Update lm_finetuning.py
Browse files- lm_finetuning.py +34 -3
lm_finetuning.py
CHANGED
@@ -40,6 +40,28 @@ logging.basicConfig(format='%(asctime)s %(levelname)-8s %(message)s', level=logg
|
|
40 |
|
41 |
PARALLEL = bool(int(os.getenv("PARALLEL", 1)))
|
42 |
RAY_RESULTS = os.getenv("RAY_RESULTS", "ray_results")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
|
44 |
|
45 |
def internet_connection(host='http://google.com'):
|
@@ -105,6 +127,8 @@ def main():
|
|
105 |
tokenizer = AutoTokenizer.from_pretrained(opt.model, local_files_only=not network)
|
106 |
model = AutoModelForSequenceClassification.from_pretrained(
|
107 |
opt.model,
|
|
|
|
|
108 |
num_labels=len(dataset[opt.split_train]['label'][0]),
|
109 |
local_files_only=not network,
|
110 |
problem_type="multi_label_classification"
|
@@ -129,7 +153,8 @@ def main():
|
|
129 |
eval_dataset=tokenized_datasets[opt.split_validation],
|
130 |
compute_metrics=compute_metric_search,
|
131 |
model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
|
132 |
-
opt.model, return_dict=True, num_labels=len(dataset[opt.split_train]['label'][0]))
|
|
|
133 |
)
|
134 |
# parameter search
|
135 |
if PARALLEL:
|
@@ -166,7 +191,10 @@ def main():
|
|
166 |
best_model_path,
|
167 |
num_labels=len(dataset[opt.split_train]['label'][0]),
|
168 |
local_files_only=not network,
|
169 |
-
problem_type="multi_label_classification"
|
|
|
|
|
|
|
170 |
trainer = Trainer(
|
171 |
model=model,
|
172 |
args=TrainingArguments(
|
@@ -182,7 +210,10 @@ def main():
|
|
182 |
return_dict=True,
|
183 |
num_labels=len(dataset[opt.split_train]['label'][0]),
|
184 |
local_files_only=not network,
|
185 |
-
problem_type="multi_label_classification"
|
|
|
|
|
|
|
186 |
)
|
187 |
summary_file = pj(opt.output_dir, opt.summary_file)
|
188 |
if not opt.skip_eval:
|
|
|
40 |
|
41 |
PARALLEL = bool(int(os.getenv("PARALLEL", 1)))
|
42 |
RAY_RESULTS = os.getenv("RAY_RESULTS", "ray_results")
|
43 |
+
LABEL2ID = {
|
44 |
+
"arts_&_culture": 0,
|
45 |
+
"business_&_entrepreneurs": 1,
|
46 |
+
"celebrity_&_pop_culture": 2,
|
47 |
+
"diaries_&_daily_life": 3,
|
48 |
+
"family": 4,
|
49 |
+
"fashion_&_style": 5,
|
50 |
+
"film_tv_&_video": 6,
|
51 |
+
"fitness_&_health": 7,
|
52 |
+
"food_&_dining": 8,
|
53 |
+
"gaming": 9,
|
54 |
+
"learning_&_educational": 10,
|
55 |
+
"music": 11,
|
56 |
+
"news_&_social_concern": 12,
|
57 |
+
"other_hobbies": 13,
|
58 |
+
"relationships": 14,
|
59 |
+
"science_&_technology": 15,
|
60 |
+
"sports": 16,
|
61 |
+
"travel_&_adventure": 17,
|
62 |
+
"youth_&_student_life": 18
|
63 |
+
}
|
64 |
+
ID2LABEL = {v: k for k, v in LABEL2ID.items()}
|
65 |
|
66 |
|
67 |
def internet_connection(host='http://google.com'):
|
|
|
127 |
tokenizer = AutoTokenizer.from_pretrained(opt.model, local_files_only=not network)
|
128 |
model = AutoModelForSequenceClassification.from_pretrained(
|
129 |
opt.model,
|
130 |
+
id2label=ID2LABEL,
|
131 |
+
label2id=LABEL2ID,
|
132 |
num_labels=len(dataset[opt.split_train]['label'][0]),
|
133 |
local_files_only=not network,
|
134 |
problem_type="multi_label_classification"
|
|
|
153 |
eval_dataset=tokenized_datasets[opt.split_validation],
|
154 |
compute_metrics=compute_metric_search,
|
155 |
model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
|
156 |
+
opt.model, return_dict=True, num_labels=len(dataset[opt.split_train]['label'][0])),
|
157 |
+
id2label=ID2LABEL, label2id=LABEL2ID,
|
158 |
)
|
159 |
# parameter search
|
160 |
if PARALLEL:
|
|
|
191 |
best_model_path,
|
192 |
num_labels=len(dataset[opt.split_train]['label'][0]),
|
193 |
local_files_only=not network,
|
194 |
+
problem_type="multi_label_classification",
|
195 |
+
id2label=ID2LABEL,
|
196 |
+
label2id=LABEL2ID
|
197 |
+
)
|
198 |
trainer = Trainer(
|
199 |
model=model,
|
200 |
args=TrainingArguments(
|
|
|
210 |
return_dict=True,
|
211 |
num_labels=len(dataset[opt.split_train]['label'][0]),
|
212 |
local_files_only=not network,
|
213 |
+
problem_type="multi_label_classification",
|
214 |
+
id2label=ID2LABEL,
|
215 |
+
label2id=LABEL2ID
|
216 |
+
)
|
217 |
)
|
218 |
summary_file = pj(opt.output_dir, opt.summary_file)
|
219 |
if not opt.skip_eval:
|