1 Pull request merged by 1 person

• Fix the evaluation problem

  #29 merged Feb 7, 2025

7 Issues closed by 6 people

• The speed of training in one node

  #20 closed Feb 5, 2025

• training steps in wandb

  #24 closed Feb 4, 2025

• Setups for returns

  #23 closed Feb 2, 2025

• A strange error when saving model

  #19 closed Jan 31, 2025

• training data from MATH

  #7 closed Jan 29, 2025

• File not Found Error in ray job submit

  #14 closed Jan 28, 2025

• I encountered a hang issue when loading the critic model.

  #6 closed Jan 27, 2025

22 Issues opened by 22 people

• Speed of 4 node training

  #30 opened Feb 7, 2025

• Does rStar-Math use RL data?

  #28 opened Feb 6, 2025

• can it support Qwen2.5-Math-1.5B model?

  #27 opened Feb 6, 2025

• Checkpoint Release

  #26 opened Feb 5, 2025

• worker process died

  #25 opened Feb 3, 2025

• Why a critic model is needed?

  #22 opened Feb 1, 2025

• Have you tried to train the model by Lora? It requires less computation resources.

  #21 opened Jan 31, 2025

• Why use PPO instead of GRPO?

  #18 opened Jan 30, 2025

• RuntimeError: Connection closed by peer when training on a single node

  #17 opened Jan 29, 2025

• Where is the long CoT data and preprocess script?

  #16 opened Jan 28, 2025

• Docker Container for Reproducing

  #15 opened Jan 28, 2025

• what is used for critic model

  #13 opened Jan 27, 2025

• rStar-Math-7B row in the first table seems to be inaccurate?

  #12 opened Jan 27, 2025

• PoT style response?

  #11 opened Jan 27, 2025

• Could you point to me how do you preprocess the chat message?

  #10 opened Jan 27, 2025

• Please give a repro on ray cluster setup?

  #9 opened Jan 26, 2025

• Help reproducing training run

  #8 opened Jan 26, 2025

• Thanks for your fast movement and open source spirit

  #5 opened Jan 26, 2025

• Is there a response increase compared to initial length after RL?

  #4 opened Jan 26, 2025

• the minimum hardware resource configuration required for training?

  #3 opened Jan 26, 2025

• What is the reward?

  #2 opened Jan 26, 2025

• A stupid question: Why researchers call such methods as "scaling test-time computation"?

  #1 opened Jan 26, 2025