flan-t5-base-imdb-text-classification

This model is a fine-tuned version of google/flan-t5-base on the IMDB dataset. It achieves the following results on the evaluation set:

• Loss: 0.0767
• F1: 95.084
• Gen Len: 2.4976

              precision    recall  f1-score   support

           0       0.97      0.88      0.92     12500
           1       0.89      0.97      0.93     12500

    accuracy                           0.93     25000
   macro avg       0.93      0.93      0.93     25000
weighted avg       0.93      0.93      0.93     25000

Model description

In this implementation, using the Flan T5 large language model, we performed the Text Classification task on the IMDB dataset and obtained a very good accuracy of 93%.

Training and evaluation data

This model was trained on the imdb train dataset with 25,000 data and then tested and evaluated on the imdb test dataset with 25,000 data.

Usage

1. Install dependencies

!pip install transformers==4.28.1 datasets

2. Load IMDB Corpus

from datasets import load_dataset

dataset_id = "imdb"

# Load dataset from the hub
dataset = load_dataset(dataset_id)

3. Load fine tune flan t5 model

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("mohammadtaghizadeh/flan-t5-base-imdb-text-classification")
model = AutoModelForSeq2SeqLM.from_pretrained("mohammadtaghizadeh/flan-t5-base-imdb-text-classification")
model.to('cuda')

4. Test the model

from tqdm.auto import tqdm

samples_number = len(dataset['test'])
progress_bar = tqdm(range(samples_number))
predictions_list = []
labels_list = []
for i in range(samples_number):
  text = dataset['test']['text'][i]
  inputs = tokenizer.encode_plus(text, padding='max_length', max_length=512, return_tensors='pt').to('cuda')
  outputs = model.generate(inputs['input_ids'], attention_mask=inputs['attention_mask'], max_length=150, num_beams=4, early_stopping=True)
  prediction = tokenizer.decode(outputs[0], skip_special_tokens=True)
  predictions_list.append(prediction)
  labels_list.append(dataset['test']['label'][i])

  progress_bar.update(1)

5. Classification report

from sklearn.metrics import classification_report

str_labels_list = []
for i in range(len(labels_list)): str_labels_list.append(str(labels_list[i]))

report = classification_report(str_labels_list, predictions_list)
print(report)

Output

              precision    recall  f1-score   support

           0       0.97      0.88      0.92     12500
           1       0.89      0.97      0.93     12500

    accuracy                           0.93     25000
   macro avg       0.93      0.93      0.93     25000
weighted avg       0.93      0.93      0.93     25000

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 2

Training results

Training Loss	Epoch	Step
0.100500	1.0	3125
0.043600	2.0	6250

Framework versions

• Transformers 4.28.1
• Pytorch 2.0.0+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3