Issues

• [Feature]: Save important AutoRound hyperparameters in the llmc format, just like in the AutoRound format.

  enhancement
  intel/auto-round#2469 · Filter by author wenhuach21 opened on Oct 10, 2026 · 0.17.0

• [Bug]: offloder cleaning failed

  bug
  intel/auto-round#2467 · Filter by author wenhuach21 opened on Oct 10, 2026
  

• [ARK] Per-head hybrid attention

  Feature

  intel/auto-round#2465 · Filter by author chensuyue opened on Oct 10, 2026 · 0.17.0
  

• [ARK] Fused HMT Quant

  Feature

  intel/auto-round#2464 · Filter by author chensuyue opened on Oct 10, 2026 · 0.17.0
  

• [ARK] SageAttention V3 (Mixed 4/8)

  Feature

  intel/auto-round#2463 · Filter by author chensuyue opened on Oct 10, 2026 · 0.17.0
  

• [ARK] Sparse attention kernel

  Feature

  intel/auto-round#2462 · Filter by author chensuyue opened on Oct 10, 2026 · 0.17.0
  

• [ARK] Fused kernel for SVDQuant

  Feature

  intel/auto-round#2461 · Filter by author chensuyue opened on Oct 10, 2026 · 0.17.0
  

• [ARK] W2(4)A8(FP8) GEMM/MoE

  Feature

  intel/auto-round#2460 · Filter by author chensuyue opened on Oct 10, 2026 · 0.17.0
  

• AI-based GH issue assign/analysis/auto-fix

  intel/auto-round#2455 · Filter by author chensuyue opened on Oct 9, 2026 · 0.17.0
  

• [Bug] Unexpectedly high accuracy with 2-bit fake quantization, possibly due to FP32 tensor restoration

  bug
  intel/auto-round#2453 · Filter by author wenhuach21 opened on Oct 9, 2026
  2

  

• [Feature]: Extract an extensible shared smoothing engine for AWQ and SVDQuant

  enhancementhigh priority
  intel/auto-round#2451 · Filter by author changwangss opened on Oct 8, 2026 · 0.17.0
  

• [Feature]: code refinements

  enhancement
  intel/auto-round#2447 · Filter by author wenhuach21 opened on Oct 8, 2026
  

• [Feature]: validate AutoRound SVDQuant NVFP4 export with vLLM-Omni

  enhancementhigh priority
  intel/auto-round#2446 · Filter by author changwangss opened on Oct 8, 2026 · 0.17.0
  

• Quality Assurance for v0.17.0

  intel/auto-round#2442 · Filter by author XuehaoSun opened on Oct 8, 2026 · 0.17.0
  1

  

• [Model]: Qwen 3.8 Flash Next on 4x B70s - HUGE SUCCESS

  Model Support
  intel/auto-round#2440 · Filter by author KaloyanNaumov opened on Oct 8, 2026
  

• [Bug] GLM-5.3-Flash (Glm5Next) W4A16 release quantizes the MTP/nextn draft layer - vLLM MTP load fails (KeyError routed_experts.w2_qweight)

  intel/auto-round#2427 · Filter by author ming616 opened on Sep 29, 2026
  3

  

• [Model]: MXFP4 tuning enhancement

  Model Support
  intel/auto-round#2424 · Filter by author lkk12014402 opened on Sep 28, 2026
  

• [Bug]: vllm eval failed

  bug
  intel/auto-round#2414 · Filter by author wenhuach21 opened on Sep 24, 2026 · 0.17.0
  

• [llama3.1 8b][mxfp8] inference error on sm120 when using FLASHINFER attention backend

  intel/auto-round#2413 · Filter by author XuehaoSun opened on Sep 23, 2026 · 0.17.0
  1

  

• [gemma-4-31B] 4-GPUs RuntimeError: cannot enter context: <_contextvars.Context object at 0x7e35759202c0> is already entered

  intel/auto-round#2410 · Filter by author XuehaoSun opened on Sep 23, 2026 · 0.17.0
  

• [Feature]: move diffusion code from signround quantizer

  enhancement
  intel/auto-round#2399 · Filter by author wenhuach21 opened on Sep 21, 2026 · 0.17.0
  

• RFC: Data-parallel block tuning (--parallel_quantization)

  enhancement
  intel/auto-round#2378 · Filter by author avtc opened on Sep 17, 2026
  2

  

• [Bug]: Kimi-K2.6_mxfp4_fp8attn_LLMC vllm eval qkv scale is not loaded

  bug
  Bug

  intel/auto-round#2370 · Filter by author chensuyue opened on Sep 16, 2026 · 0.17.0
  4

  

• [Bug]: MiniMax-M2.7_mxfp4_mixed_fp8kv_LLMC vllm eval failed

  bug
  Bug

  intel/auto-round#2369 · Filter by author chensuyue opened on Sep 16, 2026 · 0.17.0
  

• [Bug]: crash + resume (AR_RESUME_DIR, immediate saving) silently duplicates already-quantized modules as stale fp weights in the final artifact

  bug
  intel/auto-round#2350 · Filter by author avtc opened on Sep 11, 2026 · 0.17.0