Missing `datasets/OpenThoughts3-1.2M-math.parquet` file in README example

[GrumpySloths]

Hi, thanks for open-sourcing this repository!

I was following the example command in the README for generating teacher responses with Qwen3-4B (Non-thinking):

python scripts/infer/vllm_rollout.py \
  --input-parquet datasets/OpenThoughts3-1.2M-math.parquet \
  --model-path model/Qwen3-4B \
  --gpu-ids 0,1,2,3,4,5,6,7 \
  --enable-thinking false \
  --enable-rejection-sampling true \
  --max-attempts-per-rollout 3

However, I could not find the file:

datasets/OpenThoughts3-1.2M-math.parquet

under the datasets/ directory in the repository.

Could you please clarify how to obtain this dataset file? For example:

1. Is it supposed to be downloaded separately from Hugging Face or another source?
2. Is there a preprocessing script to generate OpenThoughts3-1.2M-math.parquet?
3. Is the file name/path in the README outdated?
4. Could you provide the expected data schema for this parquet file?

This would be very helpful for reproducing the teacher rollout and subsequent student SFT steps.

Thanks!

[Turingzero0]

Thanks for pointing this out!

Now we have provided the dataset used for SFT here:

https://huggingface.co/datasets/llxy/OpenThought3-Qwen3-4B

You can download/load the dataset from Hugging Face and use it to apply SFT. Description in the GitHub README has also been updated.