Model Card for mt5-base-multi-label-en-iiib-02c

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

Model Description

The model was fine-tuned on a dataset of English Instant Messenger dialogs of Adolescents. The classification is multi-label and the model outputs any combination of the tags:'NO TAG', 'Informační podpora', 'Emocionální podpora', 'Začlenění do skupiny', 'Uznání', 'Nabídka pomoci': as a string joined with ', ' (ordered alphabetically). Each label indicates the presence of that category of Supportive Interactions: 'no tag', 'informational support', 'emocional support', 'social companionship', 'appraisal', 'instrumental support'. The inputs of the model are: a target utterance and its bi-directional context; the label of the example is determined by the label of the target utterance.

Developed by: Anonymous
Language(s): multilingual
Finetuned from: mt5-base

Model Sources

Repository: https://github.com/chi2024submission
Paper: Stay tuned!

Usage

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

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch

test_texts = ['Utterance2']
test_text_pairs = ['Utterance1;Utterance2;Utterance3']

checkpoint_path = "chi2024/mt5-base-multi-label-en-iiib-02c"
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint_path)\
    .to("cuda" if torch.cuda.is_available() else "cpu")
tokenizer = AutoTokenizer.from_pretrained(checkpoint_path)

def verbalize_input(text: str, text_pair: str) -> str:
    return "Utterance: %s\nContext: %s" % (text, text_pair)

def predict_one(text, pair):
    input_pair = verbalize_input(text, pair)
    inputs = tokenizer(input_pair, return_tensors="pt", padding=True,
                       truncation=True, max_length=256).to(model.device)
    outputs = model.generate(**inputs)
    decoded = [text.split(",")[0].strip() for text in
               tokenizer.batch_decode(outputs, skip_special_tokens=True)]
    return decoded

dec = predict_one(test_texts[0], test_text_pairs[0])
print(dec)