How to Use

Installation

git clone git@github.com:odedsc/CLARO.git
cd CLARO
pyenv virtualenv 3.11.7 CLARO
pyenv activate CLARO
pip install -r requirements.txt
pip install flash-attn==2.7.4.post1

Replicate Results

To replicate results for CLARO , you can use scripts in scripts/replicate.

1. Prepare data:

./scripts/replicate/prepare_data.sh

2. Evaluate models:

./scripts/replicate/eval_model.sh odedsc/CLARO-1.5B
./scripts/replicate/eval_model.sh odedsc/CLARO-7B

Train Models

You can skip this step if you want to use our pre-trained models.

You can run scripts in scripts/train to train your own models. Make sure to specify the correct data path.

Evaluate Models

Use one of scripts/eval to evaluate your models. Make sure to specify the correct model path.

For example, evaluate CLARO on the AIME2025 dataset:

./scripts/eval/eval_model.sh --model path/to/your/model --num-tokens <num_tokens> --datasets aime2025

Prepare Your Own Dataset

You can use scripts in scripts/data to prepare your own dataset.

For CLARO:

python scripts/data/deepscaler_dataset.py --use_both_both

For Evaluation on AIME2025, GPQA, LSAT and MMLU, you can use scripts in scripts/eval:

python scripts/data/generate_aime.py
python scripts/data/generate_gpqa.py
python scripts/data/generate_lsat.py
python scripts/data/generate_mmlu.py

Models

We release the CLARO-optimized models via Hugging Face.

Model	Size	Link
CLARO-1.5B	1.5B	🤗 Hugging Face
CLARO-7B	7B	🤗 Hugging Face

Acknowledgments

We would like to thank rLLM and L3 Lab for codebase, and opensourcing their models. This codebase is built on top of their work.

Citation

If you use CLARO in your research, please cite:

@inproceedings{schlesinger2026claro,
  title={CLARO: Controlled Attribute-Driven Reasoning Optimization for Efficient Chain-of-Thought},
  author={Schlesinger, Oded and Kim, Young Kyung and Di Martino, J Matias and Sapiro, Guillermo},
  booktitle={Findings of the Association for Computational Linguistics: ACL 2026},
  pages={26779--26799},
  year={2026}
}