[Bug]: loss is logged after 1 step, and logging (all) is off by 1 step

[R-N]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What happened?

1. Logging is off by 1 step, for the loss CSV, model, and previews

I put a counter which increments after optimizer.step(). I set the training to log every 10 steps, but it does so after 11.

This happened because step is defined by i (+ ititial_steps). i indicates the current step and is 0-indexed. It doesn't indicate the count of steps done. But the logging is done with step % save_hypernetwork_every == 0. So when it logs at step=10, it has actually done 11 steps (0-10).

2. Loss logging is done after 1 step.

Loss log in CSV always starts with steps being 1. This happened because it uses step % training_write_csv_every, while step is 0-indexed as previously mentioned. Then it logs with step+1. So every next log is off by 1 step. It didn't check if step > 0 either.

Steps to reproduce the problem

1. Set a value (10 here) for csv log in settings
2. Set a value (10 here) for model and preview logs
3. Train a hypernetwork or textual embedding

What should have happened?

Logging should be done every n steps as set by user.

Commit where the problem happens

737eb28

What platforms do you use to access UI ?

Other/Cloud

What browsers do you use to access the UI ?

Google Chrome

Command Line Arguments

--autolaunch --ckpt {workspace3}/novelai/final_pruned.ckpt --deepdanbooru --disable-safe-unpickle  --no-half-vae

Additional information, context and logs

I'll try