Training Dataset

[Lucas-TY]

Hi,
Great work on the paper.

I'm currently trying to reproduce the experiments, but I haven't been able to find the dataset used in the paper. I found a version on ZLab's Hugging Face page(z-lab/qwen3-4b-instruct-100k), but it does not seem to be enough to fully reproduce the experiments described in the paper.

Could you let me know where I can access the complete dataset?

[jianc99]

https://huggingface.co/datasets/nvidia/Nemotron-Post-Training-Dataset-v2
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1

We did't use the full dataset of Nemotron-Post-Training-Dataset-v2.

[exhyy]

@jianc99 Hi, thanks for your excellent work!
Would it be possible to share the processed 800K-sample training data? Alternatively, could you provide more details about how the training data was constructed, such as the exact sampling ratios and whether the generated responses include CoT? I would greatly appreciate any clarification. Thank you!