
|
9 |
+
"num_bytes": 254582292,
|
10 |
+
"checksum": "10b48aa187fc9c904b30f76ca97e2da0de8d3a1238acc26acadef93e2001af90"
|
11 |
+
},
|
12 |
+
"https://raw.githubusercontent.com/EdinburghNLP/XSum/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json": {
|
13 |
+
"num_bytes": 2720574,
|
14 |
+
"checksum": "9c0c5d8f048a90bd68b19a34e4c30577ed270d3247b2119fa06a04ef46292068"
|
15 |
+
}
|
16 |
+
},
|
17 |
+
"download_size": 257302866,
|
18 |
+
"features": {
|
19 |
+
"document": {
|
20 |
+
"dtype": "string",
|
21 |
+
"_type": "Value"
|
22 |
+
},
|
23 |
+
"summary": {
|
24 |
+
"dtype": "string",
|
25 |
+
"_type": "Value"
|
26 |
+
},
|
27 |
+
"id": {
|
28 |
+
"dtype": "string",
|
29 |
+
"_type": "Value"
|
30 |
+
}
|
31 |
+
},
|
32 |
+
"homepage": "https://github.com/EdinburghNLP/XSum/tree/master/XSum-Dataset",
|
33 |
+
"license": "",
|
34 |
+
"size_in_bytes": 789558247,
|
35 |
+
"splits": {
|
36 |
+
"train": {
|
37 |
+
"name": "train",
|
38 |
+
"num_bytes": 479206363,
|
39 |
+
"num_examples": 204045,
|
40 |
+
"dataset_name": "xsum"
|
41 |
+
},
|
42 |
+
"validation": {
|
43 |
+
"name": "validation",
|
44 |
+
"num_bytes": 26292877,
|
45 |
+
"num_examples": 11332,
|
46 |
+
"dataset_name": "xsum"
|
47 |
+
},
|
48 |
+
"test": {
|
49 |
+
"name": "test",
|
50 |
+
"num_bytes": 26756141,
|
51 |
+
"num_examples": 11334,
|
52 |
+
"dataset_name": "xsum"
|
53 |
+
}
|
54 |
+
},
|
55 |
+
"supervised_keys": {
|
56 |
+
"input": "document",
|
57 |
+
"output": "summary"
|
58 |
+
},
|
59 |
+
"version": {
|
60 |
+
"version_str": "1.2.0",
|
61 |
+
"major": 1,
|
62 |
+
"minor": 2,
|
63 |
+
"patch": 0
|
64 |
+
}
|
65 |
+
}
|
nob/train/state.json
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_data_files": [
|
3 |
+
{
|
4 |
+
"filename": "data-00000-of-00001.arrow"
|
5 |
+
}
|
6 |
+
],
|
7 |
+
"_fingerprint": "09e18f5a398d83ac",
|
8 |
+
"_format_columns": null,
|
9 |
+
"_format_kwargs": {},
|
10 |
+
"_format_type": null,
|
11 |
+
"_output_all_columns": false,
|
12 |
+
"_split": "train"
|
13 |
+
}
|
nob/validation/data-00000-of-00001.arrow
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0dccddfe9fc0209ae45e4ccf9279b0330b8a480dd5d1819a2ba8136e3695a320
|
3 |
+
size 44194752
|
nob/validation/dataset_info.json
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"builder_name": "xsum",
|
3 |
+
"citation": "\n@article{Narayan2018DontGM,\n title={Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization},\n author={Shashi Narayan and Shay B. Cohen and Mirella Lapata},\n journal={ArXiv},\n year={2018},\n volume={abs/1808.08745}\n}\n",
|
4 |
+
"config_name": "default",
|
5 |
+
"dataset_size": 532255381,
|
6 |
+
"description": "\nExtreme Summarization (XSum) Dataset.\n\nThere are three features:\n - document: Input news article.\n - summary: One sentence summary of the article.\n - id: BBC ID of the article.\n\n",
|
7 |
+
"download_checksums": {
|
8 |
+
"data/XSUM-EMNLP18-Summary-Data-Original.tar.gz": {
|
9 |
+
"num_bytes": 254582292,
|
10 |
+
"checksum": "10b48aa187fc9c904b30f76ca97e2da0de8d3a1238acc26acadef93e2001af90"
|
11 |
+
},
|
12 |
+
"https://raw.githubusercontent.com/EdinburghNLP/XSum/master/XSum-Dataset/XSum-TRAINING-DEV-TEST-SPLIT-90-5-5.json": {
|
13 |
+
"num_bytes": 2720574,
|
14 |
+
"checksum": "9c0c5d8f048a90bd68b19a34e4c30577ed270d3247b2119fa06a04ef46292068"
|
15 |
+
}
|
16 |
+
},
|
17 |
+
"download_size": 257302866,
|
18 |
+
"features": {
|
19 |
+
"document": {
|
20 |
+
"dtype": "string",
|
21 |
+
"_type": "Value"
|
22 |
+
},
|
23 |
+
"summary": {
|
24 |
+
"dtype": "string",
|
25 |
+
"_type": "Value"
|
26 |
+
},
|
27 |
+
"id": {
|
28 |
+
"dtype": "string",
|
29 |
+
"_type": "Value"
|
30 |
+
}
|
31 |
+
},
|
32 |
+
"homepage": "https://github.com/EdinburghNLP/XSum/tree/master/XSum-Dataset",
|
33 |
+
"license": "",
|
34 |
+
"size_in_bytes": 789558247,
|
35 |
+
"splits": {
|
36 |
+
"train": {
|
37 |
+
"name": "train",
|
38 |
+
"num_bytes": 479206363,
|
39 |
+
"num_examples": 204045,
|
40 |
+
"dataset_name": "xsum"
|
41 |
+
},
|
42 |
+
"validation": {
|
43 |
+
"name": "validation",
|
44 |
+
"num_bytes": 26292877,
|
45 |
+
"num_examples": 11332,
|
46 |
+
"dataset_name": "xsum"
|
47 |
+
},
|
48 |
+
"test": {
|
49 |
+
"name": "test",
|
50 |
+
"num_bytes": 26756141,
|
51 |
+
"num_examples": 11334,
|
52 |
+
"dataset_name": "xsum"
|
53 |
+
}
|
54 |
+
},
|
55 |
+
"supervised_keys": {
|
56 |
+
"input": "document",
|
57 |
+
"output": "summary"
|
58 |
+
},
|
59 |
+
"version": {
|
60 |
+
"version_str": "1.2.0",
|
61 |
+
"major": 1,
|
62 |
+
"minor": 2,
|
63 |
+
"patch": 0
|
64 |
+
}
|
65 |
+
}
|
nob/validation/state.json
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_data_files": [
|
3 |
+
{
|
4 |
+
"filename": "data-00000-of-00001.arrow"
|
5 |
+
}
|
6 |
+
],
|
7 |
+
"_fingerprint": "3d58f12b55350560",
|
8 |
+
"_format_columns": null,
|
9 |
+
"_format_kwargs": {},
|
10 |
+
"_format_type": null,
|
11 |
+
"_output_all_columns": false,
|
12 |
+
"_split": "validation"
|
13 |
+
}
|
translator.py
ADDED
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import re
|
3 |
+
from functools import partial
|
4 |
+
from pathlib import Path
|
5 |
+
from typing import Optional, Union
|
6 |
+
|
7 |
+
import nltk
|
8 |
+
import torch
|
9 |
+
from datasets import load_dataset
|
10 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
11 |
+
|
12 |
+
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
13 |
+
|
14 |
+
|
15 |
+
def to_lang_code(texts, lang_code, model, tokenizer, max_words=500):
|
16 |
+
is_string = isinstance(texts, str)
|
17 |
+
if is_string:
|
18 |
+
texts = [texts]
|
19 |
+
batch_size = len(texts)
|
20 |
+
to_translate = []
|
21 |
+
merges = []
|
22 |
+
for index, text in enumerate(texts):
|
23 |
+
# Split in sentences if too long
|
24 |
+
merges.append(0)
|
25 |
+
if text.count(" ") > max_words:
|
26 |
+
sentences = nltk.sent_tokenize(text, "norwegian")
|
27 |
+
text_to_translate = ""
|
28 |
+
for sentence in sentences:
|
29 |
+
spaces = (text_to_translate + " " + sentence).count(" ")
|
30 |
+
if spaces >= max_words:
|
31 |
+
to_translate.append(text_to_translate.strip())
|
32 |
+
merges[-1] += 1
|
33 |
+
else:
|
34 |
+
text_to_translate += sentence + " "
|
35 |
+
else:
|
36 |
+
to_translate.append(text)
|
37 |
+
translated_texts = []
|
38 |
+
# Split in batches for translation
|
39 |
+
to_translate_batchs = [to_translate[i:i + batch_size] for i in range(0, len(to_translate), batch_size)]
|
40 |
+
for to_translate_batch in to_translate_batchs:
|
41 |
+
inputs = tokenizer(to_translate_batch, return_tensors="pt", padding=True, truncation=True).to(DEVICE)
|
42 |
+
translated_tokens = model.generate(
|
43 |
+
**inputs,
|
44 |
+
forced_bos_token_id=tokenizer.lang_code_to_id[lang_code],
|
45 |
+
max_length=int(len(inputs.tokens()) * 1.25) # 25% more tokens for the translation just in case
|
46 |
+
)
|
47 |
+
translated_texts += tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)
|
48 |
+
# Merge outputs properly
|
49 |
+
outputs = []
|
50 |
+
for merge in merges:
|
51 |
+
output = ""
|
52 |
+
if merge:
|
53 |
+
for i in range(len(outputs), len(outputs) + merge):
|
54 |
+
output += translated_texts[i] + " "
|
55 |
+
outputs.append(output.strip())
|
56 |
+
else:
|
57 |
+
outputs.append(translated_texts[len(outputs)].strip())
|
58 |
+
return outputs[0] if is_string else outputs
|
59 |
+
|
60 |
+
|
61 |
+
def main(
|
62 |
+
dataset_name: str,
|
63 |
+
dataset_columns: Union[list, tuple],
|
64 |
+
model_name: Optional[str]="facebook/nllb-200-3.3B", # "facebook/nllb-200-distilled-600M"
|
65 |
+
model_revision: Optional[str]=None,
|
66 |
+
dataset_splits: Union[list, tuple]=("test", "validation", "train"),
|
67 |
+
dataset_config: Optional[str]=None,
|
68 |
+
dataset_revision: Optional[str]=None,
|
69 |
+
source_lang: Optional[str]="eng_Latn",
|
70 |
+
target_langs: Optional[Union[list, tuple]]=("nob_Latn", "nno_Latn"),
|
71 |
+
batch_size: Optional[int]=24,
|
72 |
+
output_dir: Optional[Path]=Path("./"),
|
73 |
+
) -> None:
|
74 |
+
|
75 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(model_name, use_auth_token=True, torch_dtype=torch.float32)
|
76 |
+
model.to(DEVICE, torch.float32, True)
|
77 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
78 |
+
model_name, revision=model_revision, use_auth_token=True, src_lang=source_lang,
|
79 |
+
)
|
80 |
+
|
81 |
+
ds = load_dataset(dataset_name, name=dataset_config, revision=dataset_revision)
|
82 |
+
dss = {}
|
83 |
+
for lang_code in target_langs:
|
84 |
+
translate = partial(to_lang_code, lang_code=lang_code, model=model, tokenizer=tokenizer)
|
85 |
+
dss[lang_code] = ds.map(
|
86 |
+
lambda batch: {col: translate(batch[col]) for col in dataset_columns},
|
87 |
+
batched=True,
|
88 |
+
batch_size=batch_size,
|
89 |
+
desc=f"Translating to {lang_code}",
|
90 |
+
)
|
91 |
+
lang_code_short = re.split(r"[-_ /]", lang_code)[0]
|
92 |
+
dss[lang_code].save_to_disk(output_dir / lang_code_short, max_shard_size="1GB")
|
93 |
+
for split in dataset_splits:
|
94 |
+
json_filename = f"{lang_code_short}_{split}.json.tar.gz".lower()
|
95 |
+
dss[lang_code][split].to_pandas().to_json(
|
96 |
+
output_dir / lang_code_short / json_filename, orient='records', lines=True
|
97 |
+
)
|
98 |
+
|
99 |
+
|
100 |
+
|
101 |
+
if __name__ == "__main__":
|
102 |
+
parser = argparse.ArgumentParser(description="Translate datasets using Facebook's NLLB models")
|
103 |
+
parser.add_argument('dataset_name')
|
104 |
+
parser.add_argument('dataset_columns', help="Comma separated column names to translate")
|
105 |
+
parser.add_argument('--dataset_splits', default="test,validation,train", help="Comma separated splits to translate")
|
106 |
+
parser.add_argument('--dataset_config')
|
107 |
+
parser.add_argument('--dataset_revision')
|
108 |
+
parser.add_argument('--model_name', default="facebook/nllb-200-3.3B")
|
109 |
+
parser.add_argument('--model_revision')
|
110 |
+
parser.add_argument('--source_lang', default="eng_Latn")
|
111 |
+
parser.add_argument('--target_langs', default="nob_Latn,nno_Latn", help="Comma separated target languages to translate to")
|
112 |
+
parser.add_argument('--batch_size', '-bs', default=24, type=int, help='Number of inputs per batch for prediction')
|
113 |
+
parser.add_argument('--output_dir', '-o', default="./", type=str)
|
114 |
+
args = parser.parse_args()
|
115 |
+
main(
|
116 |
+
dataset_name=args.dataset_name,
|
117 |
+
dataset_columns=args.dataset_columns.split(","),
|
118 |
+
dataset_splits=args.dataset_splits.split(","),
|
119 |
+
dataset_config=args.dataset_config,
|
120 |
+
dataset_revision=args.dataset_revision,
|
121 |
+
model_name=args.model_name,
|
122 |
+
model_revision=args.model_revision,
|
123 |
+
source_lang=args.source_lang,
|
124 |
+
target_langs=args.target_langs.split(","),
|
125 |
+
batch_size=args.batch_size,
|
126 |
+
output_dir=Path(args.output_dir),
|
127 |
+
)
|