Commit history

1. Commits on Oct 23, 2026

   • Fix unquantized correction inputs during low-rank candidate scoring (#23)

     rootonchair

     authored
     e8a623cView commit details

     Copy full SHA for e8a623c

     Browse repository at this point
2. Commits on Oct 17, 2026

   • Qwen image 2.1 calibration and gptq fixes (#20)

     Show description for 2f4893b
     

     rootonchair

     and

     claude

     authored
     2f4893bView commit details

     Copy full SHA for 2f4893b

     Browse repository at this point
3. Commits on Aug 25, 2026

   • Fix/torch dequant runtime (#19)

     Show description for 0965874
     

     rootonchair

     and

     claude

     authored
     0965874View commit details

     Copy full SHA for 0965874

     Browse repository at this point

   • Fix smoothing-basis mismatches in low-rank search scoring (review findings #27, #09) (#18)

     Show description for 581d04d
     

     rootonchair

     and

     claude

     authored
     581d04dView commit details

     Copy full SHA for 581d04d

     Browse repository at this point
4. Commits on Aug 18, 2026

   • Bound smoothing candidate batches by memory, not just candidate count (#16)

     Show description for 777ebbe
     

     rootonchair

     and

     claude

     authored
     777ebbeView commit details

     Copy full SHA for 777ebbe

     Browse repository at this point

   • Materialize offloaded weights instead of detaching Accelerate hooks (#15)

     Show description for d5076a0
     

     rootonchair

     and

     claude

     authored
     d5076a0View commit details

     Copy full SHA for d5076a0

     Browse repository at this point

   • Add Krea-2-Turbo example with chained calibration scopes (#17)

     Show description for eeeee80
     

     rootonchair

     and

     claude

     authored
     eeeee80View commit details

     Copy full SHA for eeeee80

     Browse repository at this point
5. Commits on Aug 16, 2026

   • Feat/ram improvement (#14)

     Show description for 25ca35e
     

     rootonchair

     and

     claude

     authored
     25ca35eView commit details

     Copy full SHA for 25ca35e

     Browse repository at this point

   • Default generic scanners to all-targets scope capture (#13)

     Show description for b57281e
     

     rootonchair

     and

     claude

     authored
     b57281eView commit details

     Copy full SHA for b57281e

     Browse repository at this point
6. Commits on Jul 28, 2026

   • Accelerate quantize time by batching smooth candidate score (#11)

     Show description for fc84fb6
     

     rootonchair

     and

     claude

     authored
     fc84fb6View commit details

     Copy full SHA for fc84fb6

     Browse repository at this point
7. Commits on Jul 26, 2026

   • Fix Accelerate hook lifecycle crash during LTX-2.3 quantization, speed up scope replay (#10)

     Show description for 423931e
     

     rootonchair

     and

     claude

     authored
     423931eView commit details

     Copy full SHA for 423931e

     Browse repository at this point
8. Commits on Jul 19, 2026

   • Avoid NVFP4 quantization distance-matrix OOM (#9)

     Show description for 7e65dab
     rootonchair

     authored
     7e65dabView commit details

     Copy full SHA for 7e65dab

     Browse repository at this point

   • Auto quantize HF model (#7)

     Show description for 751108b
     rootonchair

     authored
     751108bView commit details

     Copy full SHA for 751108b

     Browse repository at this point

   • Add GPTQ Residual Rounding for SVDQuant (#3)

     Show description for b7de5b4
     htrvu

     authored
     b7de5b4View commit details

     Copy full SHA for b7de5b4

     Browse repository at this point
9. Commits on Jul 5, 2026

   • Remove smooth_factor_orig from exports (#6)

     rootonchair

     authored
     ec9da89View commit details

     Copy full SHA for ec9da89

     Browse repository at this point

   • Refactor projector quantization dispatch (#4)

     Show description for e4ca14f
     rootonchair

     authored
     e4ca14fView commit details

     Copy full SHA for e4ca14f

     Browse repository at this point
10. Commits on Jul 1, 2026

    • Merge pull request #2 from rootonchair/fix/nunchaku-biasless-export

      Show description for f146b02
      rootonchair

      authored
      f146b02View commit details

      Copy full SHA for f146b02

      Browse repository at this point

    • Preserve shifted wrapper target paths

      rootonchair

      committed
      2ada028View commit details

      Copy full SHA for 2ada028

      Browse repository at this point

    • Limit FLUX.1 extra weight targets to NVFP4

      rootonchair

      committed
      1b615c3View commit details

      Copy full SHA for 1b615c3

      Browse repository at this point

    • Simplify Nunchaku shift bias handling

      rootonchair

      committed
      71fef56View commit details

      Copy full SHA for 71fef56

      Browse repository at this point

    • Preserve shifted Nunchaku runtime bias semantics

      rootonchair

      committed
      21e3000View commit details

      Copy full SHA for 21e3000

      Browse repository at this point

    • Match FLUX.1 activation shift target scope

      rootonchair

      committed
      3f6c268View commit details

      Copy full SHA for 3f6c268

      Browse repository at this point

    • Remove global activation shift option

      rootonchair

      committed
      4b1cf2bView commit details

      Copy full SHA for 4b1cf2b

      Browse repository at this point
11. Commits on Jun 30, 2026

    • Fix Nunchaku biasless export handling

      rootonchair

      committed
      202ee08View commit details

      Copy full SHA for 202ee08

      Browse repository at this point

    • Merge pull request #1 from rootonchair/fix/flux2-klein-examples

      Show description for cd5b05f
      rootonchair

      authored
      cd5b05fView commit details

      Copy full SHA for cd5b05f

      Browse repository at this point

    • Update FLUX2 Klein text-to-image examples

      rootonchair

      committed
      c08a7d3View commit details

      Copy full SHA for c08a7d3

      Browse repository at this point
12. Commits on Jun 24, 2026

    • Fix replay tensor handling

      rootonchair

      committed
      5e5df13View commit details

      Copy full SHA for 5e5df13

      Browse repository at this point
13. Commits on Jun 18, 2026

    • Ensure exported tensors are contiguous

      rootonchair

      committed
      6249894View commit details

      Copy full SHA for 6249894

      Browse repository at this point
14. Commits on Jun 17, 2026

    • Skip LTX2 gate logit quantization

      rootonchair

      committed
      85f3c0eView commit details

      Copy full SHA for 85f3c0e

      Browse repository at this point
15. Commits on Jun 16, 2026

    • Use default Nunchaku layout in FLUX Klein examples

      rootonchair

      committed
      ba3e5c2View commit details

      Copy full SHA for ba3e5c2

      Browse repository at this point

    • Add upstream example calibration updates

      rootonchair

      committed
      25aa7bfView commit details

      Copy full SHA for 25aa7bf

      Browse repository at this point

    • Default SVDQuant targets to Nunchaku layout

      rootonchair

      committed
      f60c585View commit details

      Copy full SHA for f60c585

      Browse repository at this point
16. Commits on Jun 8, 2026

    • Format source with Ruff

      rootonchair

      committed
      f12f700View commit details

      Copy full SHA for f12f700

      Browse repository at this point

    • Fix Accelerate hook removal for inference tensors

      rootonchair

      committed
      f0a67c7View commit details

      Copy full SHA for f0a67c7

      Browse repository at this point

    • Route core logs through QuantizationLogger

      rootonchair

      committed
      eea2f67View commit details

      Copy full SHA for eea2f67

      Browse repository at this point