unbreading
Results
3
comments of
unbreading
想问一下题主有遇到数据集解压问题不，下载的压缩文件试了几个解压软件都解压不了
I was able to solve this issue by converting the model to bfloat16. `model= model.to(torch.bfloat16)` But I'm curious about the underlying reason for this behavior. Could an assertion or warning...
试试去掉colocate_actor_ref？我和你其它参数一样，除了没用vllm(环境问题)，micro_rollout_batch_size还比你大。