Spaces:
Runtime error
Runtime error
Update Space (evaluate main: c447fc8e)
Browse files- frugalscore.py +13 -25
- requirements.txt +1 -1
frugalscore.py
CHANGED
@@ -13,9 +13,6 @@
|
|
13 |
# limitations under the License.
|
14 |
"""FrugalScore metric."""
|
15 |
|
16 |
-
from dataclasses import dataclass
|
17 |
-
from typing import Optional
|
18 |
-
|
19 |
import datasets
|
20 |
import torch
|
21 |
from transformers import AutoModelForSequenceClassification, AutoTokenizer, Trainer, TrainingArguments
|
@@ -57,28 +54,13 @@ Examples:
|
|
57 |
"""
|
58 |
|
59 |
|
60 |
-
@dataclass
|
61 |
-
class FRUGALSCOREConfig(evaluate.info.Config):
|
62 |
-
|
63 |
-
name: str = "default"
|
64 |
-
|
65 |
-
batch_size: int = 32
|
66 |
-
max_length: int = 128
|
67 |
-
device: Optional[str] = None
|
68 |
-
|
69 |
-
|
70 |
@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
|
71 |
class FRUGALSCORE(evaluate.Metric):
|
72 |
-
|
73 |
-
CONFIG_CLASS = FRUGALSCOREConfig
|
74 |
-
ALLOWED_CONFIG_NAMES = ["default"]
|
75 |
-
|
76 |
-
def _info(self, config):
|
77 |
return evaluate.MetricInfo(
|
78 |
description=_DESCRIPTION,
|
79 |
citation=_CITATION,
|
80 |
inputs_description=_KWARGS_DESCRIPTION,
|
81 |
-
config=config,
|
82 |
features=datasets.Features(
|
83 |
{
|
84 |
"predictions": datasets.Value("string"),
|
@@ -96,20 +78,26 @@ class FRUGALSCORE(evaluate.Metric):
|
|
96 |
self.model = AutoModelForSequenceClassification.from_pretrained(checkpoint)
|
97 |
self.tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
98 |
|
99 |
-
def _compute(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
"""Returns the scores"""
|
101 |
assert len(predictions) == len(
|
102 |
references
|
103 |
), "predictions and references should have the same number of sentences."
|
104 |
-
if
|
105 |
-
assert
|
106 |
-
device = self.config.device
|
107 |
else:
|
108 |
device = "gpu" if torch.cuda.is_available() else "cpu"
|
109 |
training_args = TrainingArguments(
|
110 |
"trainer",
|
111 |
fp16=(device == "gpu"),
|
112 |
-
per_device_eval_batch_size=
|
113 |
report_to="all",
|
114 |
no_cuda=(device == "cpu"),
|
115 |
log_level="warning",
|
@@ -119,7 +107,7 @@ class FRUGALSCORE(evaluate.Metric):
|
|
119 |
|
120 |
def tokenize_function(data):
|
121 |
return self.tokenizer(
|
122 |
-
data["sentence1"], data["sentence2"], max_length=
|
123 |
)
|
124 |
|
125 |
tokenized_datasets = raw_datasets.map(tokenize_function, batched=True)
|
|
|
13 |
# limitations under the License.
|
14 |
"""FrugalScore metric."""
|
15 |
|
|
|
|
|
|
|
16 |
import datasets
|
17 |
import torch
|
18 |
from transformers import AutoModelForSequenceClassification, AutoTokenizer, Trainer, TrainingArguments
|
|
|
54 |
"""
|
55 |
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
|
58 |
class FRUGALSCORE(evaluate.Metric):
|
59 |
+
def _info(self):
|
|
|
|
|
|
|
|
|
60 |
return evaluate.MetricInfo(
|
61 |
description=_DESCRIPTION,
|
62 |
citation=_CITATION,
|
63 |
inputs_description=_KWARGS_DESCRIPTION,
|
|
|
64 |
features=datasets.Features(
|
65 |
{
|
66 |
"predictions": datasets.Value("string"),
|
|
|
78 |
self.model = AutoModelForSequenceClassification.from_pretrained(checkpoint)
|
79 |
self.tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
80 |
|
81 |
+
def _compute(
|
82 |
+
self,
|
83 |
+
predictions,
|
84 |
+
references,
|
85 |
+
batch_size=32,
|
86 |
+
max_length=128,
|
87 |
+
device=None,
|
88 |
+
):
|
89 |
"""Returns the scores"""
|
90 |
assert len(predictions) == len(
|
91 |
references
|
92 |
), "predictions and references should have the same number of sentences."
|
93 |
+
if device is not None:
|
94 |
+
assert device in ["gpu", "cpu"], "device should be either gpu or cpu."
|
|
|
95 |
else:
|
96 |
device = "gpu" if torch.cuda.is_available() else "cpu"
|
97 |
training_args = TrainingArguments(
|
98 |
"trainer",
|
99 |
fp16=(device == "gpu"),
|
100 |
+
per_device_eval_batch_size=batch_size,
|
101 |
report_to="all",
|
102 |
no_cuda=(device == "cpu"),
|
103 |
log_level="warning",
|
|
|
107 |
|
108 |
def tokenize_function(data):
|
109 |
return self.tokenizer(
|
110 |
+
data["sentence1"], data["sentence2"], max_length=max_length, truncation=True, padding=True
|
111 |
)
|
112 |
|
113 |
tokenized_datasets = raw_datasets.map(tokenize_function, batched=True)
|
requirements.txt
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
-
git+https://github.com/huggingface/evaluate@
|
2 |
torch
|
3 |
transformers
|
|
|
1 |
+
git+https://github.com/huggingface/evaluate@c447fc8eda9c62af501bfdc6988919571050d950
|
2 |
torch
|
3 |
transformers
|