Wire up CLI for scan and host offloading #156
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In the previous PR (#157) we added the ability to replace a sequence of layers with scan and in an earlier PR (#151) we added the ability to offload tensors by transforming a graph. This PR changes the CLI to be able to select either for-loop, scan, or scan+host offloading.
In order to mix-and-match
sharding
configs with scan/offloading configs without creating a combinatorial explosion, I had to add a new config group calledremat
and add scan and offloading there. Sinceactivation_checkpointing_layers
and the optimization barrier is also part of remat, I moved it there too. This means the only left config in thescaling
group issharding
, so I've hoistedsharding
to be top level. This moves a bunch of files but it's mostlys/scaling/sharding
.