qiji2023

Results 17 comments of qiji2023

[QA] How to tell if radeon gpu profiler is running in c++ code？

@chesik-amd Can you help me?

[QA] How to tell if radeon gpu profiler is running in c++ code？

ok, thank you very much!

[QA] wavefront occupancy is hard to understand

@chesik-amd can you help me？

[QA] wavefront occupancy is hard to understand

@chesik-amd ok but i don't know how to give you my profile file

[QA] wavefront occupancy is hard to understand

![image](https://github.com/GPUOpen-Tools/radeon_gpu_profiler/assets/133076922/8dcbabdb-0410-48ac-bada-af516c4ddcfd) I noticed a very large time gap here. Can you explain it @chesik-amd

[QA] Can you give a paper about mixed-radix ntt algorithm?

@jeremyfelder I have some questions， could you give me some questions？ 1. Can you explain `fast twiddle`? I can't understand your codes and your codes don't have any comments!!!! 2....

8卡4090*48g, lora微调wan2.1 14B I2V模型OOM，怎么解决？

加了use_gradient_checkpointing_offload和training_strategy deepspeed_stage_3 同样会OOM

8卡4090*48g, lora微调wan2.1 14B I2V模型OOM，怎么解决？

@Artiprocher @wenmengzhou Could you help me？thanks！

8卡4090*48g, lora微调wan2.1 14B I2V模型OOM，怎么解决？

@Artiprocher 你们是否有支持（模型）流水并行微调的计划，把模型分割到多个显卡上来训练，普通玩家很难有80G显存的，拿出24GB都很不错了

8卡4090*48g, lora微调wan2.1 14B I2V模型OOM，怎么解决？

@Artiprocher 有没有qlora的方法，起码让48G的卡跑起来呗