Dataset · open-r1
DAPO math problems for GRPO
open-r1/DAPO-Math-17k-Processed
Cleaned DAPO-Math prompts and answers compatible with the TRL GRPO trainer.
What's inside
A processed version of BytedTsinghua-SIA/DAPO-Math-17k. Prompts are deduplicated, and formulations and reference answers are formatted to be compatible with the TRL GRPO trainer. Separate English and Chinese subsets are also highlighted; metadata size is 10k–100k examples.
What you can do with it
- Train a math reasoning model using GRPO
- Test the TRL GRPO pipeline on ready-made prompts
- Evaluate the model on English and Chinese subsets
- Size
- 10K–100K rows
- Downloads this month
- 20.3K
- Likes
- 90
Models trained on it
- mradermacher/Llama3.1-8B-Thinking-R1-i1-GGUF 1.1K
- mradermacher/Llama3.1-8B-Thinking-R1-GGUF 909
- FrancescoArno94/Qwen3-1.7B-GRPO-16bit-DAPO-mathText generation 224
- CaptainHPY/Qwen2.5-7B-R1-GGUFText generation 103
- Palind/Qwen3-4B-PG-repro-bestText generation 66
- hdong0/Qwen3-8B-base-Open-R1-GRPO_dapo_acc_2048_to_16384_noklText generation 52
- Stormtrooperaim/Small-Math-Q8_0-GGUF 45
- Palind/Qwen3-4B-SC-TIS-repro-bestText generation 43
How to load it
from datasets import load_dataset
ds = load_dataset("open-r1/DAPO-Math-17k-Processed")