Re-release consolidated OPT / OPT-IML checkpoints

[suchenzang]

After https://github.com/facebookresearch/metaseq/pull/459 and https://github.com/facebookresearch/metaseq/pull/556, we can now release updated checkpoints that are consolidated from FSDP shards with different model parallelism as well. We should update all of our checkpoints as a start to help address some of the following painpoints that users are facing:

• https://github.com/facebookresearch/metaseq/issues/595
• https://github.com/facebookresearch/metaseq/issues/594
• https://github.com/facebookresearch/metaseq/issues/574
• https://github.com/facebookresearch/metaseq/issues/567
• https://github.com/facebookresearch/metaseq/issues/475
• https://github.com/facebookresearch/metaseq/issues/407
• https://github.com/facebookresearch/metaseq/issues/233
• https://github.com/facebookresearch/metaseq/issues/211

and previous issues:

• https://github.com/facebookresearch/metaseq/issues/31
• https://github.com/facebookresearch/metaseq/issues/125

We have internal consolidated versions for 2.7B and 30B to check against, and will also need to confirm that generation looks roughly sane after consolidation.

[EIFY]

I can work around it, but could the following issue be considered related as well?

• #552

[andrewPoulton]

Yeah looks like it, at least tangentially - the loading logic there could probably do with simplifying. It should be possible to identify naming convention by just reading the checkpoint directory.