[FEA] Handle SplitAndRetryOOM for BatchedByKey
goal in GpuCoalesceBatches
#7778
Labels
reliability
Features to improve reliability or bugs that severly impact the reliability of the plugin
For the
BatchedByKey
goal we are required to preserve that sets of keys stay within a batch (that no key is split between two batches). This fits with the split/merge approach for SplitAndRetryOOM and we should be able to produce smaller batches satisfying this requirement rather than OOMing.The text was updated successfully, but these errors were encountered: