Flux lora training Error with Batch >1

[Desm0nt]

On the current version in linux, if you set batch to more than 1, the start of Flux Lora training aborts with this error. The old version (commit https://github.com/ostris/ai-toolkit/commit/6d31c6db730a9cba3004eae1a3d7283c663d9295) does not have this problem.

[WarAnakin]

this is not caused by the batch size, it's caused by linear_timesteps being true. Comment it out like in my example

[jaretburkett]

This should be resolved with and without linear_timestep now.