XiaoHongbo-Hope opened a new pull request, #11056: URL: https://github.com/apache/paimon/pull/11056
### Purpose Allow `load_from_lerobot` to accept an ordered list of LeRobot v3 sources and import them directly into one new table group, without an intermediate merged dataset. Renumber global frames and Episodes, unify task/subtask vocabularies, and aggregate statistics while preserving Episode-local timestamps and separate video payloads. Validate metadata before creating tables, commit Frames once, and publish the root training Tag last. The first version requires `parallelism=1` for multiple sources. Splits become one training split; global quantiles and renumbered-field Episode quantiles are omitted rather than approximated. Existing single-source and Ray imports remain unchanged. ### Tests - 127 passed, 1 skipped with LeRobot 0.4.4, PyAV 15.1 and PyTorch 2.7.1 CPU. The skipped test requires TorchCodec. - Compare real local datasets against official LeRobot aggregation; verify FileIO imports, statistics, task mappings, training windows, pixel-correct video reads and failure/tag boundaries. - Add the new tests to the LeRobot 0.4.4/0.6.0/0.6.1 CI matrix. - Python 3.6 syntax, flake8 and `git diff --check` passed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
