Dylan Prins comments

Repositories
Issues
Comments

Results 4 comments of


                                            Dylan Prins

ARS Returns 0 actions on evaluation env

We have an action space of the following shape: self.action_space = gym.spaces.Box(low=-1, high=1, shape=(198,)) during evaluation the predict returns 0 actions, but during training everything seems fine. Even when using...

SD3 unnecessary train_t5xxl check results in error

Please update the README for SD3 training, it states you need to use the networks.lora instead of networks.lora_sd3 module. That's why the network wasn't initialised properly.

_flash_attention_3 in dispatch_attention_fn is not compatible with the latest flash-atten interface.

Let me now if you need help @ryanpyc27, I have the same issue.

[Bug] WAN 2.2 Quality Degrated Drastically

@helloyongyang @wangshankun