from transformers import pipeline

question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="riswanahamed/SmolLM2-FT-MyDataset", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training procedure

Training Methods

What I Did

I fine-tuned a pre-trained language model using the Hugging Face transformers library. The base model was adapted to perform better on specific task by training it on a domain-specific dataset.

How I Did It

Fine-Tuning Setup:

Configured the model training parameters, including the learning rate, batch size, and number of steps.
Used SFTTrainer from Hugging Face for seamless training with built-in evaluation capabilities.
Trained the model for 1 epoch to prevent overfitting, as the dataset was relatively small and hardware resources were limited.

Training Environment:

The training was performed in Google Colab using a CPU/GPU environment.
Adjusted batch sizes and learning rates to balance between performance and available resources.

Evaluation:
- Monitored training loss and validation loss at regular intervals to ensure the model was learning effectively.
- Evaluated the model using metrics like [accuracy, F1 score, or other task-specific metrics].
Saving the Model:
- The fine-tuned model was saved to a specified output directory for reuse.

What the User Should Do

Use the Model:
- Load the model using the Hugging Face transformers library.
- Tokenize your inputs and pass them to the model for inference.
- If your task or domain differs, fine-tune the model further on your dataset.
- Follow the same process: prepare the dataset, set training configurations, and monitor evaluation metrics.
Experiment with Parameters:
- If you have access to better hardware, experiment with larger batch sizes or additional epochs to improve results.
- Use hyperparameter tuning to find the best configuration for your use case.

Framework versions

TRL: 0.13.0
Transformers: 4.47.1
Pytorch: 2.5.1+cu121
Datasets: 3.2.0
Tokenizers: 0.21.0

Citations

Cite TRL as:

bibtex

@misc{vonwerra2022trl,
	title        = {{TRL: Transformer Reinforcement Learning}},
	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
	year         = 2020,
	journal      = {GitHub repository},
	publisher    = {GitHub},
	howpublished = {\url{https://github.com/huggingface/trl}}
}