Skip to content

3.3.0

Latest
Latest

Choose a tag to compare

View all tags
@gblong1 gblong1 released this 23 Jun 15:25
· 10 commits to main since this release
3.3.0
4d9afa6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

What's New

This release upgrades the runtime to OpenVINO 2026.2+, consolidates Python dependency management into setup.py, and extends the ModelManager API for callers that pre-detect NPU capabilities.

Highlights

  • OpenVINO 2026.2+ / OpenVINO-GenAI 2026.2+ — the plugin now requires and pins to the latest OV runtime.
  • Single source of truth for dependencies — requirements.txt has been removed; all runtime dependencies are declared in setup.py install_requires.
  • ModelManager API extension — ModelManager(weight_path) accepts new optional kwargs npu_arch= and npu_is_available=, letting host applications pass in NPU state they already detected instead of forcing redundant ov.Core() queries.

What's new

Runtime

  • Bumped openvino to >=2026.2 and openvino-genai to >=2026.2.
  • transformers pinned to the >=4.37.0,<=4.56.2 range (avoids breakage in 4.57+).
  • diffusers capped at <=0.37.1.
  • Suppressed the noisy controlnet_aux mediapipe UserWarning at package import — mediapipe is an optional dependency we don't use, and the warning was confusing users into thinking something was wrong.

Packaging

  • All dependencies (optimum, optimum-intel, peft, pydantic, tomesd, hf_xet, protobuf, sentencepiece, etc.) now live in setup.py.
  • install.sh / install.bat simplified — they no longer reference requirements.txt; pip install . pulls everything.

API

  • ModelManager.__init__(weight_path, npu_arch=None, npu_is_available=None) — both new kwargs are optional and default to existing auto-detection behavior, so existing callers are unaffected.

Documentation

  • README clarifies that the inpainting model dropdown now lists both SD 1.5 and SDXL inpainting models.
  • Minor copy fixes in the inpainting workflow steps.

Breaking changes

  • Python ≥ 3.10 is now required. setup.py previously declared >=3.7, but the codebase had already moved past that. Users on 3.8 or 3.9 will fail at install.
  • requirements.txt is gone. External tooling, CI pipelines, or Docker images that did pip install -r requirements.txt must switch to pip install . (which is what the bundled installers already do).
  • OpenVINO 2025.x is no longer supported. Environments pinned to 2025.4 or earlier will need to bump to 2026.2+.

Upgrade notes

For most users, the bundled installer handles everything:

# Linux
./install.sh

# Windows
install.bat

If you maintain your own environment:

pip install --upgrade .

The first launch after upgrade may take longer as OpenVINO recompiles model caches against the new 2026.2 runtime.

Full changelog

[3.2.1...3.3.0](https://github.com/intel/openvino-ai-plugins-gimp/compare/3.2.1...3.3.0)

Assets 2
Loading