Pull Requests intel/auto-round

Move the zero point to CPU for asymmetric layers too

#2394 by Shubham-Padkonde was merged 2026-09-21 06:54 0.16.0

Apply autocast only when mixed precision is enabled

#2393 by Shubham-Padkonde was merged 2026-09-21 06:52 0.16.0

fix: fake format has no FP32 tensor restoration

#2392 by xin3he was merged 2026-09-21 02:34 0.16.0

fix model_free saving with single shard and MXFP fake format evaluation

#2391 by xin3he was merged 2026-09-21 02:35 0.16.0

Fix CWE-22 (path traversal) in checkpoint loading issues

#2389 by lvliang-intel was merged 2026-09-24 01:17 0.16.0

fix: update import to include get_model_path for model retrieval

#2388 by xin3he was merged 2026-09-20 06:49 0.16.0

Cast float32 weights to the export dtype when saving

#2385 by Shubham-Padkonde was merged 2026-09-21 02:52 0.16.0

fix:reduce memory usage when loading native fused MoE checkpoints

#2382 by n1ck-guo was merged 2026-09-20 02:39 0.16.0

fix: share NVFP4 global scales for Wan QKV projections bug

#2379 by changwangss was merged 2026-09-18 07:29 0.16.0

patch triangular_solve for Qwen3.8 model for hpu

#2377 by lkk12014402 was merged 2026-09-18 02:48 0.16.0

feat: implement optimized RTN toggle for NVFP4 E5M3 quantization and update CLI handling

#2376 by xin3he was closed 2026-09-21 03:19 0.16.0

fix: restore FP32 tensors saved as FP16/BF16 during tensor copying

#2375 by xin3he was merged 2026-09-18 08:07 0.16.0

bug fix: fix NVFP4 evaluation with input_ global_scale handling and testing

#2374 by xin3he was merged 2026-09-17 08:41 0.16.0

Remove deprecated files

#2373 by chensuyue was merged 2026-09-17 13:44 0.16.0

remove bagel specific constraints

#2372 by WeiweiZhang1 was merged 2026-09-17 06:09 0.16.0

feat: use OpenS2V for diffusion calibration

#2368 by changwangss was merged 2026-09-21 08:36 0.16.0

docs: clarify diffusion export format

#2367 by changwangss was merged 2026-09-17 03:40

feat: add NVFP4 static KV cache quantization support

#2366 by yiliu30 was merged 2026-09-17 05:05 0.16.0

refine: drop no-op RotationConfig.__init__ (#2036)

#2364 by lkk12014402 was closed 2026-09-15 08:56

[Accuracy improvement] Re-implement opt-rtn for both imatrix and model-free, covering INT & NVFP

#2362 by xin3he was closed 2026-09-22 01:57 0.17.0

Refactor test timeouts and update dependencies for performance

#2361 by XuehaoSun was merged 2026-09-16 03:42

refactor: decouple datatype quantization from algorithms

#2360 by n1ck-guo was merged 2026-09-16 08:15 0.16.0

fix: preserve declared FP32 tensors during model dtype conversion

#2359 by changwangss was merged 2026-09-15 06:30

perf: avoid unused SVD reconstruction and select the CUDA driver in SVDQuant

#2356 by changwangss was merged 2026-09-16 06:05

fix: reload Diffusers blocks from component checkpoints

#2353 by changwangss was merged 2026-09-14 04:18

fix qwen35 regression

#2352 by wenhuach21 was merged 2026-09-12 12:38

fix: update key cache access for compatibility with newer DynamicCache implementation

#2348 by chensuyue was merged 2026-09-14 07:44

fix: support list inputs in diffusion tuning cache

#2346 by changwangss was merged 2026-09-15 02:17

add fineweb-edu dataset as calibration bakeup ready

#2345 by WeiweiZhang1 was merged 2026-09-17 05:21 0.16.0

Update README.md

#2343 by wenhuach21 was merged 2026-09-11 06:35