You can deploy this model and start earning money today!

joel-crasto/
TEXT-01

generated_from_trainer

autotrain_compatible

text-generation-inference

endpoints_compatible

Views

No views yet

FACT-1

This model is based on the GPT-2 model shape, trained from scratch on the wikimedia/wikipedia dataset. It achieves the following results on the evaluation set:

eval_loss: 2.0494
eval_runtime: 443.6756
eval_samples_per_second: 225.39
eval_steps_per_second: 28.174
epoch: 4.2665
step: 60000

Model description

It is purely an experimental model for learning purposes, and due to the small dataset and train time, it may function as autocomplete with more tuning, but in its current state does not produce coherent sentences.

Intended uses & limitations

Not intended for use.

Sample prompt:

generator = pipeline("text-generation", model=model, tokenizer=tokenizer, device=0)
print(generator("Planes are designed", 
    max_length=100,
    do_sample=True,
    top_p=0.9,    
    temperature=0.8,   
    repetition_penalty=1.2, 
    no_repeat_ngram_size=2 ))

Sample response:

[{'generated_text': 'Planes are designed to to be\n1., for in-Aoonoono...'}]

Training and evaluation data

Subset of wikimedia/wikipedia dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 64
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 1000
training_steps: 100000
mixed_precision_training: Native AMP

Framework versions

Transformers 4.56.0
Pytorch 2.5.1+cu121
Datasets 4.0.0
Tokenizers 0.22.0

joel-crasto/TEXT-01