MegaScience/Qwen2.5-3B-MegaScience

MegaScience: Pushing the Frontiers of Post-Training Datasets for Science Reasoning

This repository contains the Qwen2.5-3B-MegaScience model, one of the models trained as part of the MegaScience project.

For the official code, data processing pipeline, and evaluation system, please refer to the MegaScience GitHub repository.

Qwen2.5-3B-MegaScience

Usage

You can use this model with the Hugging Face transformers library:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "MegaScience/Qwen2.5-3B-MegaScience"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Example text generation
prompt = "The capital of France is"
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
model_inputs = tokenizer([text], return_tensors="pt")

generated_ids = model.generate(model_inputs.input_ids, max_new_tokens=20)
print(tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0])

Training Recipe

• LR: 5e-6
• LR Schedule: Cosine
• Batch Size: 512
• Max Length: 4,096
• Warm Up Ratio: 0.05
• Epochs: 3

Evaluation Results

More about MegaScience

Citation

Check out our paper for more details. If you use our dataset or find our work useful, please cite

@article{fan2025megascience,
  title={MegaScience: Pushing the Frontiers of Post-Training Datasets for Science Reasoning},
  author={Fan, Run-Ze and Wang, Zengzhi and Liu, Pengfei},
  year={2025},
  journal={arXiv preprint arXiv:2507.16812},
  url={https://arxiv.org/abs/2507.16812}
}