Model Card for SmolLM2-360M-GRPO-v1

This model is a fine-tuned version of HuggingFaceTB/SmolLM2-360M. It has been trained using TRL and using the lamini/taylor_swift dataset.

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="sky-2002/SmolLM2-360M-GRPO-v1", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Evals

Referring this blog post, used a similar evaluation method.

However, since llm-judge was being used in one of the reward functions, I tried with different models as judges and observed the changes.

Model	Average ROUGE-L	LLM-Judge Model
Qwen-0.5B finetuned	0.3313	Qwen-0.5B
SmolLM2-360M-GRPO-v0	0.1644	llama3.2:1B
SmolLM2-360M-GRPO-v1	0.1672	deepseek-r1:1.5b

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="sky-2002/SmolLM2-360M-GRPO-v1", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training procedure

This model was trained with GRPO, a method introduced in DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models.

Framework versions

• TRL: 0.15.0.dev0
• Transformers: 4.49.0.dev0
• Pytorch: 2.5.1
• Datasets: 3.2.0
• Tokenizers: 0.21.0

Citations

Cite GRPO as:

@article{zhihong2024deepseekmath,
    title        = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
    author       = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
    year         = 2024,
    eprint       = {arXiv:2402.03300},
}

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}}
}