All pull requests

Pull requests

• bug fix: layer_config mismatch due to transformers rename; fix model free stream mode detection

  #2411·xin3he opened Sep 23, 2026Contributor·

  ·

  · 0.16.0
  13

• enhance attention module with original implementation retrieval…

  #2408·xin3he opened Sep 22, 2026Contributor·

  ·

  · 0.16.0
  12

• refactor: move diffusion tuning data handling out of SignRound

  #2404·changwangss opened Sep 21, 2026Contributor·

  ·
  3

• Reset context singletons between tests

  #2390·Shubham-Padkonde opened Sep 20, 2026Contributor·

  ·
  2

• Don't rewrite packed modules as float weights when resuming

  #2386·Shubham-Padkonde opened Sep 20, 2026Contributor·

  ·

  · 0.16.0
  1

  6

• feat: chunked tuning for lm_head-class layers

  #2383·avtc opened Sep 20, 2026Collaborator·

  ·
  16

• feat: add mxfp8 mxfp4 moe bdpas prefill

  #2371·a32543254 opened Sep 16, 2026Contributor·

  ·
  10

• [bugfix][refine]: drop no-op RotationConfig.__init__

  #2365·lkk12014402 opened Sep 15, 2026Contributor·

  ·
  1

  3

• refine dataset

  #2363·wenhuach21 opened Sep 15, 2026Contributor·

  ·
  6

• feat: multi-GPU quantization improvements -- batched device-local sea…

  #2357·avtc opened Sep 14, 2026Collaborator·

  ·
  5

• feat: data-parallel block tuning via --parallel_quantization

  #2351·avtc opened Sep 11, 2026Collaborator·

  ·

  · 0.16.0
  23

• Refine device operations into a unified DeviceManager and ARDevice abstraction

  #2344·lvliang-intel opened Sep 11, 2026Contributor·

  ·

  · 0.17.0
  2

• NeUQI grid search for optimized RTN (--enable_neuqi: joint asym scale/zp search, two-stage sym search, frozen-init anchor)

  #2341·avtc opened Sep 10, 2026Collaborator·

  ·
  18

• feat(ark): add INT4 S4 pre-packed Q*K kernel for SageAttention

  #2319·luoyu-intel opened Sep 8, 2026Contributor·

  ·

• Enhance offload cleanup handling for exception cases

  #2317·lvliang-intel opened Sep 8, 2026Contributor·

  ·
  8

• Recurrent Residual Quantization (RRQ) for LLMs

  #2308·luoyu-intel opened Sep 6, 2026Contributor·

  ·
  60

• support teq algo

  experimentalWIP
  #2301·WeiweiZhang1 opened Sep 4, 2026Contributor·

  ·
  2

• Xpu hmt mxfp4

  #2264·a32543254 opened Aug 31, 2026Contributor·

  ·

• Add lagrangian solver in AutoScheme

  #2221·wenhuach21 opened Aug 24, 2026Contributor·

  ·
  2

• feat: W4A8 ARK XPU MoE kernel (int4 weight / int8 compute) with prefill + decode

  #2143·a32543254 with CopilotAI opened Aug 11, 2026·

  ·
  48

  

• Fix AutoRound gradients for Kimi-VL routed experts

  #2097·lesj0610 opened Jul 28, 2026·

  ·

  · 0.16.0
  2

• Stream targeted block re-quantization's calibration pass

  #2064·aquilarubra opened Jul 19, 2026Collaborator·

  ·

  · 0.16.0
  18

• Add hpu inference in CI test

  #1976·chensuyue opened Jul 1, 2026Contributor·

  ·

• Add quantization support for DiffusionGemma

  Stale
  #1935·lvliang-intel opened Jun 17, 2026Contributor·

  ·
  24