Version 2.26.0

Full Changelog: 2.15.0...2.26.0

Version 2.25.1

2.25.1

• Bug fixes and Improvements

  • ONNX

    • Fix for encoding propagation for concat layers (5084af3)

  • Torch

    • Fix to reduce GPU RAM usage for AdaScale for Qwen 3 VL model (ee3d193)

Version 2.25.0

2.25.0

• Bug fixes and Improvements

  • ONNX

    • Reduced peak CPU memory usage for AdaScale and SeqMSE techniques (28f89a7)
    • Reduced peak CUDA memory usage for AdaScale technique (a29f44f)
    • Added support for Qwen3 VL models in GenAITests (c014961)
    • ONNX-IR based supergroup pattern detection and replacement (9972c1b)
    • Tie concat and interpolation ops by default (a8ac6f4)

  • Torch

    • Bug fix for onnx qdq export with control flow ops (ae1abd1)
    • Use Triton kernels by default if available (3adcbee)
    • Introduces block_size parameter to EncodingAnalyzer (e250abd)
    • Always export encodings as uint (ae7d5ef)
    • float4/8 QDQ export support (135a0af)
    • Support loading zero_point_shift with sim.load_encodings() (624ba30)
    • Support built-in quantization of SyncBatchNorm (1e8eceb)

Version 2.24.0

• Bug fixes and Improvements

  • ONNX

    • Add Windows ARM64 wheel build/test support, distribute Windows ARM64 wheel on GitHub releases (1390b96)
    • Add transpose MatMul support in Sequential MSE (ff7a284)

  • Torch

    • Expose block-level AdaScale API (72246db)
    • Improve numerical stability of zero point shifting ([-1.5, -.5, .5, 1.5]) implementation (489f7df)
    • Fix :func:replace_lora_layers_with_quantizable_layers to inherit train/eval flag (af5a82d)
    • Fix SpinQuant evaluation by untying lm_head and embed_tokens prior to loading the state_dict (47f574d)
    • Experimental - Implement Progressive Gradient Scaling (PGS) support for Triton-based quantization kernels (b58b00b)

  • Common

    • Fix TFEnhanced incorrectly producing negative scales when encountering empty (size‑0) inputs (ea4af6a)
    • Unpin numpy dependency (8a999a1)
    • Add an alias for referencing the eNPU configuration file (b79611c)

Version 2.23.0

• Bug fixes and Improvements

  • ONNX

    • Disable per-channel quantization for ConvTranspose ops (9395e32)
    • New top level API for configuring parameter quantization type (a1c197d)

  • Torch

    • Enable Torch Dynamo ONNX export (59e0125)

  • Common

    • Enable per-channel matmul quantization in config files (7137849)
    • LLM quantization recipes in docs (6561f0e)
    • Fix CUDA discrepancies against CPU wheel (01e7422)

Version 2.22.0

• Bug fixes and Improvements

  • ONNX

    • Allow loading 2.0.0 encoding format to sim (e8cb098)
    • Fix Cast unpacking error (6761a19)
    • Enable exporting non-LPBQ encodings with zero_point shift (7b3cc4c)
    • Implement aimet-onnx LPBQEncoding (5ad7ea6)

  • Common

    • Support exporting 1x1 Conv LPBQ to ONNX QDQ (58ce71d)

Version 2.21.0

• Bug fixes and Improvements

  • ONNX

    • Fix IndexError when Conv or Linear layers are reused in the model (65c4b3b)
    • Add optional argument export_int32_bias to aimet-onnx export (3b8e0f0)
    • Unpin PyTorch version in aimet-onnx (d99b6c4)
    • Align NaN handling with ORT CPU Execution Provider (e4c49eb)
    • Fix quantization axis handling for transposed MatMul operations (6ca06d6)

  • PyTorch

    • Fix quantization logic to enable input quantizers for layers following ignored layers (80fb4fe_)

Version 2.20.0

2.20.0

• Bug fixes and Improvements

  • Common

    • Update supported python version to >=3.10 (2bc8c94)
    • Repackage aimet_common as alias to aimet_onnx.common or aimet_torch.common (074e85f)
    • Remove Pad op from data movement ops (21cddb6)

  • ONNX

    • Export data movement op output encoding in sim.export by default (550c029)
    • Assign generic node names if node name is missing or duplicate (273dd82)
    • Add PyTorch Pad modules to nn.Module -> onnx op mapping (7e5342b)
    • Add LSTM cell state int32 quantization mechanism for LPAI (3a8659b)
    • Support stacked RNN/GRU/LSTM (552ad83)
    • Make exclude/include node argument naming consistent (ec22d86)
    • Implement LPBQ support in aimet-onnx SeqMSE (495567f)
    • Add support for dilation, grouping, stride to Quantized Conv (f94f3e2)
    • Remove block type from adascale config (b55b058)
    • Skip tying concat encoding if input has multiple consumers (3136828)
    • Tie quantizers upstream first and downstream later (59aac3e)
    • Fix ValidationError in LazyExtractor when external files are missing or inconsistent (a8f32fc)
    • Align torch and onnx GenAI recipes (7d4659d)

  • Torch

    • Use separate input quantizer for each concat input (755c54a)
    • Add predict and fallback later approach for batched matmul in aimet-torch seq mse (8874173)
    • Refactored MMP to not use rounding mode (fd7e40d)
    • Use tuple for strided slice indexing (4ddbd66)
    • Fix symmetry bug in _from_qnn_encoding_dict (35602ea)
    • Align onnx 1.0.0 BQ encoding export ordering with QAIRT expectation (0182b7a)

Version 2.19.0

• New Features

• Bug fixes and Improvements

  • ONNX

    • Make LiteMP API percentage float (69f96ff)
    • Set layernorm int16 weight to symmetric by default (8560e13)
    • Automatically insert data movement op output qdq during to_onnx_qdq (15c8b9b)
    • Create LazyExtractor to handle external data for onnx Extractor utils (104e7e8)
    • Tie input/output encodings across maximum Concat subgraph (832ea91)
    • Tie hidden state quantizers of RNN/GRU/LSTM (c18fd05)

  • Torch

    • Fix histogram observer rebinning logic (2c88364)
    • Fix connectedgraph input ordering for non-trivial layer types (2b7b548)

  • Common

    • Disable per-channel quantization of RNN/GRU/LSTM for all HTP backends (df8b875)

Version 2.18.0

• New Features

  • Torch
    • Promoted aimettorch.onnx.export and QuantizationSimModel.onnx.export as production APIs (99160d2, e026fd1)
    • Added utility functions to exclude some or all unknown nn.Modules from quantization (5a419f3, 501eebd)

• Bug fixes and Improvements

  • ONNX

    • Fixed supergroup misidentification bug upon MatMul-MatMul-Add sequence (ab63866)

  • Torch

    • Made compatible with PyTorch 1.13 (47fae94)
    • Made compatible with PyTorch 2.9 (283ecc1)

  • Common

    • Set priority among supergroups (6676a6c)