intel/auto-round

Pull Requests Commits

fix: lazily import diffusion tuning cache

changwangss committed 55 days ago

b178a7c7

refactor: rename tuning batch forward helper

changwangss committed 55 days ago

ff91ae3d

refactor: simplify tuning cache cleanup after review

changwangss committed 55 days ago

45620473

refactor: retain existing tuning cache cleanup

changwangss committed 55 days ago

b4910456

refactor: extract diffusion tuning cache helpers

changwangss committed 55 days ago

1465d4fd

feat: use OpenS2V for diffusion calibration (#2368)

changwangss committed 55 days ago

Verified 924b3987

Honour the padding mask in SignRoundV2's plain loss (#2395)

Shubham-Padkonde committed 55 days ago

Verified 08832178

Move the zero point to CPU for asymmetric layers too (#2394)

Shubham-Padkonde committed 55 days ago

Verified 89c3dfe9

Apply autocast only when mixed precision is enabled (#2393)

Shubham-Padkonde committed 55 days ago

Verified 3f6c8190

Cast float32 weights to the export dtype when saving (#2385)

Shubham-Padkonde committed 55 days ago

Verified 0aa5e351

fix model_free saving with single shard and MXFP fake format evaluation (#2391)

xin3he committed 55 days ago

Verified f4d2c162

fix: fake format has no FP32 tensor restoration (#2392)

xin3he committed 55 days ago

Verified 8447db40

fix: update import to include get_model_path for model retrieval (#2388)

xin3he committed 56 days ago

Verified c29847ef

fix:reduce memory usage when loading native fused MoE checkpoints (#2382)

n1ck-guo committed 56 days ago

Verified a5f6ddd8

fix: restore FP32 tensors saved as FP16/BF16 during tensor copying (#2375)

xin3he committed 58 days ago

Verified a6c612fa

fix: share NVFP4 global scales for Wan QKV projections (#2379)

changwangss committed 58 days ago

Verified 36079c49

[ARK] Improve XPU W4A16 WOQ decode with dense S4 DPAS path (#2331)

Zhenzhong1 committed 58 days ago

Verified eaf57a2a

patch triangular_solve for Qwen3.8 model for hpu (#2377)

lkk12014402 committed 58 days ago

Verified 8cd9b878

Remove deprecated files (#2373)

chensuyue committed 59 days ago

Verified fd077878

bug fix: fix NVFP4 evaluation with input_ global_scale handling and testing (#2374)

xin3he committed 59 days ago

Verified 27daf7f8

remove bagel specific constraints (#2372)

WeiweiZhang1 committed 59 days ago

Verified 4b3680b9

add fineweb-edu dataset as calibration bakeup (#2345)

WeiweiZhang1 committed 59 days ago

Verified 3938e152

feat: add NVFP4 static KV cache quantization support (#2366)

yiliu30 committed 59 days ago

Verified 04fca622

docs: clarify diffusion export format (#2367)

changwangss committed 59 days ago

Verified ea838e56

Add blockwise rotation. (#2087)

lkk12014402 committed 60 days ago

Verified 9b746ff9

fix: recognize Modular Diffusers pipelines as diffusion models (#2120)

Ar4ikov committed 60 days ago

Verified 0d76712e

refactor: decouple datatype quantization from algorithms (#2360)

n1ck-guo committed 60 days ago

Verified 2e82448a

fix: save the config of diffusers models on export (#2119)

Ar4ikov committed 60 days ago

Verified 1983b65a

feat(ark): fused Hadamard + MXFP4 activation quantization kernel

jijiaz committed 60 days ago

Verified ecc5b005

perf: avoid unused SVD reconstruction and select the CUDA driver in SVDQuant (#2356)

changwangss committed 60 days ago

Verified 903ba0ce

Older