Model Card for llm-course-hw2-reward-model

This model is a fine-tuned version of HuggingFaceTB/SmolLM-135M-Instruct on the HumanLLMs/Human-Like-DPO-Dataset dataset. It is trained as a reward model for PPO using TRL library as a part of homework on VK NLP Course. . The model is of class AutoModelForSequenceClassification with num_labels=1: only last linear layer is trained, and the rest of the model remains unchanged.

Quick start

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="X1716/llm-course-hw2-reward-model", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training procedure

Training logs are available here:

This model was trained with Reward.

Framework versions

• TRL: 0.16.0
• Transformers: 4.50.0
• Pytorch: 2.3.0a0+6ddf5cf85e.nv24.4
• Datasets: 3.4.1
• Tokenizers: 0.21.1

Citations

Cite TRL as:

@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}}
}