What's the purpose for adding Identity() in chunked blocks?

[wangh09]

I saw the following code for using FSDP:

   # this is to keep the block index consistent if we chunk the block list
   chunked_blocks.append([nn.Identity()] * i + blocks_list[i : i + chunksize])

But I can't figure out why the extra Identity()s are necessary here. Does anyone know? Thanks!

[qasfb]

so, as per the comment, this allows retrieving block indices easily when they are wrapped inside a block chunk so instead of having blocks as : blocks.0.[0..8[ blocks.1.[0..8[

you would have blocks as : blocks.0.[0..8[ and blocks.1.[8..16[ where blocks.1.[0..8[ is just padding