- [2025年05月23日] 🎉 We release FastCuRL-1.5B-V3 and FastCuRL-1.5B-V2. Specifically, FastCuRL-1.5B-V2 and FastCuRL-1.5B-V3 are models with continued training based on the Preview version.
- [2025年03月17日] We release FastCuRL-1.5B-Preview, a slow-thinking reasoning model that outperforms 📈 the previous SOTA DeepScaleR-1.5B-Preview with 🚀 50% training steps! We propose a curriculum RL framework with stage-wise context scaling to achieve efficient training and concise CoT reasoning based on DeepSeek-R1-Distil-Qwen-1.5B and observe continuous performance improvement as training steps increase. To better reproduce our work and advance research progress, we open-source our code, model, and data.
We report Pass@1 accuracy averaged over 16 samples for each problem.
| Model | AIME 2024 | MATH 500 | AMC 2023 | Minerva Math | OlympiadBench | Avg. |
|---|---|---|---|---|---|---|
| Qwen2.5-Math-7B-Instruct | 13.3 | 79.8 | 50.6 | 34.6 | 40.7 | 43.8 |
| rStar-Math-7B | 26.7 | 78.4 | 47.5 | - | 47.1 | - |
| Eurus-2-7B-PRIME | 26.7 | 79.2 | 57.8 | 38.6 | 42.1 | 48.9 |
| Qwen2.5-7B-SimpleRL | 26.7 | 82.4 | 62.5 | 39.7 | 43.3 | 50.9 |
| DeepSeek-R1-Distill-Qwen-1.5B | 28.8 | 82.8 | 62.9 | 26.5 | 43.3 | 48.9 |
| Still-1.5B | 32.5 | 84.4 | 66.7 | 29.0 | 45.4 | 51.6 |
| DeepScaleR-1.5B-Preview | 43.1 | 87.8 | 73.6 | 30.2 | 50.0 | 57.0 |
| FastCuRL-1.5B-Preview | 43.1 | 88.0 | 74.2 | 31.6 | 50.4 | 57.5 |
| FastCuRL-1.5B-V2 | 47.5 | 89.3 | 77.0 | 32.8 | 53.3 | 60.0 |
| FastCuRL-1.5B-V3 | 49.6 | 90.5 | 78.5 | 34.7 | 54.5 | 61.6 |
| Model | Training Steps | Training Stages | Number of GPUs Used in Each Stage |
|---|---|---|---|
| DeepScaleR-1.5B-Preview | ~1,750 | 3 | 8, 16, 32 |
| FastCuRL-1.5B-Preview | ~860 | 4 | 8, 8, 8, 8 |
| FastCuRL-1.5B-V2 | ~1,710 | 5 | 8, 8, 8, 8, 8 |
| FastCuRL-1.5B-V3 | ~2,620 | 5 | 8, 8, 8, 8, 8 |
Here, we uniformly set the batch size to 128 for counting training steps, meaning two steps with a batch size of 64 are counted as one with a batch size of 128.
# Installing Python 3.10 Environment. conda create -n rllm python=3.10 -y conda activate rllm # Installing RLLM dependencies. cd rllm pip install -e ./verl pip install -e .
Following DeepScaleR, our training dataset consists of 40,315 unique problem-answer pairs compiled from:
- AIME problems (1984-2023)
- AMC problems (before 2023)
- Omni-MATH dataset
- Still dataset
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 export VLLM_ATTENTION_BACKEND=XFORMERS # Run 8K context length training, 160 steps bash ./scripts/train/run_fastcurl_1.5b_8k_stage1.sh | tee -a fastcurl-1.5b-stage1.log # Run 16K context length training, 590 steps bash ./scripts/train/run_fastcurl_1.5b_16k_stage2.sh | tee -a fastcurl-1.5b-stage2.log # Run 24K context length training, 230 steps bash ./scripts/train/run_fastcurl_1.5b_24k_stage3.sh | tee -a fastcurl-1.5b-stage3.log # Run 16K context length training, 580 steps bash ./scripts/train/run_fastcurl_1.5b_16k_stage4.sh | tee -a fastcurl-1.5b-stage4.log
python3 -m verl.trainer.main_generation \ trainer.nnodes=1 \ trainer.n_gpus_per_node=8 \ data.path=./fastcurl/data/test/xxx.parquet \ data.output_path=${OUTPUT_DIR}/xxx.parquet \ data.n_samples=16 \ data.batch_size=2048 I am running a few minutes late; my previous meeting is running over. model.path=${MODEL_PATH} \ rollout.temperature=0.6 \ rollout.response_length=32768 \ rollout.top_k=-1 \ rollout.top_p=1 \ rollout.gpu_memory_utilization=0.9 \ rollout.tensor_model_parallel_size=1
If you find our work helpful, feel free to give us a cite.
@misc{fastcurl, title={FastCuRL: Curriculum Reinforcement Learning with Stage-wise Context Scaling for Efficient Training R1-like Reasoning Models}, author={Mingyang Song and Mao Zheng and Zheng Li and Wenjie Yang and Xuan Luo and Yue Pan and Feng Zhang}, year={2025}, eprint={2503.17287}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2503.17287}, }
- Our model is trained on top of
DeepSeek-R1-Distill-Qwen-1.5B. - Our training experiments are powered by our heavily modified fork of verl.
- We directly use DeepScaleR's code to finish our experiments. However, we have modified parts of the code related to naming conflicts to avoid confusion.