AIMET
QualcommAIMET (AI Model Efficiency Toolkit) quantizes and compresses trained models to cut the compute and memory they need on edge devices such as phones and laptops. It works on PyTorch and ONNX models and combines post-training techniques with fine-tuning that recovers accuracy lost to quantization. Qualcomm Innovation Center develops it.
Openness
5 high confidence- license
- BSD-3-Clause(OSI)
- source
- public
- core features withheld
- no
The LICENSE file contains the BSD-3-Clause text. The repository is public and the README describes no paid tier, enterprise edition or license-gated build, so the core is ungated.
- https://raw.githubusercontent.com/qualcomm/aimet/develop/LICENSE recorded 2026-09-30
LICENSE opens with "Copyright (c) 2020, Qualcomm Innovation Center, Inc. All rights reserved." and then the three-clause BSD text, "Redistribution and use in source and binary forms, with or without" and "Neither the name of the copyright holder nor the names of its contributors".
- https://raw.githubusercontent.com/qualcomm/aimet/develop/pyproject.toml recorded 2026-09-30
pyproject.toml lists the authors as "Qualcomm Innovation Center, Inc." and sets build-backend = "scikit_build_core.build" so the package builds from the repository. Searched the body for enterprise, license key, paid and proprietary: none.
- https://raw.githubusercontent.com/qualcomm/aimet/develop/README.md recorded 2026-09-30
README says "is a software toolkit for quantizing trained ML models" and points to "Build from source" instructions, with "aimet-onnx" and "aimet-torch" on PyPI. Searched the body for enterprise, license key, paid, commercial and proprietary: none.
Adoption
2 high confidencePyPI downloads of `aimet-torch` are the adoption signal here.
- https://pypistats.org/api/packages/aimet-torch/recent recorded 2026-08-18
last_month downloads = 23,007 for aimet-torch
Capability
3 medium confidenceAIMET combines post-training and fine-tuning quantization with channel pruning and spatial SVD compression, covering PyTorch and ONNX models for edge deployment, with the README citing the Qualcomm Hexagon DSP as its measured target. That is one optimization family, well short of the retargetable, multi-vendor compilation Apache TVM covers.
- https://raw.githubusercontent.com/qualcomm/aimet/develop/README.md recorded 2026-09-30
README says "AIMET employs post-training and fine-tuning techniques to minimize accuracy loss during quantization and compression." and "AIMET supports models from the ONNX and PyTorch frameworks."
- https://raw.githubusercontent.com/qualcomm/aimet/develop/README.md recorded 2026-09-30
README documents "Spatial SVD*: Tensor decomposition technique to split a large layer into two smaller ones" and "Channel Pruning*: Removes redundant input channels from a layer and reconstructs layer weights" beside post-training quantization for ONNX and PyTorch.
- https://raw.githubusercontent.com/qualcomm/aimet/develop/README.md recorded 2026-09-30
README says models run "5x-15x faster on the Qualcomm Hexagon DSP than on the Qualcomm Kyro CPU" and that quantized models facilitate deployment "on edge devices like mobile phones or laptops".
Verified 2026-09-24