time_embed elements location

[miasik]

The elements below are listed in your table as members of IN00 block.
It was a coincident but I've noticed the the next elements are outside IN00.
They have "Not Merge" in the "Block" column in the "Elements" tab.

model.diffusion_model.time_embed.0.bias
model.diffusion_model.time_embed.0.weight
model.diffusion_model.time_embed.2.bias
model.diffusion_model.time_embed.2.weight

Where is the truth?

[miasik]

Actually, there's no "skip_connection" in IN01

[hako-mikan]

Yes. Exactly outside of IN00, but when the merge is processed, it is processed as IN00. This is also true for OUT11. This is also true for MBW, from which this script is derived.

                if 'time_embed' in key:
                    weight_index = 0                # before input blocks
                elif '.out.' in key:
                    weight_index = NUM_TOTAL_BLOCKS - 1     # after output blocks

This script is quoted from Merge Block Weighted and time_embed is considered IN00.

And thanks to reporting mistake of table.
I fixed table.