intel/auto-round

Pull Requests Commits

fix: handle GLM draft layers through predefined ignores

Shubham-Padkonde committed 10 days ago

1330a096

fix: preserve GLM nextn weights during checkpoint copying

Shubham-Padkonde committed 26 days ago

f944f16a

add mxuint (#2425)

wenhuach21 committed 26 days ago

Verified 7c4b6e92

update README.md (#2428)

wenhuach21 committed 26 days ago

Verified 4286cf93

fix: return the repaired string from escape_invalid_json_backslashes (#2419)

Shubham-Padkonde committed 26 days ago

Verified ba12c951

Keep the case of the AR_WORK_SPACE path (#2417)

Shubham-Padkonde committed 26 days ago

Verified 11949312

fix: honor AR_WORK_SPACE for the AudioCaps dataset cache (#2420)

Shubham-Padkonde committed 26 days ago

Verified 1ace02c5

INC integration test bug fix (#2423)

chensuyue committed 26 days ago

Verified 5606837a

NeUQI grid search for optimized RTN (--enable_neuqi: joint asym scale/zp search, two-stage sym search, frozen-init anchor) (#2341)

avtc committed 30 days ago

Verified 44b48d60

fix(xpu): make ARK SDPA graph-capture safe for torch.xpu.XPUGraph (#2416)

luoyu-intel committed 32 days ago

Verified a4f01f85

bug fix: prune static kv/attention granularity keys when dtype not set (#2415)

yiliu30 committed 32 days ago

Verified a20f5297

enhance attention module with original implementation retrieval and context restoration (#2408)

xin3he committed 32 days ago

Verified 5eceba6c

bug fix: layer_config mismatch due to transformers rename; fix model free stream mode detection (#2411)

xin3he committed 32 days ago

Verified 24d14371

Fix CWE-22 (path traversal) in checkpoint loading issues (#2389)

lvliang-intel committed 32 days ago

Verified fe973526

fix: materialize checkpoint-backed cpu buffers (#2412)

n1ck-guo committed 32 days ago

Verified 6c4d95cb

fix: avoid excessive memory during calibration dataset casting (#2405)

changwangss committed 34 days ago

Verified 22d6e28c

Revert "refactor: decouple datatype quantization from algorithms (#2406)

n1ck-guo committed 34 days ago

Verified 86e97adf

fix: update get_predefined_ignore_layers to include shared_expert_gat… (#2397)

xin3he committed 34 days ago

Verified 6db9435f

fix: update tensor summary creation to directly use tensors_to_restor… (#2401)

xin3he committed 34 days ago

Verified bff3deab

ark: add BF16/FP16 Sparge variants and reduce compilation RAM (#2145)

yiliu30 committed 35 days ago

Verified 402b0edd

feat: use OpenS2V for diffusion calibration (#2368)

changwangss committed 35 days ago

Verified 924b3987

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

Shubham-Padkonde committed 35 days ago

Verified 08832178

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

Shubham-Padkonde committed 35 days ago

Verified 89c3dfe9

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

Shubham-Padkonde committed 35 days ago

Verified 3f6c8190

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

Shubham-Padkonde committed 35 days ago

Verified 0aa5e351

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

xin3he committed 35 days ago

Verified f4d2c162

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

xin3he committed 35 days ago

Verified 8447db40

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

xin3he committed 36 days ago

Verified c29847ef

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

n1ck-guo committed 36 days ago

Verified a5f6ddd8

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

xin3he committed 38 days ago

Verified a6c612fa

Older