workaround for train_set batching during inference time#784
Open
workaround for train_set batching during inference time#784
Conversation
varisd
force-pushed
the
batching_workaround
branch
from
299c1bc to
0f5649c
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
We can currently use max_len parameter in decoders to avoid OOM exceptions during inference time.
However, this is not enough, e.g. when specifying batch_size in number of tokens.
For simplicity, imagine a decoder-only scenario. For example, let the token-level batch_size be 9, which barely fits into memory, and max_len=3. We can get a batch of [4, 2] (batch_size, seq_len). During inference we can easily generate a result of size [4, 3] which will cause OOM.
This PR suggests one possible solution and is open for discussion.