Pack training sequences with the right constraints.
What it does
Guides configuration and validation of sequence packing and long-context training, including batch sizing, padding, and context-parallel constraints.
When to use it
Use when configuring offline LLM packing, collate-time VLM packing, Energon online packing, or long-context training.
How to use it
It changes how the agent configures and validates sequence packing, enforcing compatible batch sizes, sequence lengths, padding, and parallelism settings.
What you provide
- An existing project
What you need · 3
The prescribed verification commands require a Python runtime; no version is specified.
The prescribed verification commands use uv to run the checked-in tests.
The prescribed verification commands require pytest to run the checked-in unit tests.
Access · 0
No additional actions listed.