v0.66.0 - New AI models added with breaking export command changes
This release brings exciting updates to the AI model hub, including fresh additions like AG-Pose and LFM3. It also introduces important export changes where old python commands are replaced by cleaner CLI tools, alongside better quantization support for models like minilm_v2 to help things run smoother and faster.
New AI Models
This release welcomes several powerful new models to the platform. You can now access AG-Pose for pose estimation, netvlad for place recognition, Parakeet TDT for speech tasks, and LFM3 (including its vision variant) running on GGUF via GenieX llama.cpp.
Breaking Changes & Improvements
There are important updates to how you interact with the system. The old deprecation warnings for python export commands are gone, and running them directly will now throw an error; you must switch to using the new qai-hub-models export command instead. On the bright side, we now have phi_4_mini_instruct onboarding, better w8a16 quantization support replacing w8a8 for minilm_v2, and updated Qwen3 assets.
Under the Hood
For developers and system builders, configuration files like perf.yaml can now be easily parsed using standard protobuf libraries thanks to newly added proto definitions. SDK dependencies have also been refreshed, bumping PyTorch to version 2.14.1 and ONNX max compatibility to 1.22.0.
View selection rationale
Importance 36/100
About this brief
- Content type
- News summary
- Source type
- First-party / official source
- Original source
- Qualcomm AI Hub Releases ↗
- Source published
- Process
- AI-assisted discovery, translation, summarisation, classification and scoring; standards and corrections are maintained by the editorial desk.
- Verification
- The original URL and publication metadata are retained. This brief is not independently fact-checked unless explicitly stated.
- Published by
- Polyemergence Editorial Desk · PolyEmergence Pte Ltd.
- Conflict disclosure
- No financial or commercial conflict declared.