Purpose

• Standardize oneshot callbacks and prepare for oneshot pipeline extraction

Changes

• Implement the sequential_epoch_end callback event which should be called after one sequential layer has been calibrated with one epoch
  • Add callback event to sequential and layer_sequential pipelines
• Trigger module compression both on the SEQUENTIAL_EPOCH_END event as well as on finalize
• Prepare smoothquant for pipeline extraction
  • Replace uses of _hf_hook. pre_forward with align_module_device for clarity
  • Specify resolved_mappings_ type hint for clarity
  • Trigger _apply_smoothing on the SEQUENTIAL_EPOCH_END as well as on finalize
  • Add a guard which allows the _apply_smoothing function to be called multiple times per session (as is required by sequential pipeline)

Testing

• Cannot test since cannot call events before initialization