RichardErkhov's picture
uploaded readme
c0000ef verified

Quantization made by Richard Erkhov.

Github

Discord

Request more models

dadjokes-tuned-opt - AWQ

Original model description:

license: mit base_model: facebook/opt-350m tags: - trl - sft - gnumanth/dadjokes-trained-opt model-index: - name: tmp_trainer results: [] datasets: - gnumanth/dad-jokes language: - en pipeline_tag: text-generation widget: - text: "joke"

This model is a fine-tuned version of facebook/opt-350m on an gnumanth/dad-jokes dataset.

Model description

SFT Trained simple model for fun!

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-05
  • 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: 3.0

Training results

TrainOutput(global_step=18, training_loss=2.2378472222222223, metrics={'train_runtime': 149.7511, 'train_samples_per_second': 0.881, 'train_steps_per_second': 0.12, 'total_flos': 9828797644800.0, 'train_loss': 2.2378472222222223, 'epoch': 3.0})

Framework versions

  • Transformers 4.38.1
  • Pytorch 2.1.0+cu121
  • Datasets 2.17.1
  • Tokenizers 0.15.1