|
|
from typing import Dict, Any |
|
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
import torch |
|
|
import re |
|
|
|
|
|
class EndpointHandler(): |
|
|
def __init__(self, path=""): |
|
|
self.tokenizer = AutoTokenizer.from_pretrained(path) |
|
|
self.model = AutoModelForSequenceClassification.from_pretrained(path) |
|
|
self.model.eval() |
|
|
self.id2label = {0: "Human", 1: "Mixed", 2: "AI"} |
|
|
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
|
|
self.model.to(self.device) |
|
|
|
|
|
def split_into_sentences(self, text: str): |
|
|
sentences = re.split(r'(?<=[.!?])\s+', text) |
|
|
return [s.strip() for s in sentences if s.strip()] |
|
|
|
|
|
def get_token_predictions(self, text: str): |
|
|
tokens = self.tokenizer.tokenize(text) |
|
|
token_predictions = [] |
|
|
for i in range(len(tokens)): |
|
|
start = max(0, i - 10) |
|
|
end = min(len(tokens), i + 10) |
|
|
context = self.tokenizer.convert_tokens_to_string(tokens[start:end]) |
|
|
inputs = self.tokenizer(context, return_tensors="pt", truncation=True, max_length=512) |
|
|
inputs = {k: v.to(self.device) for k, v in inputs.items()} |
|
|
with torch.no_grad(): |
|
|
outputs = self.model(**inputs) |
|
|
probs = torch.softmax(outputs.logits, dim=1) |
|
|
ai_prob = probs[0][2].item() |
|
|
token = tokens[i].replace("Ġ", " ").replace("▁", " ").replace("Ċ", " ").strip() |
|
|
if token: |
|
|
token_predictions.append({"token": token, "ai_prob": ai_prob}) |
|
|
return token_predictions |
|
|
|
|
|
def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]: |
|
|
text = data.get("inputs", "") |
|
|
|
|
|
inputs = self.tokenizer(text, return_tensors="pt", truncation=True, max_length=512) |
|
|
inputs = {k: v.to(self.device) for k, v in inputs.items()} |
|
|
with torch.no_grad(): |
|
|
outputs = self.model(**inputs) |
|
|
probs = torch.softmax(outputs.logits, dim=1) |
|
|
pred = torch.argmax(probs, dim=1).item() |
|
|
doc_result = { |
|
|
"prediction": self.id2label[pred], |
|
|
"confidence": probs[0][pred].item(), |
|
|
"probabilities": {self.id2label[i]: float(p) for i, p in enumerate(probs[0])} |
|
|
} |
|
|
|
|
|
sentences = self.split_into_sentences(text) |
|
|
sent_results = [] |
|
|
for sent in sentences: |
|
|
inputs = self.tokenizer(sent, return_tensors="pt", truncation=True, max_length=512) |
|
|
inputs = {k: v.to(self.device) for k, v in inputs.items()} |
|
|
with torch.no_grad(): |
|
|
outputs = self.model(**inputs) |
|
|
probs = torch.softmax(outputs.logits, dim=1) |
|
|
pred = torch.argmax(probs, dim=1).item() |
|
|
sent_results.append({ |
|
|
"sentence": sent, |
|
|
"prediction": self.id2label[pred], |
|
|
"confidence": probs[0][pred].item(), |
|
|
"probabilities": {self.id2label[i]: float(p) for i, p in enumerate(probs[0])} |
|
|
}) |
|
|
|
|
|
token_results = self.get_token_predictions(text) |
|
|
return [{ |
|
|
"document": doc_result, |
|
|
"sentences": sent_results, |
|
|
"tokens": token_results |
|
|
}] |
