v0.1.1

v0.1.1 Release notes

Overview

This first public release of mlia-converters-pytorch covers development
since the repository’s initial commit.

The package integrates PyTorch model export and conversion into MLIA,
with four automatically discovered transformers.

Supported conversions

• nn_module_to_pt2: exports a torch.nn.Module to .pt2 using
  torch.export, with example inputs supplied as a tuple.
• pt2_to_tosa: converts .pt2 models to TOSA-1.0+INT, with
  post-training quantization enabled by default.
• pt2_to_pte: produces ExecuTorch .pte programs with quantization
  and Arm Ethos-U delegation.
• pte_to_delegate: extracts TOSA or VGF payloads from ExecuTorch programs.

Conversion features and diagnostics

• Optional direct TOSA lowering with enable_quantization=False.
• PyTorch node names in TOSA operator locations for traceability.
• Validation of inputs, target configuration, and delegate payloads.
• Checks that PTE conversion produces Ethos-U delegation.
• Preserved propagation into MLIA logging.
• Python and native compiler output captured at debug level.
• Robust output capture that preserves logging destinations and handles
  unavailable streams and concurrent conversions.

Requirements

The package declares Python 3.10 or later and integrates with MLIA
and downstream backend plugins.

v0.1.1rc3

What's Changed

• ci: require Black Duck before syncing main and release commits by @max-shvetsov
• ci: Pin public sync to the merged workflow commit by @max-shvetsov

Full Changelog: v0.1.1rc2...v0.1.1rc3

v0.1.1rc2

What's Changed

• ci: prepare uv dependencies for the release Black Duck scan by @max-shvetsov

Full Changelog: v0.1.1rc1...v0.1.1rc2