Upload summarizer.py
Browse files- summarizer.py +74 -0
summarizer.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
import pytorch_lightning as pl
|
4 |
+
from torch import nn
|
5 |
+
from transformers import AdamW
|
6 |
+
from transformers import T5ForConditionalGeneration
|
7 |
+
from pytorch_lightning.callbacks import ModelCheckpoint
|
8 |
+
from pytorch_lightning.loggers import TensorBoardLogger
|
9 |
+
|
10 |
+
class SummarizerModel(pl.LightningModule):
|
11 |
+
def __init__(self, model_name = None):
|
12 |
+
super().__init__()
|
13 |
+
self.model = T5ForConditionalGeneration.from_pretrained(model_name, return_dict = True)
|
14 |
+
|
15 |
+
def forward(self,
|
16 |
+
input_ids,
|
17 |
+
attention_mask,
|
18 |
+
decoder_attention_mask,
|
19 |
+
labels = None):
|
20 |
+
output = self.model(
|
21 |
+
input_ids,
|
22 |
+
attention_mask = attention_mask,
|
23 |
+
labels = labels,
|
24 |
+
decoder_attention_mask = decoder_attention_mask
|
25 |
+
)
|
26 |
+
return output.loss, output.logits
|
27 |
+
|
28 |
+
def training_step(self, batch, batch_idx):
|
29 |
+
input_ids = batch['text_input_ids']
|
30 |
+
attention_mask = batch['text_attention_mask']
|
31 |
+
labels = batch['labels']
|
32 |
+
decoder_attention_mask = batch['labels_attention_mask']
|
33 |
+
|
34 |
+
loss, outputs = self.forward(
|
35 |
+
input_ids = input_ids,
|
36 |
+
attention_mask = attention_mask,
|
37 |
+
decoder_attention_mask = decoder_attention_mask,
|
38 |
+
labels = labels
|
39 |
+
)
|
40 |
+
self.log("train_loss", loss, prog_bar = True, logger = True)
|
41 |
+
return loss
|
42 |
+
|
43 |
+
def validation_step(self, batch, batch_idx):
|
44 |
+
input_ids = batch['text_input_ids']
|
45 |
+
attention_mask = batch['text_attention_mask']
|
46 |
+
labels = batch['labels']
|
47 |
+
decoder_attention_mask = batch['labels_attention_mask']
|
48 |
+
|
49 |
+
loss, outputs = self.forward(
|
50 |
+
input_ids = input_ids,
|
51 |
+
attention_mask = attention_mask,
|
52 |
+
decoder_attention_mask = decoder_attention_mask,
|
53 |
+
labels = labels
|
54 |
+
)
|
55 |
+
self.log("val_loss", loss, prog_bar = True, logger = True)
|
56 |
+
return loss
|
57 |
+
|
58 |
+
def test_step(self, batch, batch_idx):
|
59 |
+
input_ids = batch['text_input_ids']
|
60 |
+
attention_mask = batch['text_attention_mask']
|
61 |
+
labels = batch['labels']
|
62 |
+
decoder_attention_mask = batch['labels_attention_mask']
|
63 |
+
|
64 |
+
loss, outputs = self.forward(
|
65 |
+
input_ids = input_ids,
|
66 |
+
attention_mask = attention_mask,
|
67 |
+
decoder_attention_mask = decoder_attention_mask,
|
68 |
+
labels = labels
|
69 |
+
)
|
70 |
+
self.log("test_loss", loss, prog_bar = True, logger = True)
|
71 |
+
return loss
|
72 |
+
|
73 |
+
def configure_optimizers(self):
|
74 |
+
return AdamW(self.model.parameters(), lr = 0.0001)
|