v0.7.0

Minor updates

• Starting from version 2.32, "QNN SDK" is now referred to as "QAIRT SDK" (with updated download link across several App READMEs).
• Tweaks to LLM tutorial for clarity.
• Add Llama3-TAIDE-LX-8B-Chat-Alpha1 to LLM tutorial.
• Change deprecated soc_id to soc_model in some HTP configurations.

Bug fixes

• Fix bug in deployable ONNX script (tutorials/deploy/onnx/build_deployable_asset.py) affecting numerical correctness.
• Fix demo video (did not work for many users)

v0.6.0

Release Notes:

• Added support for IBM Granite 3.1 in LLM Tutorial
• Fixed bug in Whisper demo on Snapdragon X Elite
• Fix poor Super Resolution outputs around black/white regions

v0.5.0

What’s new

• Added QCS8550 chipset in Android ChatApp
• Added list of Production devices and OS to run Android ChatApp

Fixes and updates

• Fixed links and documentation for llm-on-genie tutorial
• Update TFLite-QNN Maven Repository name to LiteRT-QNN

v0.4.0

New Additions

• Sample Android Chat App based on Genie C++ API

v0.3.2

New Additions

• Updated LLM on-device deployment supported device table, as well other edits
• Fixed Llama 3 prompt in ChatApp
• Fixed typo in whisper export command

v0.3.1

New Additions

• Added reference to the newly released WhisperKit for Android, by our collaborator Argmax
• Guide for running LLMs (Llama3) on-device using the Qualcomm AI Engine Direct SDK.

v0.3.0

New Additions

• Added Windows Sample Applications: LLamaV2 Chat, Whisper Speech-to-Text (Python), Classification, Super Resolution

Quality Improvements & Bug Fixes

• Android Applications now use QNN directly from Maven. (Previous code would download a copy of the SDK and manually extract it).

v0.2.1

New Additions:

• Example Windows app using ONNX runtime and NPU acceleration via the QNN execution provider:

  • ObjectDetection: Detect objects in an image and annotate them with bounding boxes

v0.1.1

New Additions:

• Example Android apps using TFLite and NPU acceleration via the QNN delegate:

  • ImageClassification: Image classification on still images
  • SemanticSegmentation: Semantic segmentation on live camera feed
  • SuperResolution: Super resolution on still images