Datasets:
Tasks:
Text Classification
Modalities:
Text
Sub-tasks:
sentiment-classification
Languages:
English
Size:
10K - 100K
ArXiv:
License:
Update lm_finetuning.py
Browse files- lm_finetuning.py +47 -4
lm_finetuning.py
CHANGED
@@ -35,10 +35,21 @@ from datasets import load_dataset, load_metric
|
|
35 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
|
36 |
from ray import tune
|
37 |
|
|
|
|
|
38 |
logging.basicConfig(format='%(asctime)s %(levelname)-8s %(message)s', level=logging.INFO, datefmt='%Y-%m-%d %H:%M:%S')
|
39 |
|
40 |
PARALLEL = bool(int(os.getenv("PARALLEL", 1)))
|
41 |
RAY_RESULTS = os.getenv("RAY_RESULTS", "ray_results")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
|
43 |
|
44 |
def internet_connection(host='http://google.com'):
|
@@ -96,7 +107,12 @@ def main():
|
|
96 |
# setup model
|
97 |
tokenizer = AutoTokenizer.from_pretrained(opt.model, local_files_only=not network)
|
98 |
model = AutoModelForSequenceClassification.from_pretrained(
|
99 |
-
opt.model,
|
|
|
|
|
|
|
|
|
|
|
100 |
tokenized_datasets = dataset.map(
|
101 |
lambda x: tokenizer(x["text"], padding="max_length", truncation=True, max_length=opt.seq_length),
|
102 |
batched=True)
|
@@ -116,7 +132,14 @@ def main():
|
|
116 |
train_dataset=tokenized_datasets[opt.split_train],
|
117 |
eval_dataset=tokenized_datasets[opt.split_validation],
|
118 |
compute_metrics=compute_metric_search,
|
119 |
-
model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
)
|
121 |
# parameter search
|
122 |
if PARALLEL:
|
@@ -150,7 +173,12 @@ def main():
|
|
150 |
|
151 |
# evaluation
|
152 |
model = AutoModelForSequenceClassification.from_pretrained(
|
153 |
-
opt.model,
|
|
|
|
|
|
|
|
|
|
|
154 |
trainer = Trainer(
|
155 |
model=model,
|
156 |
args=TrainingArguments(
|
@@ -162,7 +190,13 @@ def main():
|
|
162 |
eval_dataset=tokenized_datasets[opt.split_test],
|
163 |
compute_metrics=compute_metric_all,
|
164 |
model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
|
165 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
166 |
)
|
167 |
summary_file = pj(opt.output_dir, opt.summary_file)
|
168 |
if not opt.skip_eval:
|
@@ -181,6 +215,15 @@ def main():
|
|
181 |
tokenizer.push_to_hub(opt.model_alias, **args)
|
182 |
if os.path.exists(summary_file):
|
183 |
shutil.copy2(summary_file, opt.model_alias)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
184 |
os.system(
|
185 |
f"cd {opt.model_alias} && git lfs install && git add . && git commit -m 'model update' && git push && cd ../")
|
186 |
shutil.rmtree(f"{opt.model_alias}") # clean up the cloned repo
|
|
|
35 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
|
36 |
from ray import tune
|
37 |
|
38 |
+
from readme import get_readme
|
39 |
+
|
40 |
logging.basicConfig(format='%(asctime)s %(levelname)-8s %(message)s', level=logging.INFO, datefmt='%Y-%m-%d %H:%M:%S')
|
41 |
|
42 |
PARALLEL = bool(int(os.getenv("PARALLEL", 1)))
|
43 |
RAY_RESULTS = os.getenv("RAY_RESULTS", "ray_results")
|
44 |
+
LABEL2ID = {
|
45 |
+
"arts_&_culture": 0,
|
46 |
+
"business_&_entrepreneurs": 1,
|
47 |
+
"pop_culture": 2,
|
48 |
+
"daily_life": 3,
|
49 |
+
"sports_&_gaming": 4,
|
50 |
+
"science_&_technology": 5
|
51 |
+
}
|
52 |
+
ID2LABEL = {v: k for k, v in LABEL2ID.items()}
|
53 |
|
54 |
|
55 |
def internet_connection(host='http://google.com'):
|
|
|
107 |
# setup model
|
108 |
tokenizer = AutoTokenizer.from_pretrained(opt.model, local_files_only=not network)
|
109 |
model = AutoModelForSequenceClassification.from_pretrained(
|
110 |
+
opt.model,
|
111 |
+
num_labels=6,
|
112 |
+
local_files_only=not network,
|
113 |
+
id2label=ID2LABEL,
|
114 |
+
label2id=LABEL2ID
|
115 |
+
)
|
116 |
tokenized_datasets = dataset.map(
|
117 |
lambda x: tokenizer(x["text"], padding="max_length", truncation=True, max_length=opt.seq_length),
|
118 |
batched=True)
|
|
|
132 |
train_dataset=tokenized_datasets[opt.split_train],
|
133 |
eval_dataset=tokenized_datasets[opt.split_validation],
|
134 |
compute_metrics=compute_metric_search,
|
135 |
+
model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
|
136 |
+
opt.model,
|
137 |
+
num_labels=6,
|
138 |
+
local_files_only=not network,
|
139 |
+
return_dict=True,
|
140 |
+
id2label=ID2LABEL,
|
141 |
+
label2id=LABEL2ID
|
142 |
+
)
|
143 |
)
|
144 |
# parameter search
|
145 |
if PARALLEL:
|
|
|
173 |
|
174 |
# evaluation
|
175 |
model = AutoModelForSequenceClassification.from_pretrained(
|
176 |
+
opt.model,
|
177 |
+
num_labels=6,
|
178 |
+
local_files_only=not network,
|
179 |
+
id2label=ID2LABEL,
|
180 |
+
label2id=LABEL2ID
|
181 |
+
)
|
182 |
trainer = Trainer(
|
183 |
model=model,
|
184 |
args=TrainingArguments(
|
|
|
190 |
eval_dataset=tokenized_datasets[opt.split_test],
|
191 |
compute_metrics=compute_metric_all,
|
192 |
model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
|
193 |
+
opt.model,
|
194 |
+
num_labels=6,
|
195 |
+
local_files_only=not network,
|
196 |
+
return_dict=True,
|
197 |
+
id2label=ID2LABEL,
|
198 |
+
label2id=LABEL2ID
|
199 |
+
)
|
200 |
)
|
201 |
summary_file = pj(opt.output_dir, opt.summary_file)
|
202 |
if not opt.skip_eval:
|
|
|
215 |
tokenizer.push_to_hub(opt.model_alias, **args)
|
216 |
if os.path.exists(summary_file):
|
217 |
shutil.copy2(summary_file, opt.model_alias)
|
218 |
+
extra_desc = f"This model is fine-tuned on `{opt.split_train}` split and validated on `{opt.split_test}` split of tweet_topic."
|
219 |
+
readme = get_readme(
|
220 |
+
model_name=opt.model_alias,
|
221 |
+
metric=f"{opt.model_alias}/{summary_file}",
|
222 |
+
language_model=opt.model,
|
223 |
+
extra_desc= extra_desc
|
224 |
+
)
|
225 |
+
with open(f"{opt.model_alias}/README.md", "w") as f:
|
226 |
+
f.write(readme)
|
227 |
os.system(
|
228 |
f"cd {opt.model_alias} && git lfs install && git add . && git commit -m 'model update' && git push && cd ../")
|
229 |
shutil.rmtree(f"{opt.model_alias}") # clean up the cloned repo
|