shakespeare-distilgpt2

This model is a fine-tuned version of distilbert/distilgpt2 on the shakespeare-lines dataset. It achieves the following results on the evaluation set:

• Loss: 4.2490
• Perplexity: 74.01

Training and evaluation data

The training and evaluation data was taken from the shakespeare-lines dataset. The dataset was shuffled with a seed of 24, and split into training and evaluation with a ratio of 80:20.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("benchaffe/shakespeare-distilgpt2")
tokenizer = AutoTokenizer.from_pretrained("benchaffe/shakespeare-distilgpt2")

prompt = "What light through yonder window breaks"
inputs = tokenizer(prompt, return_tensors="pt")

outputs = model.generate(
    **inputs,
    max_length=80,
    temperature=0.8,
    top_p=0.95,
    do_sample=True
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss
3.9479	1.0	22941	4.2781
3.7527	2.0	45882	4.2111
3.5778	3.0	68823	4.2035
3.4214	4.0	91764	4.2129
3.3513	5.0	114705	4.2490

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1