Repository Issue Activity (beta)

#125 Did anyone try to launch this repo with a more recent version of veRL?

#124 Qwen2.5 1.5B with GRPO meets Entropy Collapse

#105 qwen2.5-3b OOM Issue on 8x V100 GPUs

#76 Is there a method to calculate ACCURACY？

#122 How to reproduce a qwen-3b model that supports multiplication?

#77 模型只会加减，不会乘除。。。

#92 how to run main_generation and main_eval?

#108 How to guarantee the problem is valid?

#100 How to run SFT on 2 L40 GPUs

#120 定义答案正确的奖励是不是应该从<answer>的标签里找

#117 如果我想做QA-pair的RL训练，不是rule-based reward，该看哪里，多谢多谢多谢！

#118 有没有跑QA-pair数据集的example，不是rule-based reward的math数据集，多谢多谢多谢

#19 how to prepare the dataset? Is there any reference?

#110 What Dataset should I use?

#57 follow the official code,got the error:because name 'global_poolverl_group_2:0' already exists

#7 Failed to register worker to Raylet

#115 RuntimeError: FlashAttention only support Ampere GPUs or newer despite setting VLLM_ATTENTION_BACKEND=XFORMERS

#5 1 gpu is not working , 2 gpus out of memory

#97 <|endoftext|>

#114 Support model

#88 ValueError: Model architectures ['Qwen2ForCausalLM'] are not supported for now.

#113 Why success in PPO but gpu memory overflow in GRPO?

#111 报错vllm的版本不支持 (vllm version 0.8.5.post1 not supported)

#30 Qwen 3B OOMs on 2 H100s

#85 After I trained for 500 steps, the length of think became smaller and smaller, and even disappeared.