Repository navigation

Add GPTQ quantization support for K-last MoE architectures - #2610

Merged

Ti-Tai Wang (titaiwangms) merged 53 commits into

mainfrom

b1-gptq-moe

Aug 14, 2026

Merged

Add GPTQ quantization support for K-last MoE architectures#2610
Ti-Tai Wang (titaiwangms) merged 53 commits into
mainfrom
b1-gptq-moe

Commits on May 15, 2026
Commits on May 16, 2026
- QuantLinearNbit.from_quant_tensor: reshape-only, no unpack/repack
  
  Jambay Kinley (jambayk)
  and
  Copilot
  committed
  27458c0
- Make qzeros optional on QuantLinearNbit / QuantEmbeddingNbit
  
  Jambay Kinley (jambayk)
  and
  Copilot
  committed
  6cbacad
- Drop placeholder Constant for qzeros in MatMulNBits symbolic
  Copilot
  committed
  d758ef3
- Address PR feedback: factor quant target selection
  Copilot
  committed
  46c8c7a
- Drop dead writes to self.modules_to_not_convert / self._skip_patterns
  Copilot
  committed
  4b66da5
- Make quant_info a parameter-level attribute; unify rank handling
  Copilot
  committed
  f8bc453
- Drop QuantTarget dataclass; yield (module, pname, full_name) tuples
  Copilot
  committed
  a488f23
- Address rubber-duck review findings
  Copilot
  committed
  9ecdd50
- Address Copilot PR feedback + add forward parity tests
  Copilot
  committed
  de4c04e
Commits on Jul 21, 2026
- Initial plan
  Copilot
  authored
  fbd60bd
- Merge jambayk/moe-quant: integrate QuantTensor MoE quantization onto main
  copilot
  authored
  235e6de
- Override precedence: insertion-order first-match; add ReDoS-safe regex validator + tests
  copilot
  authored
  4c50d8b
- Add MoE selection/tensor regression tests + PyTorch Native RTN docs
  copilot
  authored
  a0aa4cc
- Central 3D movement-op guard + real-HF Mixtral round-trip test + docs
  copilot
  authored
  c41a618
Commits on Jul 22, 2026
- Add Gptq+Rtn composition test; make lintrunner clean across changed files
  copilot
  authored
  8035bb9
Commits on Jul 23, 2026
- Merge branch 'main' into copilot/resume-jambayk-moe-quant-extend-rtn-weight
  Ti-Tai Wang (titaiwangms)
  authored
  3a133f1
Commits on Jul 24, 2026
- Merge branch 'main' into copilot/resume-jambayk-moe-quant-extend-rtn-weight
  Ti-Tai Wang (titaiwangms)
  authored
  59a7f49
Commits on Jul 30, 2026
- Round-1/round-2 remediation: MoE quant regex/placeholder/index safety fixes
  Ti-Tai Wang (titaiwangms)
  committed
  d5c5641
Commits on Jul 31, 2026
Commits on Aug 1, 2026
- Fix Critical/Major bugs found in full PR review
  
  Ti-Tai Wang (titaiwangms)
  and
  Copilot
  committed
  6baed50
Commits on Aug 4, 2026
- Merge remote-tracking branch 'origin/main' into copilot/resume-jambayk-moe-quant-extend-rtn-weight
  Ti-Tai Wang (titaiwangms)
  committed
  2225683
Commits on Aug 5, 2026
Commits on Aug 6, 2026
- Merge remote-tracking branch 'origin/main' into b1-gptq-moe
  Ti-Tai Wang (titaiwangms)
  committed
  f1fa238
Commits on Aug 7, 2026
- Add dual fallback-threshold condition for MoE GPTQ calibration
  Ti-Tai Wang (titaiwangms)
  committed
  9ea3de9
- Address full-team review findings on GPTQ MoE fallback logic
  Ti-Tai Wang (titaiwangms)
  committed
  1204db5
Commits on Aug 11, 2026
Commits on Aug 12, 2026
Commits on Aug 13, 2026
- Merge remote-tracking branch 'origin/main' into b1-gptq-moe-local
  Ti-Tai Wang (titaiwangms)
  committed
  cab0b8a
- Exclude Mamba/SSM submodules from generic quantization selection
  Ti-Tai Wang (titaiwangms)
  committed
  711dfad
Commits on Aug 14, 2026
- Merge remote-tracking branch 'origin/main' into b1-gptq-moe-local
  Ti-Tai Wang (titaiwangms)
  committed
  91ccd32
- lint and spellcheck
  
  Jambay Kinley (jambayk)
  authored and
  Ti-Tai Wang (titaiwangms)
  committed
  ed15186

Commits

Commits on May 15, 2026

Commits on May 16, 2026

Commits on Jul 21, 2026

Commits on Jul 22, 2026

Commits on Jul 23, 2026

Commits on Jul 24, 2026

Commits on Jul 30, 2026

Commits on Jul 31, 2026

Commits on Aug 1, 2026

Commits on Aug 4, 2026

Commits on Aug 5, 2026

Commits on Aug 6, 2026

Commits on Aug 7, 2026

Commits on Aug 11, 2026

Commits on Aug 12, 2026

Commits on Aug 13, 2026

Commits on Aug 14, 2026