Add support for LCM lora

[db0]

https://huggingface.co/blog/lcm_lora

[tazlin]

As mentioned else where, this would benefit from n_iter (n on the API), especially with high end cards.

Namely, a 4090 could likely do a 4 step image in sub-second time frames (<0.7 seconds according to the paper), but initializing a comfy pipeline takes longer than that. Workers would benefit in terms of efficiency being able to batch these jobs and users would get more images for less kudos.

[db0]

Correct. Batching is very much on my priority list. I'm hoping to get the NLNet funding to work on it soon.

[tazlin]

As of Haidra-Org/hordelib@261b259, the version of comfyui packaged supports LCM, and now hordelib needs the relevant changes to accept it, accounting for the fact it needs a certain sampler to work. (see here for more info).

[tazlin]

This was resolved as of Haidra-Org/hordelib#132 and #337