Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -7,6 +7,7 @@ import gradio as gr
|
|
7 |
import pandas as pd
|
8 |
import os
|
9 |
import spaces
|
|
|
10 |
|
11 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
12 |
print(device)
|
@@ -23,7 +24,7 @@ def load_data(file):
|
|
23 |
global global_data
|
24 |
df = pd.read_csv(file)
|
25 |
inputs = tokenizer(df['text'].tolist(), padding=True, truncation=True, return_tensors="pt") # Mã hóa văn bản
|
26 |
-
labels = torch.tensor(df['
|
27 |
global_data = TensorDataset(inputs['input_ids'], inputs['attention_mask'], labels)
|
28 |
|
29 |
print(global_data)
|
@@ -33,7 +34,7 @@ def get_dataloader(start, end, batch_size=8):
|
|
33 |
subset = torch.utils.data.Subset(global_data, range(start, end))
|
34 |
return DataLoader(subset, batch_size=batch_size)
|
35 |
|
36 |
-
@spaces.GPU(duration=
|
37 |
def train_batch(dataloader):
|
38 |
model.train()
|
39 |
start_time = time.time()
|
@@ -49,41 +50,45 @@ def train_batch(dataloader):
|
|
49 |
optimizer.step()
|
50 |
|
51 |
elapsed_time = time.time() - start_time
|
52 |
-
if elapsed_time >
|
53 |
-
print("save checkpoint")
|
54 |
torch.save(model.state_dict(), "./checkpoint/model.pt")
|
55 |
return False, "Checkpoint saved. Training paused."
|
56 |
|
57 |
return True, "Batch training completed."
|
58 |
|
59 |
-
|
60 |
def train_step(file=None):
|
61 |
if file:
|
62 |
load_data(file)
|
|
|
63 |
|
64 |
start_idx = 0
|
65 |
batch_size = 8
|
66 |
total_samples = len(global_data)
|
67 |
|
|
|
68 |
while start_idx < total_samples:
|
69 |
-
print(
|
70 |
-
|
|
|
|
|
71 |
dataloader = get_dataloader(start_idx, end_idx, batch_size)
|
72 |
|
73 |
-
|
74 |
-
|
75 |
-
|
|
|
76 |
|
77 |
-
|
78 |
-
|
79 |
-
|
|
|
80 |
|
81 |
start_idx = end_idx
|
|
|
82 |
|
83 |
torch.save(model.state_dict(), "./checkpoint/model.pt")
|
84 |
return "Training completed and model saved."
|
85 |
|
86 |
-
|
87 |
if __name__ == "__main__":
|
88 |
iface = gr.Interface(
|
89 |
fn=train_step,
|
|
|
7 |
import pandas as pd
|
8 |
import os
|
9 |
import spaces
|
10 |
+
from spaces.zero.gradio import HTMLError
|
11 |
|
12 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
13 |
print(device)
|
|
|
24 |
global global_data
|
25 |
df = pd.read_csv(file)
|
26 |
inputs = tokenizer(df['text'].tolist(), padding=True, truncation=True, return_tensors="pt") # Mã hóa văn bản
|
27 |
+
labels = torch.tensor(df['label'].tolist()).long() # Đảm bảo tên cột là 'label'
|
28 |
global_data = TensorDataset(inputs['input_ids'], inputs['attention_mask'], labels)
|
29 |
|
30 |
print(global_data)
|
|
|
34 |
subset = torch.utils.data.Subset(global_data, range(start, end))
|
35 |
return DataLoader(subset, batch_size=batch_size)
|
36 |
|
37 |
+
@spaces.GPU(duration=5)
|
38 |
def train_batch(dataloader):
|
39 |
model.train()
|
40 |
start_time = time.time()
|
|
|
50 |
optimizer.step()
|
51 |
|
52 |
elapsed_time = time.time() - start_time
|
53 |
+
if elapsed_time > 50: # Dừng trước 59 giây để đảm bảo không vượt hạn ngạch
|
|
|
54 |
torch.save(model.state_dict(), "./checkpoint/model.pt")
|
55 |
return False, "Checkpoint saved. Training paused."
|
56 |
|
57 |
return True, "Batch training completed."
|
58 |
|
|
|
59 |
def train_step(file=None):
|
60 |
if file:
|
61 |
load_data(file)
|
62 |
+
print(global_data)
|
63 |
|
64 |
start_idx = 0
|
65 |
batch_size = 8
|
66 |
total_samples = len(global_data)
|
67 |
|
68 |
+
counting = 0
|
69 |
while start_idx < total_samples:
|
70 |
+
print("Step:", counting)
|
71 |
+
print("Percent:", total_samples/start_idx * 100, "%")
|
72 |
+
counting += 1
|
73 |
+
end_idx = min(start_idx + (batch_size * 10), total_samples) # 10 batches per loop
|
74 |
dataloader = get_dataloader(start_idx, end_idx, batch_size)
|
75 |
|
76 |
+
try:
|
77 |
+
success, message = train_batch(dataloader)
|
78 |
+
if not success:
|
79 |
+
return message
|
80 |
|
81 |
+
except HTMLError as e:
|
82 |
+
print("Exceeded GPU quota, retrying in 10 seconds...")
|
83 |
+
time.sleep(10)
|
84 |
+
continue
|
85 |
|
86 |
start_idx = end_idx
|
87 |
+
time.sleep(2) # Nghỉ 2 giây giữa các phiên huấn luyện
|
88 |
|
89 |
torch.save(model.state_dict(), "./checkpoint/model.pt")
|
90 |
return "Training completed and model saved."
|
91 |
|
|
|
92 |
if __name__ == "__main__":
|
93 |
iface = gr.Interface(
|
94 |
fn=train_step,
|