metadata

license: apache-2.0
datasets:
  - ruanchaves/hatebr
language:
  - pt
metrics:
  - accuracy
library_name: transformers
pipeline_tag: text-classification
tags:
  - hate-speech
widget:
  - text: Não concordo com a sua opinião.
    example_title: Exemplo
  - text: Pega a sua opinião e vai a merda com ela!
    example_title: Exemplo

TeenyTinyLlama-460m-HateBR

TeenyTinyLlama is a series of small foundational models trained in Brazilian Portuguese.

This repository contains a version of TeenyTinyLlama-460m (TeenyTinyLlama-460m-HateBR) fine-tuned on the HateBR dataset.

Details

Number of Epochs: 3
Batch size: 16
Optimizer: torch.optim.AdamW (learning_rate = 4e-5, epsilon = 1e-8)
GPU: 1 NVIDIA A100-SXM4-40GB

Usage

Using transformers.pipeline:

from transformers import pipeline

text = "Pega a sua opinião e vai a merda com ela!"

classifier = pipeline("text-classification", model="nicholasKluge/TeenyTinyLlama-460m-HateBR")
classifier(text)

# >>> [{'label': 'TOXIC', 'score': 0.9998729228973389}]

Reproducing

To reproduce the fine-tuning process, use the following code snippet:


# Hatebr
! pip install transformers datasets evaluate accelerate -q

import evaluate
import numpy as np
from huggingface_hub import login
from datasets import load_dataset, Dataset, DatasetDict
from transformers import AutoTokenizer, DataCollatorWithPadding
from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer

# Load the task
dataset = load_dataset("ruanchaves/hatebr")

# Format the dataset
train = dataset['train'].to_pandas()
train = train[['instagram_comments', 'offensive_language']]
train.columns = ['text', 'labels']
train.labels = train.labels.astype(int)
train = Dataset.from_pandas(train)

test = dataset['test'].to_pandas()
test = test[['instagram_comments', 'offensive_language']]
test.columns = ['text', 'labels']
test.labels = test.labels.astype(int)
test = Dataset.from_pandas(test)

dataset = DatasetDict({
    "train": train,  
    "test": test                  
})

# Create a `ModelForSequenceClassification`
model = AutoModelForSequenceClassification.from_pretrained(
    "nicholasKluge/TeenyTinyLlama-460m", 
    num_labels=2, 
    id2label={0: "NONTOXIC", 1: "TOXIC"}, 
    label2id={"NONTOXIC": 0, "TOXIC": 1}
)

tokenizer = AutoTokenizer.from_pretrained("nicholasKluge/TeenyTinyLlama-460m")

# Preprocess the dataset
def preprocess_function(examples):
    return tokenizer(examples["text"], truncation=True)

dataset_tokenized = dataset.map(preprocess_function, batched=True)

# Create a simple data collactor
data_collator = DataCollatorWithPadding(tokenizer=tokenizer)

# Use accuracy as evaluation metric
accuracy = evaluate.load("accuracy")

# Function to compute accuracy
def compute_metrics(eval_pred):
    predictions, labels = eval_pred
    predictions = np.argmax(predictions, axis=1)
    return accuracy.compute(predictions=predictions, references=labels)

# Define training arguments
training_args = TrainingArguments(
    output_dir="checkpoints",
    learning_rate=4e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
    evaluation_strategy="epoch",
    save_strategy="epoch",
    load_best_model_at_end=True,
    push_to_hub=True,
    hub_token="your_token_here",
    hub_model_id="username/model-ID",
)

# Define the Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=dataset_tokenized["train"],
    eval_dataset=dataset_tokenized["test"],
    tokenizer=tokenizer,
    data_collator=data_collator,
    compute_metrics=compute_metrics,
)

# Train!
trainer.train()

Fine-Tuning Comparisons

Models	HateBr
Teeny Tiny Llama 460m	91.64
Bert-large-portuguese-cased	91.57
Bert-base-portuguese-cased	91.28
Teeny Tiny Llama 160m	90.71
Gpt2-small-portuguese	87.42

Cite as 🤗


@misc{nicholas22llama,
  doi = {10.5281/zenodo.6989727},
  url = {https://huggingface.co/nicholasKluge/TeenyTinyLlama-460m},
  author = {Nicholas Kluge Corrêa},
  title = {TeenyTinyLlama},
  year = {2023},
  publisher = {HuggingFace},
  journal = {HuggingFace repository},
}

Funding

This repository was built as part of the RAIES (Rede de Inteligência Artificial Ética e Segura) initiative, a project supported by FAPERGS - (Fundação de Amparo à Pesquisa do Estado do Rio Grande do Sul), Brazil.

License

TeenyTinyLlama-460m-HateBR is licensed under the Apache License, Version 2.0. See the LICENSE file for more details.