README.md

metadata

license: apache-2.0
language: en
datasets:
  - sst2
metrics:
  - precision
  - recall
  - f1
tags:
  - text-classification

T5-base fine-tuned for Sentiment Analysis 👍👎

OpenAI's GPT-2 medium fine-tuned on SST-2 dataset for Sentiment Analysis downstream task.

Details of T5

The GPT-2 model was presented in Language Models are Unsupervised Multitask Learners by Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever

Model fine-tuning 🏋️‍

The model has been finetuned for 10 epochs on standard hyperparameters

Val set metrics 🧾

           |precision | recall  | f1-score |support|
|----------|----------|---------|----------|-------|
|negative  |     0.92 |     0.92|      0.92|   428 |
|positive  |     0.92 |     0.93|      0.92|   444 |
|----------|----------|---------|----------|-------|
|accuracy|            |         |      0.92|   872 |
|macro avg|       0.92|     0.92|      0.92|   872 |
|weighted avg|    0.92|     0.92|      0.92|   872 |

Model in Action 🚀

from transformers import GPT2Tokenizer, GPT2ForSequenceClassification

tokenizer = GPT2Tokenizer.from_pretrained("michelecafagna26/gpt2-medium-finetuned-sst2-sentiment")
model = GPT2ForSequenceClassification.from_pretrained("michelecafagna26/gpt2-medium-finetuned-sst2-sentiment")

inputs = tokenizer("I love it", return_tensors="pt")

model(**inputs).logits.argmax(axis=1)

# 1: Positive, 0: Negative
# Output: tensor([1])

This model card is based on "mrm8488/t5-base-finetuned-imdb-sentiment" by Manuel Romero/@mrm8488