All pull requests

Pull requests

• [TorchFX] Store per-channel zero points as int32 to avoid wraparound in dequantization

  NNCF PT
  #4313·WeakyWeak opened Nov 10, 2026·

  ·
  1

• [TorchFX] Fix post-quantize weight compression for asymmetric weights

  NNCF PT
  #4312·WeakyWeak opened Nov 10, 2026·

  ·
  1

• [ci] Pin node version to 26.13.0 in pre-commit

  #4309·AlexanderDokuchaev opened Nov 10, 2026Collaborator·

  ·

• Add prepare-release skill

  documentation
  #4308·AlexanderDokuchaev opened Nov 9, 2026Collaborator·

  ·

• Bump Version to 3.6.0

  #4307·AlexanderDokuchaev opened Nov 9, 2026Collaborator·

  ·
  1

• Release notes for NNCF v3.5.0

  documentationrelease target
  #4306·AlexanderDokuchaev opened Nov 9, 2026Collaborator·

  ·
  1

• [release_v350] Openvino 2026.5.0rc1

  documentationNNCF ONNXNNCF OpenVINONNCF PTrelease target
  #4305·AlexanderDokuchaev opened Nov 9, 2026Collaborator·

  ·
  1

• [TorchFX] Erase only the replaced constant in constant_update

  NNCF PT
  #4301·BigSmoKe07 opened Nov 6, 2026·

  ·
  1

  6

• [ONNX] Replace an existing external data dir metadata entry instead of adding a duplicate

  NNCF ONNX
  #4300·BigSmoKe07 opened Nov 6, 2026·

  ·
  1

  2

• [OV] SparseAtten operation

  #4298·AlexanderDokuchaev opened Nov 4, 2026Collaborator·

  ·

• [DRAFT] Add INT1_SYM and narrow_range; store INT2_SYM codes as signed i2

  documentationNNCF OpenVINO
  #4276·devajithvs opened Oct 21, 2026·

  ·

• [DRAFT] Added possibility to launch int8 with group_size and scale estimation

  APIdocumentationNNCF OpenVINO
  #4273·AsyaPronina opened Oct 19, 2026Collaborator·

  ·
  4

• debug usage of microsoft runner

  #4270·gkoscins opened Oct 14, 2026Contributor·

  ·
  4

• Bump datasets from 5.0.1 to 5.1.0

  python
  #4263·dependabot[bot] opened Oct 11, 2026Bot·

  ·
  1

  

• [ONNX] Fix incorrect weight tensor name determination when intermediate nodes exist between the weight and the node

  NNCF ONNX
  #4259·andrey-churkin opened Oct 6, 2026Contributor·

  ·
  4

• Bump the pip group across 15 directories with 1 update

  NNCF OpenVINONNCF PTNNCF PTQpython
  #4258·dependabot[bot] opened Oct 6, 2026Bot·

  ·

• Enable validation on python 3.15

  #4254·AlexanderDokuchaev opened Oct 5, 2026Collaborator·

  ·
  1

• [Torch][MoE] Support GroupedMM in Torch Backend

  #4209·anzr299 opened Sep 10, 2026Collaborator·

  ·

• Fix PT2E Mul weight quantization for parameters

  #4173·ricardoasouz opened Aug 15, 2026·

  ·
  5

  

• [FBC] Support models without the batch dimension

  #4149·darshil929 opened Jul 20, 2026Contributor·

  ·
  1

• Cache statistics in OpenVINO/ONNX Accuracy Restorer

  #4113·andersendsa opened Jun 25, 2026·

  ·
  2

• [Torch] Lora kernels compilation

  #4100·daniil-lyakhov opened Jun 15, 2026Contributor·

  ·
  6

• [OpenVINO] calibration_device parameter

  APINNCF ONNXNNCF OpenVINONNCF PTNNCF PTQ
  #4055·daniil-lyakhov opened Apr 28, 2026Contributor·

  ·
  7

• feat: add HQQ data-free weight compression algorithm

  NNCF OpenVINO
  #3981·abhayuvi opened Mar 13, 2026·

  ·
  8

• Register SingleConfigQuantizationPoint in CommonStatefulClassesRegistry

  NNCF Common
  #3978·abhayuvi opened Mar 12, 2026·

  ·
  2