lucidmorto
commited on
Commit
•
085809d
1
Parent(s):
fea89cd
Initial commit with training script
Browse files- humanizer.py +94 -0
- requirements.txt +3 -0
- run_trainer.py +5 -0
humanizer.py
ADDED
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datasets import load_dataset, DatasetDict
|
2 |
+
from transformers import AutoTokenizer, T5ForConditionalGeneration, Seq2SeqTrainingArguments, Seq2SeqTrainer
|
3 |
+
from transformers import EarlyStoppingCallback
|
4 |
+
from transformers.integrations import TensorBoardCallback
|
5 |
+
import logging
|
6 |
+
|
7 |
+
# Set up logging
|
8 |
+
logging.basicConfig(level=logging.INFO)
|
9 |
+
logger = logging.getLogger(__name__)
|
10 |
+
|
11 |
+
def generate_formal_text(text):
|
12 |
+
# Implement formal text generation here
|
13 |
+
return text # Placeholder
|
14 |
+
|
15 |
+
def prepare_data(example):
|
16 |
+
example["formal_text"] = generate_formal_text(example["text"])
|
17 |
+
return example
|
18 |
+
|
19 |
+
def tokenize_function(examples, tokenizer):
|
20 |
+
model_inputs = tokenizer(examples["formal_text"], max_length=128, truncation=True, padding="max_length")
|
21 |
+
labels = tokenizer(examples["text"], max_length=128, truncation=True, padding="max_length")
|
22 |
+
model_inputs["labels"] = labels["input_ids"]
|
23 |
+
return model_inputs
|
24 |
+
|
25 |
+
def main():
|
26 |
+
# Load the dataset and take only 10000 samples
|
27 |
+
logger.info("Loading dataset...")
|
28 |
+
dataset = load_dataset("LucasChu/reddit_comments")
|
29 |
+
dataset = dataset.shuffle(seed=42)
|
30 |
+
dataset["train"] = dataset["train"].select(range(10000))
|
31 |
+
logger.info("Dataset loaded, shuffled, and truncated to 10,000 samples.")
|
32 |
+
|
33 |
+
# Split the train dataset into train and test
|
34 |
+
train_testvalid = dataset["train"].train_test_split(test_size=0.2, seed=42)
|
35 |
+
test_valid = train_testvalid["test"].train_test_split(test_size=0.5, seed=42)
|
36 |
+
|
37 |
+
dataset = DatasetDict({
|
38 |
+
"train": train_testvalid["train"],
|
39 |
+
"test": test_valid["test"],
|
40 |
+
"validation": test_valid["train"]
|
41 |
+
})
|
42 |
+
|
43 |
+
# Prepare the dataset
|
44 |
+
logger.info("Preparing dataset...")
|
45 |
+
processed_dataset = {split: data.map(prepare_data) for split, data in dataset.items()}
|
46 |
+
logger.info("Dataset prepared.")
|
47 |
+
|
48 |
+
# Tokenize the dataset
|
49 |
+
model_name = "t5-small"
|
50 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
51 |
+
|
52 |
+
logger.info("Tokenizing dataset...")
|
53 |
+
tokenized_dataset = {split: data.map(lambda examples: tokenize_function(examples, tokenizer), batched=True) for split, data in processed_dataset.items()}
|
54 |
+
logger.info("Dataset tokenized.")
|
55 |
+
|
56 |
+
# Check available splits in the dataset
|
57 |
+
available_splits = list(tokenized_dataset.keys())
|
58 |
+
logger.info(f"Available splits in the dataset: {available_splits}")
|
59 |
+
|
60 |
+
# Set up the model and trainer
|
61 |
+
logger.info("Setting up model and trainer...")
|
62 |
+
model = T5ForConditionalGeneration.from_pretrained(model_name)
|
63 |
+
|
64 |
+
training_args = Seq2SeqTrainingArguments(
|
65 |
+
output_dir="./results",
|
66 |
+
num_train_epochs=1,
|
67 |
+
per_device_train_batch_size=16,
|
68 |
+
warmup_steps=100,
|
69 |
+
weight_decay=0.01,
|
70 |
+
logging_dir="./logs",
|
71 |
+
logging_steps=100,
|
72 |
+
evaluation_strategy="steps" if "test" in available_splits else "no",
|
73 |
+
eval_steps=500,
|
74 |
+
save_steps=1000,
|
75 |
+
load_best_model_at_end=True,
|
76 |
+
metric_for_best_model="eval_loss",
|
77 |
+
greater_is_better=False
|
78 |
+
)
|
79 |
+
|
80 |
+
trainer = Seq2SeqTrainer(
|
81 |
+
model=model,
|
82 |
+
args=training_args,
|
83 |
+
train_dataset=tokenized_dataset["train"],
|
84 |
+
eval_dataset=tokenized_dataset.get("test"),
|
85 |
+
tokenizer=tokenizer,
|
86 |
+
callbacks=[EarlyStoppingCallback(early_stopping_patience=3), TensorBoardCallback()]
|
87 |
+
)
|
88 |
+
logger.info("Model and trainer set up.")
|
89 |
+
|
90 |
+
# Return the trainer object
|
91 |
+
return trainer
|
92 |
+
|
93 |
+
if __name__ == "__main__":
|
94 |
+
main()
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
datasets
|
2 |
+
transformers
|
3 |
+
torch
|
run_trainer.py
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from humanizer import main
|
2 |
+
|
3 |
+
if __name__ == "__main__":
|
4 |
+
trainer = main()
|
5 |
+
trainer.train()
|