Commits

Branch selector

 develop

User selector

quic-akhobare

Datepicker

All time

Commit History

Commits on Jan 2, 2025

• Fix AMP v2 slowdown issue (#3705)

  Show description for 421aece
  

  quic-akhobare

  and

  quic-seokan

  authored
  421aece

  Copy full SHA for 421aece

Commits on Oct 28, 2024

• Implement encoding rescaler (#3441)

  Show description for aa32576
  quic-akhobare

  authored
  aa32576

  Copy full SHA for aa32576

Commits on May 12, 2024

• Disable yaml export by default (#2952)

  Show description for c2d31dd
  

  quic-akhobare

  and

  quic-klhsieh

  authored
  c2d31dd

  Copy full SHA for c2d31dd

Commits on Apr 29, 2023

• Add fp32_acc instance variable (#2195)

  Show description for a8d02b8
  

  quic-akhobare

  and

  quic-saksrai

  authored
  a8d02b8

  Copy full SHA for a8d02b8

• Fixes accuarcy drop and export issues with fold to scale on Keras (#2194)

  Show description for c6c5c10
  

  quic-akhobare

  and

  quic-ernst

  authored
  c6c5c10

  Copy full SHA for c6c5c10

Commits on Apr 26, 2023

• Added support for half-float tensors for per-channel mode. Also added tests for static grid and learned grid for half-float models.

  Show description for 1f5b69c
  quic-akhobare

  committed
  1f5b69c

  Copy full SHA for 1f5b69c

• Added a test to check if we can tie quantizers and run eval and train

  Show description for e198e29
  quic-akhobare

  committed
  e198e29

  Copy full SHA for e198e29

• Added benchmarking tests for half models. Handle half-float tensors when updating stats in PyTorch tensor quantizers

  Show description for 03e49e3
  quic-akhobare

  committed
  03e49e3

  Copy full SHA for 03e49e3

Commits on Apr 5, 2023

• Disable editable mode stages (#2066)

  Show description for 25ea64d
  quic-akhobare

  authored
  25ea64d

  Copy full SHA for 25ea64d

Commits on Mar 29, 2023

• Added tests for QuantSim creation, eval and training for Conv3D models

  Show description for fac96c8
  quic-akhobare

  committed
  fac96c8

  Copy full SHA for fac96c8

Commits on Mar 28, 2023

• Added test models for Conv3d. Added BN fold code changes to handle Conv3d and BN3d layers.

  Show description for a3c3a1a
  quic-akhobare

  committed
  a3c3a1a

  Copy full SHA for a3c3a1a

Commits on Feb 28, 2023

• Moved imagenet_dataloader from aimet_torch to acceptance tests

  Show description for 730fd8a
  quic-akhobare

  committed
  730fd8a

  Copy full SHA for 730fd8a

• Moved example models from aimet_torch to tests, acceptance tests, and onnx folders

  Show description for e282817
  quic-akhobare

  committed
  e282817

  Copy full SHA for e282817

Commits on Feb 27, 2023

• Add jupyter notebook for AutoQuant 2.0 (#1967)

  Show description for 247161b
  

  quic-akhobare

  and

  quic-kyunggeu

  authored
  247161b

  Copy full SHA for 247161b

• Autoquant 2.0 user guide & API doc

  Show description for 8ff1477
  

  quic-akhobare

  and

  quic-kyunggeu

  authored
  8ff1477

  Copy full SHA for 8ff1477

Commits on Oct 28, 2022

• Added links to keras notebooks

  Show description for f91625a
  quic-akhobare

  committed
  f91625a

  Copy full SHA for f91625a

• Added links to TF adaround and autoquant

  Show description for 8044a97
  quic-akhobare

  committed
  8044a97

  Copy full SHA for 8044a97

• Removed links to missing notebooks. Organized the installation steps more clearly.

  Show description for ed27222
  quic-akhobare

  committed
  ed27222

  Copy full SHA for ed27222

• Added mechanism to invoke nbsphinx to render html versions of example notebooks. Added links to the same from the user guide.

  Show description for 57554d6
  quic-akhobare

  committed
  57554d6

  Copy full SHA for 57554d6

Commits on Oct 20, 2022

• Fixed: When new PerChannelQuantizer or AdaRoundQuantizer objects were created from existing other quantizers, unsigned symmetric and strict symmetric flags were not propagated.

  Show description for 9cb6157
  quic-akhobare

  committed
  9cb6157

  Copy full SHA for 9cb6157

Commits on Sep 29, 2022

• Removed unnecessary line ending colons

  Show description for 6b05d82
  quic-akhobare

  committed
  6b05d82

  Copy full SHA for 6b05d82

• Updated API docs to add cross-links to relevant User Guide sections

  Show description for d9bc076
  quic-akhobare

  committed
  d9bc076

  Copy full SHA for d9bc076

Commits on Jul 15, 2022

• Added updated default runtime config, also a per-channel one. Fixed necessary unit tests.

  Show description for 17bcc52
  quic-akhobare

  committed
  17bcc52

  Copy full SHA for 17bcc52

Commits on Jun 21, 2022

• Removed logic to export disabled parameters (bias) with 32-bit precision

  Show description for 6db0cb5
  quic-akhobare

  committed
  6db0cb5

  Copy full SHA for 6db0cb5

Commits on Jun 11, 2022

• Add capability to quantize weight shared parameters for tf quantsim

  Show description for 761c796
  quic-akhobare

  authored
  761c796

  Copy full SHA for 761c796

• Fixed random seed on one test to make it more deterministic

  Show description for 1d891bf
  quic-akhobare

  committed
  1d891bf

  Copy full SHA for 1d891bf

• Cleared stats before computing encodings in QcQuantizeOp in one-shot mode

  Show description for 1082302
  quic-akhobare

  committed
  1082302

  Copy full SHA for 1082302

Commits on May 31, 2022

• For parameters, removing the logic to compute encodings after TF QuantSim calibration

  Show description for 10bfdba
  quic-akhobare

  committed
  10bfdba

  Copy full SHA for 10bfdba

Commits on May 24, 2022

• Temporarily disabling some tests - parameter quantization for the embedding layers needs to be disabled.

  Show description for 3f87f2d
  quic-akhobare

  committed
  3f87f2d

  Copy full SHA for 3f87f2d

• Added mechanism to pre-compute weight encodings before starting calibration

  Show description for bcdbced
  quic-akhobare

  committed
  bcdbced

  Copy full SHA for bcdbced

Commits on May 3, 2022

• Minor formatting updates

  Show description for 32d61c3
  quic-akhobare

  committed
  32d61c3

  Copy full SHA for 32d61c3

• Fixed cmake to use variables found via the Find_Python package. Fixed build for TF CPU to compile the newly added unit tests.

  Show description for 774d1c3
  quic-akhobare

  committed
  774d1c3

  Copy full SHA for 774d1c3

• Added code to compute encodings using TF min/max functions. Added a unit test to test just the encoding computation logic.

  Show description for 5405510
  quic-akhobare

  committed
  5405510

  Copy full SHA for 5405510

Commits on Apr 26, 2022

• Added copyright header

  Show description for 21d57b3
  quic-akhobare

  committed
  21d57b3

  Copy full SHA for 21d57b3

• Added initial test code to enable us to unit test functions in the custom AIMET C++ ops for tensorflow

  Show description for ed5d601
  quic-akhobare

  committed
  ed5d601

  Copy full SHA for ed5d601

Pagination

Previous

Next