jayksharma
commited on
Commit
•
da9ac04
1
Parent(s):
2d8a6d5
Update train.py
Browse files
train.py
ADDED
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# train.py
|
2 |
+
|
3 |
+
import torch
|
4 |
+
import torch.nn as nn
|
5 |
+
import torch.optim as optim
|
6 |
+
from torch.utils.data import DataLoader, Dataset
|
7 |
+
from super_large_language_model import TransformerModel
|
8 |
+
|
9 |
+
class TextDataset(Dataset):
|
10 |
+
def __init__(self, texts, vocab):
|
11 |
+
self.texts = texts
|
12 |
+
self.vocab = vocab
|
13 |
+
|
14 |
+
def __len__(self):
|
15 |
+
return len(self.texts)
|
16 |
+
|
17 |
+
def __getitem__(self, idx):
|
18 |
+
text = self.texts[idx]
|
19 |
+
text_indices = [self.vocab[char] for char in text]
|
20 |
+
return torch.tensor(text_indices)
|
21 |
+
|
22 |
+
def train_model(model, dataset, num_epochs=10, batch_size=32, learning_rate=0.001):
|
23 |
+
dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True)
|
24 |
+
criterion = nn.CrossEntropyLoss()
|
25 |
+
optimizer = optim.Adam(model.parameters(), lr=learning_rate)
|
26 |
+
|
27 |
+
for epoch in range(num_epochs):
|
28 |
+
model.train()
|
29 |
+
for batch in dataloader:
|
30 |
+
optimizer.zero_grad()
|
31 |
+
output = model(batch[:-1], batch[1:])
|
32 |
+
loss = criterion(output.view(-1, output.size(-1)), batch[1:].view(-1))
|
33 |
+
loss.backward()
|
34 |
+
optimizer.step()
|
35 |
+
print(f'Epoch {epoch+1}, Loss: {loss.item()}')
|
36 |
+
|
37 |
+
if __name__ == "__main__":
|
38 |
+
# Example texts and vocabulary
|
39 |
+
texts = ["hello world", "pytorch is great"]
|
40 |
+
vocab = {char: idx for idx, char in enumerate(set("".join(texts)))}
|
41 |
+
|
42 |
+
dataset = TextDataset(texts, vocab)
|
43 |
+
model = TransformerModel(vocab_size=len(vocab), d_model=512, nhead=8, num_encoder_layers=6, num_decoder_layers=6, dim_feedforward=2048)
|
44 |
+
|
45 |
+
train_model(model, dataset)
|