File size: 9,785 Bytes
5e89ee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8e7d1ea
5e89ee6
8e7d1ea
 
 
 
 
 
 
 
5e89ee6
 
 
 
 
 
 
 
8e7d1ea
 
 
 
 
 
 
 
5e89ee6
 
 
 
 
 
 
 
8e7d1ea
 
 
5e89ee6
 
 
 
 
8e7d1ea
 
5e89ee6
 
 
 
 
8e7d1ea
 
5e89ee6
 
8e7d1ea
 
 
 
 
 
 
 
 
 
 
 
 
5e89ee6
 
8e7d1ea
 
 
5e89ee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9d4c2df
 
 
 
5e89ee6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
import streamlit as st

from datasets import load_dataset

import numpy as np
import os
from sklearn.metrics import accuracy_score, precision_recall_fscore_support

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader

from transformers import AutoModelForTokenClassification, AutoTokenizer, DataCollatorForTokenClassification
from transformers import DebertaV2Config, DebertaV2ForTokenClassification

os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True" 

# print weights
def print_trainable_parameters(model):
    pytorch_total_params = sum(p.numel() for p in model.parameters())
    torch_total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
    print(f'total params: {pytorch_total_params}. tunable params: {torch_total_params}')

device = torch.device('cpu')
print(f"Is CUDA available: {torch.cuda.is_available()}")
# True
if torch.cuda.is_available():
    print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
    device = torch.device('cuda')

# Load models
st.write('Loading the pretrained model ...')
teacher_model_name = "iiiorg/piiranha-v1-detect-personal-information"
teacher_model = AutoModelForTokenClassification.from_pretrained(teacher_model_name)
tokenizer = AutoTokenizer.from_pretrained(teacher_model_name)
print(teacher_model)
print_trainable_parameters(teacher_model)
label2id = teacher_model.config.label2id
id2label = teacher_model.config.id2label 

st.write("id2label: ", id2label)
st.write("label2id: ", label2id)
dimension = len(id2label)
st.write("dimension", dimension)

student_model_config = teacher_model.config
student_model_config.num_attention_heads = 8
student_model_config.num_hidden_layers = 4
student_model = DebertaV2ForTokenClassification.from_pretrained(
    "microsoft/mdeberta-v3-base",
    config=student_model_config)
#    ignore_mismatched_sizes=True)
print(student_model)
print_trainable_parameters(student_model)

if torch.cuda.is_available():
    teacher_model = teacher_model.to(device)
    student_model = student_model.to(device)

# Load data. 
raw_dataset = load_dataset("ai4privacy/pii-masking-400k", split='train')
raw_dataset = raw_dataset.filter(lambda example: example["language"].startswith("en"))
#raw_dataset = raw_dataset.select(range(2000))
raw_dataset = raw_dataset.filter(lambda example, idx: idx % 11 == 0, with_indices=True)
raw_dataset = raw_dataset.train_test_split(test_size=0.2)
print(raw_dataset)
print(raw_dataset.column_names)

# inputs = tokenizer(
#     raw_dataset['train'][0]['mbert_tokens'],
#     truncation=True,
#     is_split_into_words=True)
# print(inputs)
# print(inputs.tokens())
# print(inputs.word_ids())

# function to align labels with tokens 
# --> special tokens: -100 label id (ignored by cross entropy),
# --> if tokens are inside a word, replace 'B-' with 'I-' 
def align_labels_with_tokens(labels, word_ids, max_length):
    aligned_label_ids = []
    for word_id in word_ids:
        if word_id is None:  
            aligned_label_ids.append(-100)
        else: 
            aligned_label_ids.append(label2id[labels[word_id]].replace("B-", "I-"))

    # Pad to max length
    aligned_label_ids += [-100] * (max_length - len(aligned_label_ids))
    return aligned_label_ids

# create tokenize function
def tokenize_function(examples):
    inputs = tokenizer(
        examples['mbert_tokens'],
        is_split_into_words=True,
        truncation=True,
        max_length=512,
        padding="max_length"
    )
    word_ids = inputs.word_ids()
    inputs["labels"] = [
        align_labels_with_tokens(labels, word_ids, tokenizer.model_max_length)
        for labels in examples['mbert_token_classes']
    ]
    return inputs

# tokenize training and validation datasets
tokenized_data = raw_dataset.map(
    tokenize_function,
    batched=True)
tokenized_data.set_format(type='torch', columns=['input_ids', 'attention_mask', 'labels'])
# data collator
data_collator = DataCollatorForTokenClassification(
    tokenizer, padding=True, truncation=True, max_length=512
)

st.write(tokenized_data["train"][:2]["labels"])

# Function to evaluate model performance
def evaluate_model(model, dataloader, device):
    model.eval()
    all_preds, all_labels = [], []

    with torch.no_grad():
        for batch in dataloader:
            input_ids = batch['input_ids'].to(device)
            attention_mask = batch['attention_mask'].to(device)
            labels = batch['labels'].to(device)

            outputs = model(input_ids, attention_mask=attention_mask)
            logits = outputs.logits
            preds = torch.argmax(logits, dim=-1)

            # Mask out padding tokens (-100 in labels)
            mask = labels != -100
            valid_preds = preds[mask]
            valid_labels = labels[mask]

            all_preds.extend(valid_preds.cpu().numpy())
            all_labels.extend(valid_labels.cpu().numpy())

    # Convert to numpy arrays for metrics calculation
    all_preds = np.array(all_preds)
    all_labels = np.array(all_labels)

    accuracy = accuracy_score(all_labels, all_preds)
    precision, recall, f1, _ = precision_recall_fscore_support(
        all_labels, all_preds, average='micro'
    )

    return accuracy, precision, recall, f1

# Function to compute distillation and hard-label loss
def distillation_loss(student_logits, teacher_logits, true_labels, temperature, alpha):
    # print("Distillation loss sizes")
    # print(teacher_logits.size())
    # print(student_logits.size())
    # print(true_labels.size())
    # Compute soft targets from teacher logits
    soft_targets = nn.functional.softmax(teacher_logits / temperature, dim=-1)
    student_soft = nn.functional.log_softmax(student_logits / temperature, dim=-1)

    # KL Divergence loss for distillation
    distill_loss = nn.functional.kl_div(student_soft, soft_targets, reduction='batchmean') * (temperature ** 2)

    # Cross-entropy loss for hard labels
    student_logit_reshape = torch.transpose(student_logits, 1, 2) # transpose to match the labels dimension
    hard_loss = nn.CrossEntropyLoss()(student_logit_reshape, true_labels)

    # Combine losses
    loss = alpha * distill_loss + (1.0 - alpha) * hard_loss

    return loss

# hyperparameters
batch_size = 32
lr = 1e-4
num_epochs = 30
temperature = 2.0
alpha = 0.5

# define optimizer
optimizer = optim.Adam(student_model.parameters(), lr=lr)

# create training data loader
dataloader = DataLoader(tokenized_data['train'], batch_size=batch_size, collate_fn=data_collator)
# create testing data loader
test_dataloader = DataLoader(tokenized_data['test'], batch_size=batch_size, collate_fn=data_collator)

# TEMPORARY - for testing
teacher_accuracy, teacher_precision, teacher_recall, teacher_f1 = evaluate_model(teacher_model, test_dataloader, device)
print(f"Teacher (test) - Accuracy: {teacher_accuracy:.4f}, Precision: {teacher_precision:.4f}, Recall: {teacher_recall:.4f}, F1 Score: {teacher_f1:.4f}")

# put student model in train mode
student_model.train()

# train model
for epoch in range(num_epochs):
    for batch in dataloader:
        # Prepare inputs
        input_ids = batch['input_ids'].to(device)
        attention_mask = batch['attention_mask'].to(device)
        labels = batch['labels'].to(device)

        # Disable gradient calculation for teacher model
        with torch.no_grad():
            teacher_outputs = teacher_model(input_ids, attention_mask=attention_mask)
            teacher_logits = teacher_outputs.logits

        # Forward pass through the student model
        student_outputs = student_model(input_ids, attention_mask=attention_mask)
        student_logits = student_outputs.logits

        # Compute the distillation loss
        loss = distillation_loss(student_logits, teacher_logits, labels, temperature, alpha)

        # Backpropagation
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

    print(f"Epoch {epoch + 1} completed with loss: {loss.item()}")

    # Evaluate the teacher model
    teacher_accuracy, teacher_precision, teacher_recall, teacher_f1 = evaluate_model(teacher_model, test_dataloader, device)
    print(f"Teacher (test) - Accuracy: {teacher_accuracy:.4f}, Precision: {teacher_precision:.4f}, Recall: {teacher_recall:.4f}, F1 Score: {teacher_f1:.4f}")

    # Evaluate the student model
    student_accuracy, student_precision, student_recall, student_f1 = evaluate_model(student_model, test_dataloader, device)
    print(f"Student (test) - Accuracy: {student_accuracy:.4f}, Precision: {student_precision:.4f}, Recall: {student_recall:.4f}, F1 Score: {student_f1:.4f}")
    print("\n")

    # put student model back into train mode
    student_model.train()

#Compare the models
# create testing data loader
validation_dataloader = DataLoader(tokenized_data['test'], batch_size=8, collate_fn=data_collator)
# Evaluate the teacher model
teacher_accuracy, teacher_precision, teacher_recall, teacher_f1 = evaluate_model(teacher_model, validation_dataloader, device)
print(f"Teacher (validation) - Accuracy: {teacher_accuracy:.4f}, Precision: {teacher_precision:.4f}, Recall: {teacher_recall:.4f}, F1 Score: {teacher_f1:.4f}")
# Evaluate the student model
student_accuracy, student_precision, student_recall, student_f1 = evaluate_model(student_model, validation_dataloader, device)
print(f"Student (validation) - Accuracy: {student_accuracy:.4f}, Precision: {student_precision:.4f}, Recall: {student_recall:.4f}, F1 Score: {student_f1:.4f}")
     

st.write('Pushing model to huggingface')

# Push model to huggingface
hf_name = 'CarolXia' # your hf username or org name
mode_name = "pii-kd-deberta-v2"
model_id = hf_name + "/" + mode_name
student_model.push_to_hub(model_id, token=st.secrets["HUGGINGFACE_TOKEN"])