Model Card for small-e-czech-multi-label-online-risks-cs

This model is fine-tuned for multi-label text classification of Online Risks in Instant Messenger dialogs of Adolescents.

Model Description

The model was fine-tuned on a dataset of Instant Messenger dialogs of Adolescents. The classification is multi-label and the model outputs probablities for labels {0,1,2,3,4,5}:

0. None
1. Aggression, Harassing, Hate
2. Mental Health Problems
3. Alcohol, Drugs
4. Weight Loss, Diets
5. Sexual Content

• Developed by: Anonymous
• Language(s): cs
• Finetuned from: small-e-czech

Model Sources

• Repository: https://github.com/justtherightsize/supportive-interactions-and-risks
• Paper: Stay tuned!

Usage

Here is how to use this model to classify a context-window of a dialogue:

import numpy as np
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# Prepare input texts. This model is pretrained on multi-lingual data 
# and fine-tuned on English
test_texts = ['Utterance1;Utterance2;Utterance3']

# Load the model and tokenizer
model = AutoModelForSequenceClassification.from_pretrained(
    'justtherightsize/small-e-czech-multi-label-online-risks-cs', num_labels=6).to("cuda")

tokenizer = AutoTokenizer.from_pretrained(
    'justtherightsize/small-e-czech-multi-label-online-risks-cs',
    use_fast=False, truncation_side='left')
assert tokenizer.truncation_side == 'left'

# Define helper functions
def predict_one(text: str, tok, mod, threshold=0.5):
    encoding = tok(text, return_tensors="pt", truncation=True, padding=True,
                   max_length=256)
    encoding = {k: v.to(mod.device) for k, v in encoding.items()}
    outputs = mod(**encoding)
    logits = outputs.logits
    sigmoid = torch.nn.Sigmoid()
    probs = sigmoid(logits.squeeze().cpu())
    predictions = np.zeros(probs.shape)
    predictions[np.where(probs >= threshold)] = 1
    return predictions, probs

def print_predictions(texts):
    preds = [predict_one(tt, tokenizer, model) for tt in texts]
    for c, p in preds:
        print(f'{c}: {p.tolist():.4f}')

# Run the prediction
print_predictions(test_texts)