model.py

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

class EmotionModel:
    def __init__(self):
        self.model_name = "joeddav/distilbert-base-uncased-go-emotions-student"
        self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
        self.model = AutoModelForSequenceClassification.from_pretrained(self.model_name)
        self.labels = self.model.config.id2label

    def predict(self, text):
        inputs = self.tokenizer(text, return_tensors="pt", truncation=True, padding=True)
        with torch.no_grad():
            logits = self.model(**inputs).logits
        probs = torch.sigmoid(logits)[0]
        return {
            self.labels[i]: float(probs[i])
            for i in range(len(probs)) if probs[i] > 0.4
        }

class SuicidalIntentModel:
    def __init__(self):
        self.model_name = "sentinet/suicidality"
        self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
        self.model = AutoModelForSequenceClassification.from_pretrained(self.model_name)

    def _score_text(self, text):
        inputs = self.tokenizer(text, return_tensors="pt", truncation=True, padding=True)
        with torch.no_grad():
            logits = self.model(**inputs).logits
        probs = torch.nn.functional.softmax(logits, dim=1)
        return float(probs[0][1])  # Probability of suicidal intent

    def predict(self, text, window_size=20, stride=10):
        tokens = self.tokenizer.tokenize(text)
        if len(tokens) <= window_size:
            return self._score_text(text)

        scores = []
        for i in range(0, len(tokens) - window_size + 1, stride):
            window_tokens = tokens[i:i + window_size]
            window_text = self.tokenizer.convert_tokens_to_string(window_tokens)
            score = self._score_text(window_text)
            scores.append(score)

        return max(scores, default=0.0)