search

lavinal712 / control-lora-v3

MIT License
4 stars 0 forks source link

run with distilled model #1

Open kmpartner opened 3 months ago

kmpartner commented 3 months ago

Is it possible to run distilled model (Block-removed Knowledge-distilled Stable Diffusion, https://github.com/Nota-NetsPresso/BK-SDM).

When I tried to some distilled model. I got error message similar to "unknown mid_block_type" , probably because of unet structure changed.

lavinal712 commented 3 months ago

You should train your own model to fit the distilled model.

kmpartner commented 2 months ago

Thank you for response.

kmpartner commented 2 months ago

What model and dataset are used for test images generation?

lavinal712 commented 2 months ago

What model and dataset are used for test images generation?

SD1.5 test image is in https://github.com/lllyasviel/ControlNet/discussions/188

kmpartner commented 2 months ago

I tried to train ControlLoRA by using code in page, then tried to resume from created checkpoint. But resume from the checkpoint does not work in ControlLoRA training. Error message say, config.json not found.

How can I solve this problem?

lavinal712 commented 2 months ago

I tried to train ControlLoRA by using code in page, then tried to resume from created checkpoint. But resume from the checkpoint does not work in ControlLoRA training. Error message say, config.json not found.

How can I solve this problem?

Well, the code doesn't support from resume from created checkpoint for now as it only stores lora layers. Please retrain a new model.