Skip to content

model-explorer-v0.1.34

Latest

Choose a tag to compare

@yijie-yang yijie-yang released this 20 Oct 16:33
· 10 commits to main since this release

What's Changed

πŸš€ Highlights & Features

  • Hugging Face Transformers Adapter (model-explorer-hfgraph): Added the Hugging Face Transformers adapter as a built-in adapter, including branch status colors (groupNodeConfigs) and support for keepLayersWithASingleChild (#752, #762, #763, #776).
  • Quantization Support: Added blockwise and multi-axis quantization support in the LiteRT FlatBuffer converter (#748).
  • Graph Picker UX: Users can now expand and collapse subgraphs directly in the graph picker (#739).
  • Diagnostics & Progress Reporting: Introduced backend diagnostic collection, structured logging, JSON serialization, and ABI-safe C-API progress callbacks for graph conversions (#718, #726, #728).

πŸ› Bug Fixes

  • Fixed SIGSEGV crash when printing 0-D non-splat tensors (#740).
  • Secured file handling by keeping uploaded model files confined to their temporary upload directory (#765).
  • Corrected typos across adapter and metadata modules in the server (#764).

πŸ”§ Build, CI & Infrastructure

  • Workflow Improvements: Hardened GitHub Actions workflows against zizmor security findings (#692, #727).
  • Adapter CI/CD: Configured self-hosted Linux ARC job containers, resolved build failures, and improved Bazel caching / C-ABI packaging (#689, #720, #721).
  • Dialect & Dependencies: Synchronized OpDefs via TableGen AST with StableHLO dialect tags (#725), updated TensorFlow pinned dependencies (#770), and removed resolve_symbols_in_exec (#724).
  • Version Bumps: Bumped Model Explorer backend version to 0.2.1 (#769).

πŸ“– Documentation

  • Updated Model Explorer backend documentation to adopt LiteRT naming conventions (#719).

Full Changelog: model-explorer-v0.1.33...model-explorer-v0.1.34