Hugging Face Sentiment Analysis Fine Tuning Example

Steps to Use This Model and API

• pip install -r requirements.txt
• python train.py This will train the model and place the trained model data in the qa-final-model/ directory. Wait for training to complete.
• In the terminal enter: uvicorn api:app --reload This will serve the API.
• Use the cURL below to make a request to the API.

Results From Last Training

The first epoch is the selected checkpoint that the API will use as it had the lowest loss while maintaning comperable accuracy to the other epochs.

{
  'eval_loss': 0.21767772734165192,
  'eval_accuracy': 0.9212,
  'eval_f1': 0.9216076402705929,
  'eval_runtime': 14.7899,
  'eval_samples_per_second': 169.034,
  'eval_steps_per_second': 21.163,
  'epoch': 1.0
}
{
  'eval_loss': 0.28838035464286804,
  'eval_accuracy': 0.924,
  'eval_f1': 0.9230145867098866,
  'eval_runtime': 12.8606,
  'eval_samples_per_second': 194.392,
  'eval_steps_per_second': 24.338,
  'epoch': 2.0
}
{
  'eval_loss': 0.35869795083999634,
  'eval_accuracy': 0.9276,
  'eval_f1': 0.9280889948351212,
  'eval_runtime': 12.9623,
  'eval_samples_per_second': 192.867,
  'eval_steps_per_second': 24.147,
  'epoch': 3.0
}
{
  'eval_loss': 0.4029352068901062,
  'eval_accuracy': 0.928,
  'eval_f1': 0.9271844660194175,
  'eval_runtime': 12.8943,
  'eval_samples_per_second': 193.884,
  'eval_steps_per_second': 24.274,
  'epoch': 4.0
} 

Sample Postman Request

curl --location 'http://127.0.0.1:8000/model' \
--header 'Content-Type: application/json' \
--data '{
    "sequence": ["I loved that movie!", "That movie was not good."]
}'