gpt2-finetuned-headliner

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

GPT2 Fine Tuned Headline Generator

• This model is trained on the harvard/abcnews-dataset to generate news headlines
• This model is a fine-tuned version of openai-community/gpt2-medium on an unknown dataset.

Model description

The model is fine-tuned for 2 epochs and 4k training samples from the abcnews dataset. This enables the model to generate news headline like text given a simple prompt

Intended uses & limitations

This model is only for learning purposes only. The model easily hallucinates people names, locations and other artifacts & incidents.

Training and evaluation data

The model leverages 2k test samples for evaluation

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 4
• num_epochs: 2

Training results

The final output after 2 epochs is as follows: TrainOutput(global_step=130, training_loss=5.044873604407678, metrics={'train_runtime': 140.587, 'train_samples_per_second': 59.166, 'train_steps_per_second': 0.925, 'total_flos': 248723096358912.0, 'train_loss': 5.044873604407678, 'epoch': 2.0})

Framework versions

• Transformers 4.44.2
• Pytorch 2.3.1
• Datasets 2.21.0
• Tokenizers 0.19.1