File size: 6,962 Bytes
0ae85e9
 
 
 
 
 
dff6ebd
 
 
 
 
 
0ae85e9
725379f
0ae85e9
725379f
 
0ae85e9
 
725379f
 
 
 
 
0ae85e9
dff6ebd
0ae85e9
725379f
 
 
 
 
 
dff6ebd
725379f
 
 
dff6ebd
 
 
 
 
 
 
 
725379f
 
dff6ebd
725379f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dff6ebd
 
 
 
 
 
725379f
dff6ebd
725379f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dff6ebd
 
725379f
dff6ebd
725379f
 
 
dff6ebd
 
 
 
 
 
 
0ae85e9
 
7f93af3
0ae85e9
725379f
0ae85e9
 
 
dff6ebd
0ae85e9
 
 
dff6ebd
 
 
725379f
dff6ebd
 
 
725379f
0ae85e9
 
dff6ebd
0ae85e9
 
 
dff6ebd
0ae85e9
dff6ebd
0ae85e9
dff6ebd
0ae85e9
 
7f93af3
0ae85e9
 
7f93af3
0ae85e9
 
 
 
dff6ebd
 
725379f
dff6ebd
725379f
dff6ebd
 
0ae85e9
 
7f93af3
 
0ae85e9
 
 
 
7f93af3
725379f
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from datasets import load_dataset
from transformers import AutoTokenizer
from sklearn.preprocessing import LabelEncoder

# Import your model from tensor_network.py
from tensor_network import FourDimensionalTransformer  # Adjust the import path as needed

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# List of dataset identifiers for reasoning and knowledge
dataset_ids = [
    "race/all",    # For reasoning
    "squad"        # For general knowledge
]

# Update possible keys
possible_text_keys = ['question', 'sentence', 'query']
possible_context_keys = ['context', 'article', 'passage']
possible_label_keys = ['answer', 'answers', 'options']

# Initialize tokenizer
tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')

def tokenize_function_race(examples):
    texts = [q + " " + p for q, p in zip(examples['question'], examples['article'])]
    labels = examples['answer']
    tokenized_inputs = tokenizer(texts, padding='max_length', truncation=True, max_length=48)
    tokenized_inputs['labels'] = labels
    return tokenized_inputs

def tokenize_function_squad(examples):
    texts = [q + " " + c for q, c in zip(examples['question'], examples['context'])]
    labels = [ans['text'][0] if ans['text'] else '' for ans in examples['answers']]
    tokenized_inputs = tokenizer(texts, padding='max_length', truncation=True, max_length=48)
    tokenized_inputs['labels'] = labels
    return tokenized_inputs

# Initialize LabelEncoder
label_encoder = LabelEncoder()
all_labels = []

# Process RACE dataset
race_dataset = load_dataset('race', 'all')
tokenized_datasets = []
for split in race_dataset.keys():
    tokenized_race = race_dataset[split].map(
        tokenize_function_race,
        batched=True,
        remove_columns=race_dataset[split].column_names,
        load_from_cache_file=False,
    )
    tokenized_datasets.append({split: tokenized_race})
    # Collect labels
    all_labels.extend(tokenized_race['labels'])

# Process SQuAD dataset
squad_dataset = load_dataset('squad')
for split in squad_dataset.keys():
    tokenized_squad = squad_dataset[split].map(
        tokenize_function_squad,
        batched=True,
        remove_columns=squad_dataset[split].column_names,
        load_from_cache_file=False,
    )
    tokenized_datasets.append({split: tokenized_squad})
    # Collect labels
    all_labels.extend(tokenized_squad['labels'])

# Fit label encoder
label_encoder.fit(all_labels)
num_classes = len(label_encoder.classes_)
print(f"Number of unique labels: {num_classes}")

# Limit the number of classes to top 10 frequent labels
if num_classes > 10:
    print("Number of classes exceeds 10. Reducing to top 10 classes.")
    from collections import Counter
    label_counter = Counter(all_labels)
    top_10_labels = [label for label, _ in label_counter.most_common(10)]
    print(f"Top 10 labels: {top_10_labels}")
    label_mapping = {label: i for i, label in enumerate(top_10_labels)}
    label_mapping['other'] = len(top_10_labels)
    num_classes = len(top_10_labels) + 1
else:
    label_mapping = {label: i for i, label in enumerate(label_encoder.classes_)}

# Update model with correct num_classes
model = FourDimensionalTransformer(
    num_layers=16,
    embed_dim=7,
    num_heads=1,
    num_extra_tokens=16,
    num_classes=num_classes
).to(device)

def map_labels(labels):
    return [label_mapping.get(label, label_mapping['other']) for label in labels]

# Process datasets
for tokenized_dataset in tokenized_datasets:
    for split in tokenized_dataset.keys():
        tokenized_dataset[split] = tokenized_dataset[split].map(
            lambda examples: {'labels': map_labels(examples['labels'])},
            batched=True
        )
        tokenized_dataset[split] = tokenized_dataset[split].filter(
            lambda example: example['labels'] < num_classes
        )
        tokenized_dataset[split].set_format(type='torch', columns=['input_ids', 'labels'])

# Prepare DataLoaders
def prepare_dataloader(tokenized_datasets, split_name, batch_size=4):
    dataloaders = []
    for tokenized_dataset in tokenized_datasets:
        if split_name in tokenized_dataset:
            dataset_split = tokenized_dataset[split_name]
            dataloader = DataLoader(dataset_split, batch_size=batch_size, shuffle=True)
            dataloaders.append(dataloader)
    return dataloaders

train_dataloaders = prepare_dataloader(tokenized_datasets, 'train')
val_dataloaders = prepare_dataloader(tokenized_datasets, 'validation')

# Loss function and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=1e-4)

def train(model, train_dataloaders, val_dataloaders, num_epochs=10): #change number of Epochs to your liking
    for epoch in range(num_epochs):
        model.train()
        total_loss = 0
        total_batches = 0
        for dataloader in train_dataloaders:
            for batch in dataloader:
                input_ids = batch['input_ids']
                labels = batch['labels']

                # Reshape input_ids and move to device
                input_ids = input_ids[:, :48]
                input_ids = input_ids.view(-1, 3, 4, 4).float().to(device)

                # Convert labels to torch.long and move to device
                labels = labels.to(device).long()

                optimizer.zero_grad()
                outputs = model(input_ids)
                loss = criterion(outputs, labels)
                loss.backward()
                optimizer.step()

                total_loss += loss.item()
                total_batches += 1

        avg_loss = total_loss / total_batches if total_batches > 0 else 0
        print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {avg_loss:.4f}')

        # Validation loop
        model.eval()
        total_correct = 0
        total_samples = 0
        with torch.no_grad():
            for dataloader in val_dataloaders:
                for batch in dataloader:
                    input_ids = batch['input_ids']
                    labels = batch['labels']

                    input_ids = input_ids[:, :48]
                    input_ids = input_ids.view(-1, 3, 4, 4).float().to(device)
                    labels = labels.to(device).long()

                    outputs = model(input_ids)
                    _, predicted = torch.max(outputs, 1)
                    total_correct += (predicted == labels).sum().item()
                    total_samples += labels.size(0)
        accuracy = total_correct / total_samples if total_samples > 0 else 0
        print(f'Validation Accuracy: {accuracy:.4f}')

    torch.save(model.state_dict(), 'trained_model.pth')

# Start training
if train_dataloaders and val_dataloaders:
    train(model, train_dataloaders, val_dataloaders)
else:
    print("No data loaders available for training. Please check the datasets and preprocessing steps.")