In today’s meeting, we discussed two topics: (1) a proposal for managing bazel, Python, TF version dependencies, and (2) performance benchmarks in a regime with short training rounds (on the order of seconds). For (1), we’ve tentatively agreed to downgrade versions in OSS to enable broader deployment to environments that are less frequently updated, and to come up with a schedule to advance them - a revised proposal upcoming. For (2), we’ve identified overlapping data ingestion/preprocessing and training as a possible low-hanging fruit, e.g., by initiating the former K rounds ahead of the latter. TFF runtime natively supports pipelining, albeit not in a way that’s immediately available via the APIs used in TFF tutorials. New APIs/components upcoming. Meeting notes, as always, are posted on GitHub. Interactive conversations to continue on Discord, and in an additional meeting to be scheduled in a week (outside of the regular twice-monthly cycle). See you there!