Should the new anisora fp8 models generally be compatible with existing wan loras?
I plugged in these models to my existing comfyui wan2.2 workflow and am running into various errors when trying to use loras. I tested by bypassing my lora loader node and didn't have any issues.
With all my loras loading I encounter "self and mat2 must have the same dtype, but got Half and Float8_e4m3fn". I tried with a single lora and ran into a different error, ""mul_cuda" not implemented for 'Float8_e4m3fn'".
Any advice here? Am I a dummy for expecting this to work? Claude tells me to just port all my loras to fp8, no big deal 🤣
As an additional data point, I also tried using the wan2.2_i2v_low_noise_14B_fp8_scaled models which work without issue.
-edit- May also be worth noting that I'm using the RES4LYF clownshark samplers.
Ran into the same issue with ksampler advanced