First model version
Browse files
main.py
ADDED
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datasets import load_dataset
|
2 |
+
from transformers import AutoTokenizer
|
3 |
+
from transformers import DataCollatorWithPadding
|
4 |
+
import numpy as np
|
5 |
+
import evaluate
|
6 |
+
|
7 |
+
accuracy = evaluate.load("accuracy")
|
8 |
+
|
9 |
+
|
10 |
+
def compute_metrics(eval_pred):
|
11 |
+
predictions, labels = eval_pred
|
12 |
+
predictions = np.argmax(predictions, axis=1)
|
13 |
+
return accuracy.compute(predictions=predictions, references=labels)
|
14 |
+
|
15 |
+
|
16 |
+
def load_data():
|
17 |
+
### load dataset
|
18 |
+
imdb = load_dataset("imdb")
|
19 |
+
return imdb
|
20 |
+
|
21 |
+
|
22 |
+
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
|
23 |
+
|
24 |
+
|
25 |
+
def preprocess_function(examples):
|
26 |
+
return tokenizer(examples["text"], truncation=True)
|
27 |
+
|
28 |
+
|
29 |
+
def main():
|
30 |
+
imdb = load_data()
|
31 |
+
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
|
32 |
+
|
33 |
+
def preprocess_function(examples):
|
34 |
+
return tokenizer(examples["text"], truncation=True)
|
35 |
+
|
36 |
+
tokenized_imdb = imdb.map(preprocess_function, batched=True)
|
37 |
+
data_collator = DataCollatorWithPadding(tokenizer=tokenizer)
|
38 |
+
|
39 |
+
id2label = {0: "NEGATIVE", 1: "POSITIVE"}
|
40 |
+
label2id = {"NEGATIVE": 0, "POSITIVE": 1}
|
41 |
+
|
42 |
+
from transformers import (
|
43 |
+
AutoModelForSequenceClassification,
|
44 |
+
TrainingArguments,
|
45 |
+
Trainer,
|
46 |
+
)
|
47 |
+
|
48 |
+
model = AutoModelForSequenceClassification.from_pretrained(
|
49 |
+
"distilbert-base-uncased", num_labels=2, id2label=id2label, label2id=label2id
|
50 |
+
)
|
51 |
+
|
52 |
+
# FP16, Multi GPU, accelerator
|
53 |
+
# is it possible to continue training?
|
54 |
+
|
55 |
+
training_args = TrainingArguments(
|
56 |
+
output_dir="./",
|
57 |
+
learning_rate=2e-5,
|
58 |
+
per_device_train_batch_size=16,
|
59 |
+
per_device_eval_batch_size=16,
|
60 |
+
num_train_epochs=2,
|
61 |
+
weight_decay=0.01,
|
62 |
+
evaluation_strategy="epoch",
|
63 |
+
save_strategy="epoch",
|
64 |
+
load_best_model_at_end=True,
|
65 |
+
push_to_hub=True,
|
66 |
+
)
|
67 |
+
|
68 |
+
trainer = Trainer(
|
69 |
+
model=model,
|
70 |
+
args=training_args,
|
71 |
+
train_dataset=tokenized_imdb["train"],
|
72 |
+
eval_dataset=tokenized_imdb["test"],
|
73 |
+
tokenizer=tokenizer,
|
74 |
+
data_collator=data_collator,
|
75 |
+
compute_metrics=compute_metrics,
|
76 |
+
)
|
77 |
+
|
78 |
+
# trainer.train()
|
79 |
+
trainer.push_to_hub()
|
80 |
+
|
81 |
+
|
82 |
+
if __name__ == "__main__":
|
83 |
+
main()
|