fake-gpt-2-17m
This model is a GPTJ (with 17,637,632 parameters) trained from scratch on a synthetic dataset (1gb of documents created in 4 fake languages, each with a formal and informal writing style) for 1 epoch.
It achieves the following results on the evaluation set:
- Loss: 3.5592
Intended uses & limitations
This model is to be used as a base model for fine-tuning any language/task to probe the effectiveness of both pre-training on an algorithmically generated corpus and effectiveness of extremely small language models (SLMs?). It can only generate text based on its training data (which will be uploaded as a huggingface dataset soon).
Training and evaluation data
More information needed
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.001
- batch_size 64
- seed: 42
- optimizer: Adam
- lr_scheduler_type: linear
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 3.5175 | 1.0 | 46857 | 3.5592 |
Framework versions
- Transformers 4.22.1
- Pytorch 1.12.0
- Datasets 2.3.2
- Tokenizers 0.12.1