wtedw

seeing the same issue, seems to be related: https://github.com/facebook/create-react-app/issues/10500

For a basic LoRA implementation, we have to determine which nodes represent the QKV matrices. In order to figure out which nodes needs to be targeted, we currently have to...

Gotcha, that makes sense. I'm not well versed in the LoRA implementation, but it looks like you'd need a wrapper layer that keeps the original implementation, but also creates two...

Hey @seanmor5, I was able to implement LoRA with a couple of tricks. I ended up not using `map_nodes`. Instead I created new nodes by extracting them out of Axon.layer....

LCM just adapts these nodes in the unet model: https://github.com/wtedw/lorax/blob/main/lib/lorax/lcm.ex#L121-L139 The weights can be found here: https://huggingface.co/latent-consistency/lcm-lora-sdv1-5 For Bumblebee, (if trying to make it compatible w/ most LoRA files in...

Not sure if this is related to your issue, but I've been dealing with a similar crash whenever the render function is called `{:node_js_worker_exit, {{:badarg, [{:erlang, :port_command, [#Port, "[[\"server\",\"render\"]` Whenever...