0.17.0

Issues

• [Feature]: Support FP/INT W4A4 recipe

  Stale
  intel/auto-round#2009 · Filter by author chensuyue opened on Jul 7, 2026 · 0.17.0
  1

  

• [Feature]: refine lm-head quantization

  enhancement
  intel/auto-round#2176 · Filter by author wenhuach21 opened on Aug 18, 2026 · 0.17.0

• [Feature]: support reap

  enhancement
  intel/auto-round#2237 · Filter by author wenhuach21 opened on Aug 26, 2026 · 0.17.0

• [Feature]: support and eval H-Scale: Hessian-Guided Scale Refinement for NVFP4 Sub-Byte LLM Inference

  enhancement
  intel/auto-round#2262 · Filter by author wenhuach21 opened on Aug 31, 2026 · 0.17.0

• [Feature]: support and eval VSRAQ

  enhancement
  intel/auto-round#2008 · Filter by author wenhuach21 opened on Jul 7, 2026 · 0.17.0
  

• [Feature]: auto-scheme shared_layers supports vllm pattern as default

  enhancement
  intel/auto-round#2183 · Filter by author wenhuach21 opened on Aug 18, 2026 · 0.17.0
  

• [Feature]: support FP8+WOQ mix datatypes inference in transformers and vllm

  enhancement
  intel/auto-round#2225 · Filter by author wenhuach21 opened on Aug 25, 2026 · 0.17.0

• [Bug]: dump auto-round key hype-parameters to conifg.json

  bug
  intel/auto-round#2154 · Filter by author wenhuach21 opened on Aug 13, 2026 · 0.17.0
  

• [Bug]: auto-scheme wrongly changes dataset when oom

  bug
  intel/auto-round#2185 · Filter by author wenhuach21 opened on Aug 18, 2026 · 0.17.0
  2

  

• [Bug]: unexpected enable_quant_input: false in config.json for FP8_BLOCK

  bug
  intel/auto-round#2224 · Filter by author wenhuach21 opened on Aug 25, 2026 · 0.17.0
  

• [Model]: TR-HASH MoE with stacked 3D expert weights

  Model Support
  intel/auto-round#2211 · Filter by author Complexity-ML opened on Aug 21, 2026 · 0.17.0

• [Feature]: accuracy issue Intel/Qwen3.6-27B-int4-AutoRound

  enhancement
  intel/auto-round#2142 · Filter by author wenhuach21 opened on Aug 11, 2026 · 0.17.0
  5

  

• [Feature]: support data parallel

  enhancement
  intel/auto-round#2307 · Filter by author wenhuach21 opened on Sep 6, 2026 · 0.17.0

• Create default mixed calibration dataset

  high priority
  Task

  intel/auto-round#2071 · Filter by author chensuyue opened on Jul 21, 2026 · 0.17.0
  5

  

• [Feature]: support nlayers to support load a small model for debugging

  enhancementlow priority
  intel/auto-round#2336 · Filter by author wenhuach21 opened on Sep 10, 2026 · 0.17.0
  3

• Add a CI check to prevent coverage regressions in pull requests

  Task

  intel/auto-round#2347 · Filter by author chensuyue opened on Sep 11, 2026 · 0.17.0
  3

  

• [Bug]: imbalanced memory usages across cards

  bug
  intel/auto-round#2254 · Filter by author wenhuach21 opened on Aug 28, 2026 · 0.17.0
  2

• [Feature]: svdquant supports llm and other schemes(W2A16)

  enhancement
  intel/auto-round#2218 · Filter by author wenhuach21 opened on Aug 24, 2026 · 0.17.0
  

• [Bug]: COCO2014 is not a good calibration dataset for Wan quantization

  bug
  intel/auto-round#2381 · Filter by author changwangss opened on Sep 18, 2026 · 0.17.0
  

• [Model]: nvfp4 wan 2.2 t2v

  Model Support
  intel/auto-round#2269 · Filter by author changwangss opened on Sep 1, 2026 · 0.17.0
  

• [Feature]: mxfp4 dual-scale investigate

  intel/auto-round#2199 · Filter by author changwangss opened on Aug 20, 2026 · 0.17.0
  

• Track Wan2.2 dual-transformer quantization memory and tuning cost

  intel/auto-round#2235 · Filter by author changwangss opened on Aug 26, 2026 · 0.17.0
  

• [Feature]: refine hardmard code

  enhancementhigh priority
  intel/auto-round#2036 · Filter by author wenhuach21 opened on Jul 9, 2026 · 0.17.0
  2

  

• Refine device operations into a unified DeviceManager and ARDevice abstraction

  Unread
  intel/auto-round#2344·lvliang-intel opened Sep 11, 2026 · Updated Sep 18, 2026
  2

• [Feature]: Refine device operations into a unified class to simplify support for new devices

  enhancement
  intel/auto-round#1788 · Filter by author wenhuach21 opened on May 8, 2026 · 0.17.0