update distil_cnndm.py
Browse files- distil_cnndm.py +23 -11
distil_cnndm.py
CHANGED
@@ -18,8 +18,11 @@ There are two features:
|
|
18 |
- highlights: joined text of highlights with <s> and </s> around each
|
19 |
highlight, which is the target summary
|
20 |
|
21 |
-
The
|
22 |
-
|
|
|
|
|
|
|
23 |
https://github.com/huggingface/transformers/blob/main/examples/research_projects/seq2seq-distillation/precomputed_pseudo_labels.md.
|
24 |
"""
|
25 |
|
@@ -47,18 +50,18 @@ _SUPPORTED_VERSIONS = [
|
|
47 |
_DEFAULT_VERSION = datasets.Version("2.0.0", "Using cased version.")
|
48 |
|
49 |
|
50 |
-
class
|
51 |
-
"""BuilderConfig for
|
52 |
|
53 |
def __init__(self, **kwargs):
|
54 |
super().__init__(**kwargs)
|
55 |
|
56 |
|
57 |
-
class
|
58 |
"""Distilled CNN/DailyMail non-anonymized summarization dataset."""
|
59 |
|
60 |
BUILDER_CONFIGS = [
|
61 |
-
|
62 |
for version in _SUPPORTED_VERSIONS
|
63 |
]
|
64 |
|
@@ -83,17 +86,21 @@ class DistillCNNDM(datasets.GeneratorBasedBuilder):
|
|
83 |
|
84 |
Although dl_manager is not used, we still need to keep it.
|
85 |
"""
|
86 |
-
dl_paths = dl_manager.download(_DL_URLS)
|
87 |
if self.config.version == "1.0.0":
|
|
|
88 |
return [
|
89 |
datasets.SplitGenerator(
|
90 |
name=split,
|
91 |
gen_kwargs={
|
92 |
"src_path": os.path.join(
|
93 |
-
|
|
|
|
|
94 |
),
|
95 |
"tgt_path": os.path.join(
|
96 |
-
|
|
|
|
|
97 |
),
|
98 |
"num_examples": _NUM_EXAMPLES[_SPLIT_MAP[split]],
|
99 |
},
|
@@ -105,15 +112,20 @@ class DistillCNNDM(datasets.GeneratorBasedBuilder):
|
|
105 |
]
|
106 |
]
|
107 |
elif self.config.version == "2.0.0":
|
|
|
108 |
return [
|
109 |
datasets.SplitGenerator(
|
110 |
name=datasets.Split.TRAIN,
|
111 |
gen_kwargs={
|
112 |
"src_path": os.path.join(
|
113 |
-
|
|
|
|
|
114 |
),
|
115 |
"tgt_path": os.path.join(
|
116 |
-
|
|
|
|
|
117 |
),
|
118 |
"num_examples": 287112,
|
119 |
},
|
|
|
18 |
- highlights: joined text of highlights with <s> and </s> around each
|
19 |
highlight, which is the target summary
|
20 |
|
21 |
+
The pseudo labels are generated by running
|
22 |
+
1. facebook/bart-large-cnn on the CNN/DailyMail dataset, or
|
23 |
+
2. sshleifer/pegasus-cnn-ft-v2 on the CNN/DailyMail dataset.
|
24 |
+
|
25 |
+
The files used here is directly downloaded from
|
26 |
https://github.com/huggingface/transformers/blob/main/examples/research_projects/seq2seq-distillation/precomputed_pseudo_labels.md.
|
27 |
"""
|
28 |
|
|
|
50 |
_DEFAULT_VERSION = datasets.Version("2.0.0", "Using cased version.")
|
51 |
|
52 |
|
53 |
+
class DistilCNNDMConfig(datasets.BuilderConfig):
|
54 |
+
"""BuilderConfig for DistilCNNDM."""
|
55 |
|
56 |
def __init__(self, **kwargs):
|
57 |
super().__init__(**kwargs)
|
58 |
|
59 |
|
60 |
+
class DistilCNNDM(datasets.GeneratorBasedBuilder):
|
61 |
"""Distilled CNN/DailyMail non-anonymized summarization dataset."""
|
62 |
|
63 |
BUILDER_CONFIGS = [
|
64 |
+
DistilCNNDMConfig(name=str(version), description="Plain text", version=version)
|
65 |
for version in _SUPPORTED_VERSIONS
|
66 |
]
|
67 |
|
|
|
86 |
|
87 |
Although dl_manager is not used, we still need to keep it.
|
88 |
"""
|
|
|
89 |
if self.config.version == "1.0.0":
|
90 |
+
extracted_path = dl_manager.download_and_extract(_DL_URLS["cnn_bart_pl"])
|
91 |
return [
|
92 |
datasets.SplitGenerator(
|
93 |
name=split,
|
94 |
gen_kwargs={
|
95 |
"src_path": os.path.join(
|
96 |
+
extracted_path,
|
97 |
+
"cnn_bart_pl",
|
98 |
+
f"{_SPLIT_MAP[split]}.source",
|
99 |
),
|
100 |
"tgt_path": os.path.join(
|
101 |
+
extracted_path,
|
102 |
+
"cnn_bart_pl",
|
103 |
+
f"{_SPLIT_MAP[split]}.target",
|
104 |
),
|
105 |
"num_examples": _NUM_EXAMPLES[_SPLIT_MAP[split]],
|
106 |
},
|
|
|
112 |
]
|
113 |
]
|
114 |
elif self.config.version == "2.0.0":
|
115 |
+
extracted_path = dl_manager.download_and_extract(_DL_URLS["cnn_pegasus_pl"])
|
116 |
return [
|
117 |
datasets.SplitGenerator(
|
118 |
name=datasets.Split.TRAIN,
|
119 |
gen_kwargs={
|
120 |
"src_path": os.path.join(
|
121 |
+
extracted_path,
|
122 |
+
"pegasus_cnn_cnn_pls",
|
123 |
+
"train.source",
|
124 |
),
|
125 |
"tgt_path": os.path.join(
|
126 |
+
extracted_path,
|
127 |
+
"pegasus_cnn_cnn_pls",
|
128 |
+
"train.target",
|
129 |
),
|
130 |
"num_examples": 287112,
|
131 |
},
|