Sukai Huang

Results 10 comments of Sukai Huang

[Feature Request] RAINBOW

Any updates about the rainbow implementation?

ABGR8888 causes DisplayLink reconnection loop on Wayland sessions

Sorry to trouble but I do not understand the effect of` KWIN_DRM_NO_AMS=1` Will it affect any other things in the computer?

ABGR8888 causes DisplayLink reconnection loop on Wayland sessions

Greeting, does Plasma 6.0.3 solve the problem?

nvidia mode not working, nvidia modules already in use

@Askannz Hi, I faced the same issue and then I solve it by adding 'sudo' command. I am not sure what happens here.

nvidia mode not working, nvidia modules already in use

what is the use of rmmod btw？ rmmod是删除module还是 重置module呀

import of module '__prefect_loader__' failed when 'multiprocessing.Pool.apply_async' is used within the flow via remote deployments

> Hey [@madkinsz](https://github.com/madkinsz) - the main reason it is failing is because - `__prefect_loader__` module isn't available in sys.modules to find the function `bar`. > > Since the entry `prefect.engine`...

Lora PPO example scripts

目前我用LoRa + ray + zero_stage 2 + remote_rm_url 跑会卡住 ```bash export CUDA_HOME=$CONDA_PREFIX export PYTHONPATH=$PWD export WORKING_DIR=$PWD export NCCL_SOCKET_IFNAME=bond0.3027 # ! very important for deepspeed multi node export NCCL_IB_DISABLE=0 # ?...

Lora PPO example scripts

正在检测是不是因为adam_offload 导致训练速度变慢，所以看上去卡住了

Lora PPO example scripts

更新测试 vLLM + full PPO + ray + zero_stage 2 + remote_rm_url 不会卡住 no vLLM + full PPO + ray + zero_stage 2 + remote_rm_url 不会卡住，但是要disable adam_offload no vLLM +...

Lora PPO example scripts

更新结果 no vLLM + LoRA PPO + ray + zero_stage 2 + remote_rm_url 不会卡住，但是会比 full PPO 要卡在前面 很久。很诡异