Roya
Roya
同 换源了用这个也不行docker pull qwenllm/qwen:1.5-cu121
 换了以后会报错tensor类型不匹配这是为啥呀
 好像原来的代码里已经加了,我在SupervisedDataset、LazySupervisedDataset里尝试把input_ids、attention_mask和labels都显示to('cuda')了一下,还是会报这个错误
不是的,我后来跑的是instruct的版本,用的是zero3,没有做修改
 换了以后到第五轮就会报这个错误
python版本什么比较合适呀
好的,谢谢,还有想问一下为什么我指定4张3090跑不了,但是指定2张40g的a100可以跑呀,我看我增加显卡每个显卡的占用并不会减少,是因为代码里只做了数据并行吗
我直接用 pip -r requirements.txt 安装依赖环境报错 ERROR: No matching distribution found for bitsandbytes==0.43.0要怎么解决呀
Thanks a lot for your reply. I have on more question have you ever try other version of whisper to get audio feature, i tried whipser large-v3, but with yield...
This error occurs when I running the spider example, i followed the instruction to start ray cluster by running the script ../../scripts/restart_ray.sh, then when i start running sql_agent.py, the terminal...