Skip to content

v0.5.0

Latest
Compare
Choose a tag to compare
@jaybdub jaybdub released this 03 May 20:13
· 23 commits to master since this release
19a3176
  • Added tensor shape tracking to support dynamic shapes for flatten, squeeze, unsqueeze, view, reshape, interpolate, and getitem methods
  • Added EasyOCR example
  • Added the DatasetRecorder context manager, allowing to easily capture of module inputs in large pipeline for calibration and shape inference
  • Added support for legacy max_batch_size using optimization profiles
  • Added support for nested tuple, dict and list module inputs and outputs via. the Flattener class
  • Added ability to accept dataset as inputs argument, and infer optimization profiles from the data
  • Added Dataset, TensorBatchDataset, ListDataset, and FolderDatset classes
  • Added support for dynamic shapes
    • Known limitation: Currently some converters (ie: View) may have unexpected behavior if their arguments are defined with dynamic Tensor shapes.