File size: 6,182 Bytes
7f17fe7
95b43d8
7f17fe7
 
95b43d8
 
e66afc2
44d2bcd
 
 
3cc319e
44d2bcd
3cc319e
44d2bcd
 
 
 
95c8334
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8153817
95c8334
 
 
 
 
 
7f17fe7
1efa28d
7f17fe7
44d2bcd
8153817
 
 
44d2bcd
8153817
44d2bcd
 
7f17fe7
44d2bcd
8153817
 
 
44d2bcd
8153817
44d2bcd
 
7f17fe7
44d2bcd
3cc319e
44d2bcd
6ba018e
1efa28d
8153817
44d2bcd
 
1efa28d
8153817
44d2bcd
8153817
95c8334
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
58648d1
44d2bcd
 
 
8153817
 
44d2bcd
 
8153817
6ba018e
95b43d8
 
 
7f17fe7
95b43d8
 
 
 
1efa28d
8153817
7f17fe7
44d2bcd
95b43d8
44d2bcd
8153817
 
 
 
 
 
44d2bcd
 
 
 
 
8153817
44d2bcd
 
 
 
 
 
 
 
8153817
7f17fe7
95b43d8
 
 
4607c9c
689eabe
95b43d8
 
689eabe
7f17fe7
95b43d8
 
 
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
from fastapi import FastAPI, Request
from pydantic import BaseModel
import torch
import pickle
import gluonnlp as nlp
import numpy as np
import os
import sys
import collections
import logging # λ‘œκΉ… λͺ¨λ“ˆ μž„ν¬νŠΈ

from transformers import AutoTokenizer, BertModel
from torch.utils.data import Dataset, DataLoader
from huggingface_hub import hf_hub_download

logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

class BERTClassifier(torch.nn.Module):
    def __init__(self,
                 bert,
                 hidden_size = 768,
                 num_classes=5, # λΆ„λ₯˜ν•  클래슀 수 (category λ”•μ…”λ„ˆλ¦¬ 크기와 일치)
                 dr_rate=None,
                 params=None):
        super(BERTClassifier, self).__init__()
        self.bert = bert
        self.dr_rate = dr_rate

        self.classifier = torch.nn.Linear(hidden_size , num_classes)
        if dr_rate:
            self.dropout = torch.nn.Dropout(p=dr_rate)

    def gen_attention_mask(self, token_ids, valid_length):
        attention_mask = torch.zeros_like(token_ids)
        for i, v in enumerate(valid_length):
            attention_mask[i][:v] = 1
        return attention_mask.float()

    def forward(self, token_ids, valid_length, segment_ids):
        attention_mask = self.gen_attention_mask(token_ids, valid_length)

        _, pooler = self.bert(input_ids=token_ids, token_type_ids=segment_ids.long(), attention_mask=attention_mask.float().to(token_ids.device), return_dict=False)

        if self.dr_rate:
            out = self.dropout(pooler)
        else:
            out = pooler
        return self.classifier(out)

# --- 2. BERTDataset 클래슀 μ •μ˜ ---
class BERTDataset(Dataset):
    def __init__(self, dataset, sent_idx, label_idx, bert_tokenizer, vocab, max_len, pad, pair):
        transform = nlp.data.BERTSentenceTransform(
            bert_tokenizer, max_seq_length=max_len, vocab=vocab, pad=pad, pair=pair
        )
        self.sentences = [transform([i[sent_idx]]) for i in dataset]
        self.labels = [np.int32(i[label_idx]) for i in dataset]

    def __getitem__(self, i):
        return (self.sentences[i] + (self.labels[i],))

    def __len__(self):
        return len(self.labels)

app = FastAPI()
device = torch.device("cpu") # Hugging Face Spaces의 무료 ν‹°μ–΄λŠ” 주둜 CPUλ₯Ό μ‚¬μš©ν•©λ‹ˆλ‹€.

# βœ… category λ‘œλ“œ
try:
    with open("category.pkl", "rb") as f:
        category = pickle.load(f)
    logger.info("category.pkl λ‘œλ“œ 성곡.")
except FileNotFoundError:
    logger.error("Error: category.pkl νŒŒμΌμ„ 찾을 수 μ—†μŠ΅λ‹ˆλ‹€. ν”„λ‘œμ νŠΈ λ£¨νŠΈμ— μžˆλŠ”μ§€ ν™•μΈν•˜μ„Έμš”.")
    sys.exit(1)

# βœ… vocab λ‘œλ“œ
try:
    with open("vocab.pkl", "rb") as f:
        vocab = pickle.load(f)
    logger.info("vocab.pkl λ‘œλ“œ 성곡.")
except FileNotFoundError:
    logger.error("Error: vocab.pkl νŒŒμΌμ„ 찾을 수 μ—†μŠ΅λ‹ˆλ‹€. ν”„λ‘œμ νŠΈ λ£¨νŠΈμ— μžˆλŠ”μ§€ ν™•μΈν•˜μ„Έμš”.")
    sys.exit(1)

# βœ… ν† ν¬λ‚˜μ΄μ € λ‘œλ“œ
tokenizer = AutoTokenizer.from_pretrained('skt/kobert-base-v1')
logger.info("ν† ν¬λ‚˜μ΄μ € λ‘œλ“œ 성곡.")

# βœ… λͺ¨λΈ λ‘œλ“œ (Hugging Face Hubμ—μ„œ λ‹€μš΄λ‘œλ“œ)
try:
    HF_MODEL_REPO_ID = "hiddenFront/TextClassifier"
    HF_MODEL_FILENAME = "textClassifierModel.pt"
    
    model_path = hf_hub_download(repo_id=HF_MODEL_REPO_ID, filename=HF_MODEL_FILENAME)
    logger.info(f"λͺ¨λΈ 파일이 '{model_path}'에 μ„±κ³΅μ μœΌλ‘œ λ‹€μš΄λ‘œλ“œλ˜μ—ˆμŠ΅λ‹ˆλ‹€.")

    bert_base_model = BertModel.from_pretrained('skt/kobert-base-v1')
    model = BERTClassifier(
        bert_base_model,
        dr_rate=0.5, # ν•™μŠ΅ μ‹œ μ‚¬μš©λœ dr_rate κ°’μœΌλ‘œ λ³€κ²½ν•˜μ„Έμš”.
        num_classes=len(category)
    )

    loaded_state_dict = torch.load(model_path, map_location=device)
    
    new_state_dict = collections.OrderedDict()
    for k, v in loaded_state_dict.items():
        name = k
        if name.startswith('module.'):
            name = name[7:]
        new_state_dict[name] = v
    
    model.load_state_dict(new_state_dict, strict=False)
    model.to(device)
    model.eval()
    logger.info("λͺ¨λΈ λ‘œλ“œ 성곡.")

except Exception as e:
    logger.error(f"Error: λͺ¨λΈ λ‹€μš΄λ‘œλ“œ λ˜λŠ” λ‘œλ“œ 쀑 였λ₯˜ λ°œμƒ: {e}")
    sys.exit(1)


# βœ… 데이터셋 생성에 ν•„μš”ν•œ νŒŒλΌλ―Έν„°
max_len = 64
batch_size = 32

# βœ… 예츑 ν•¨μˆ˜
def predict(predict_sentence):
    data = [predict_sentence, '0']
    dataset_another = [data]
    another_test = BERTDataset(dataset_another, 0, 1, tokenizer.tokenize, vocab, max_len, True, False)
    test_dataLoader = DataLoader(another_test, batch_size=batch_size, num_workers=0)

    model.eval()

    with torch.no_grad():
        for batch_id, (token_ids, valid_length, segment_ids, label) in enumerate(test_dataLoader):
            token_ids = token_ids.long().to(device)
            segment_ids = segment_ids.long().to(device)
            
            out = model(token_ids, valid_length, segment_ids)
            
            logits = out # λͺ¨λΈμ˜ 직접 좜λ ₯은 λ‘œμ§“μž…λ‹ˆλ‹€.
            probs = torch.nn.functional.softmax(logits, dim=1) # ν™•λ₯  계산

            predicted_category_index = torch.argmax(probs, dim=1).item() # 예츑 인덱슀
            predicted_category_name = list(category.keys())[predicted_category_index] # 예츑 μΉ΄ν…Œκ³ λ¦¬ 이름
            
            # --- 예츑 상세 λ‘œκΉ… ---
            logger.info(f"Input Text: '{predict_sentence}'")
            logger.info(f"Raw Logits: {logits.tolist()}")
            logger.info(f"Probabilities: {probs.tolist()}")
            logger.info(f"Predicted Index: {predicted_category_index}")
            logger.info(f"Predicted Label: '{predicted_category_name}'")
            # --- 예츑 상세 λ‘œκΉ… 끝 ---

            return predicted_category_name

# βœ… μ—”λ“œν¬μΈνŠΈ μ •μ˜
class InputText(BaseModel):
    text: str

@app.get("/")
def root():
    return {"message": "Text Classification API (KoBERT)"}

@app.post("/predict")
async def predict_route(item: InputText):
    result = predict(item.text)
    return {"text": item.text, "classification": result}