ifsheldon

> ```rust > let weights = weights.try_add(mask.try_broadcast_like(&(H, s1, s2))?)?; > ``` I don't know how dfdx handles adding infinity, but in theory this is not sufficient since addition doesn't block...

> It shouldn't block gradient flow, but the gradients will be subtracted by inf so in practice they should go to zero. > Regarding infinity, I know huggingface usually uses...

> I think actually masking and subtracting inf should result in the same amount of computation? Note that the below is purely theoretical on paper in terms of tracking compute...

I have a similar question/confusion. If I understand Fig.3 correctly, when we do foreground and background generation, two (SDXL + LoRA) branches should run in parallel with attention sharing, thus...

I think it's fine. It's an MVP after all. Otherwise, I guess we need to do something like you did in render_input_markdown, which is to use `Area`. But I haven't...

> There's a problem: the save/restore thing breaks as soon as there's a scrollbar. I fixed this in commit `fb5b0eb` by calculating where the anchored cursor should be instead of...

I encountered this when trying to develop based on dify 0.6.8. The comment [here](https://github.com/grpc/grpc/issues/33714#issuecomment-1707275497) worked for me.

@Yeuoly FYI, here is the code for the plugin. https://github.com/ifsheldon/gpt_plugin_template_rs You can reproduce the error like below 

Thank you! I actually read some source code last night and got basically what is going on. But I encountered an issue and I don't know why. My custom state...

FYI, we've published the full weights [here](https://github.com/GempollAI/LooseControlNet). You can use it like normal ControlNet.