|
--- |
|
license: mit |
|
base_model: openai-community/gpt2 |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: gpt-2-finetuned-wikitext2 |
|
results: [] |
|
--- |
|
|
|
<!-- 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. --> |
|
|
|
# gpt-2-finetuned-wikitext2 |
|
|
|
This model is a fine-tuned version of [openai-community/gpt2](https://huggingface.co/openai-community/gpt2) on an unknown dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 3.3924 |
|
|
|
## Model Description |
|
|
|
This language model is built on the GPT-2 architecture provided by OpenAI. The tokenizer utilized for preprocessing text data is OpenAI's tikToken. For more details on tikToken, you can refer to the [official GitHub repository](https://github.com/openai/tiktoken). |
|
|
|
### Tokenizer Overview |
|
|
|
To interactively explore the functionality and behavior of the tikToken tokenizer, you can use the [tikToken interactive website](https://tiktokenizer.vercel.app/). This website allows you to quickly visualize the tokenization process and understand how the tokenizer segments input text into tokens. |
|
|
|
### Model Checkpoint |
|
|
|
The model checkpoint used in this implementation is sourced from the OpenAI community and is based on the GPT-2 architecture. You can find the specific model checkpoint at the following Hugging Face Model Hub link: [openai-community/gpt2](https://huggingface.co/openai-community/gpt2). |
|
|
|
### Training Details |
|
|
|
The model was trained for a total of 3 epochs on the provided dataset. This information reflects the number of times the entire training dataset was processed during the training phase. Training for a specific number of epochs helps control the duration and scope of the model's learning process. |
|
|
|
|
|
## Training and evaluation data |
|
|
|
#### Evaluation Data |
|
|
|
For evaluating the model's performance, the training script utilized an evaluation dataset. |
|
|
|
#### Evaluation Results |
|
|
|
After training, the model's performance was assessed using the evaluation dataset. The perplexity, a common metric for language modeling tasks was **Perplexity: 29.74** |
|
|
|
```python |
|
eval_results = trainer.evaluate() |
|
print(f"Perplexity: {math.exp(eval_results['eval_loss']):.2f}") |
|
|
|
>>> Perplexity : 29.74 |
|
``` |
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 2e-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 |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | |
|
|:-------------:|:-----:|:----:|:---------------:| |
|
| 3.4934 | 1.0 | 2334 | 3.4145 | |
|
| 3.3567 | 2.0 | 4668 | 3.3953 | |
|
| 3.2968 | 3.0 | 7002 | 3.3924 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.37.2 |
|
- Pytorch 2.1.0+cu121 |
|
- Datasets 2.17.1 |
|
- Tokenizers 0.15.2 |
|
|
