-
Notifications
You must be signed in to change notification settings - Fork 1
/
app.py
56 lines (40 loc) · 2.02 KB
/
app.py
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
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import GPT2Tokenizer
from classification_inference import inference, init_model
from attacker_model_inference import *
import torch
app = FastAPI()
app.add_middleware(
CORSMiddleware,
allow_origins=["*"], # You can specify a list of allowed origins or use "*" to allow all
allow_credentials=True,
allow_methods=["GET", "POST", "PUT", "DELETE", "OPTIONS"],
allow_headers=["*"],
)
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"
device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
inf_model = init_model(tokenizer, device)
inf_model.eval()
class InputData(BaseModel):
input_text: str
@app.get("/")
def home():
return {}
@app.post("/inference")
def run_inference(input_data: InputData):
# Perform inference on the input_text
result = inference(inf_model, tokenizer, device, input_data.input_text)
return {"result": result}
@app.post("/adversarial_example")
def gen_adversarial_example(input_data: InputData, input_label: int):
target_model = load_target_model().to(device)
attacker_model = load_attacker_model().to(device)
processed_dataset = process_dataset(tokenizer, input_data.input_text)
tm_embeddings, am_embeddings = load_embeddings(tokenizer, device, target_model, attacker_model)
original_pred, original_text, adversarial_text, adversarial_pred, precision, recall, f1_score = train(target_model, attacker_model, processed_dataset['input_ids'], label = torch.tensor(input_label, dtype=torch.long), am_embeddings = am_embeddings.to(device), tm_embeddings = tm_embeddings.to(device), device = device, tokenizer = tokenizer)
return {"original_text": original_text, "original_pred": original_pred.item(), "adversarial_text": adversarial_text, "adversarial_pred": adversarial_pred.item(), "precision": precision, "recall": recall, "f1_score": f1_score}