DAPO_E2H-countdown-gaussian_0p5_0p5/README.md

base_model, datasets, library_name, model_name, tags, licence

base_model	datasets	library_name	model_name	tags	licence
Qwen/Qwen2.5-1.5B-Instruct	gsm8k-dataset	transformers	Qwen2.5-1.5B-Instruct_math_grpo_cosine_0.5_0.5_SEC0.3DRO1.0G0.0_minpTrue_1600	generated_from_trainer trl grpo	license

Model Card for Qwen2.5-1.5B-Instruct_math_grpo_cosine_0.5_0.5_SEC0.3DRO1.0G0.0_minpTrue_1600

This model is a fine-tuned version of Qwen/Qwen2.5-1.5B-Instruct on the Countdown dataset. It has been trained using E2H on the top of TRL.

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="shubhamprshr/Qwen2.5-1.5B-Instruct_math_grpo_cosine_0.5_0.5_SEC0.3DRO1.0G0.0_minpTrue_1600", 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: 0.19.1
• Transformers: 4.53.1
• Pytorch: 2.7.0
• Datasets: 3.6.0
• Tokenizers: 0.21.4

Citations

Cite E2H as:

@inproceedings{parashar2026curriculum,
  title     = {Curriculum Reinforcement Learning from Easy to Hard Tasks Improves {LLM} Reasoning},
  author    = {Parashar, Shubham and Gui, Shurui and Li, Xiner and Ling, Hongyi and Vemuri, Sushil and Olson, Blake and Li, Eric and Zhang, Yu and Caverlee, James and Kalathil, Dileep and Ji, Shuiwang},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year      = {2026},
  url       = {https://openreview.net/forum?id=KJvHnl3kUv}
}