intel/auto-round

Pull Requests Commits

ark: restore woqgemm validation and blob size plumbing

yiliu30 committed 73 days ago

523cccab

ark: revert dense lse and varlen merge payload

yiliu30 committed 73 days ago

48ee4c8b

Merge remote-tracking branch 'origin/main' into sparse-attn-prefill-clean

yiliu30 committed 73 days ago

1b4af670

bump release version to 0.14.0 (#2002)

chensuyue committed 73 days ago

Verified 24bfce00

update rotation doc. (#1999)

lkk12014402 committed 73 days ago

Verified 5ee45db7

pin transformers (#1998)

yiliu30 committed 73 days ago

Verified fcd1c854

Fix cuda ut fail caused by used instructions for base model (#1994)

n1ck-guo committed 73 days ago

Verified 43f443e6

ark: split sparse prefill entrypoints explicitly

yiliu30 committed 73 days ago

d8cf7098

ark: keep dense python api separate from sparse

yiliu30 committed 73 days ago

f3878e0e

ark: drop dead sparse profile surface

yiliu30 committed 73 days ago

cad6f302

ark: trim sparse sdpa compile surface

yiliu30 committed 73 days ago

549b7f24

ark: isolate sparse sdpa wrapper from dense path

yiliu30 committed 73 days ago

9c8e87c1

ark: restore dense Sage headers to main

yiliu30 committed 73 days ago

c904ba9d

ark: enable decoupled qtile256 sparse path on HND

yiliu30 committed 74 days ago

f0a6595e

ark: add sparse Wan and Flux runners

yiliu30 committed 74 days ago

69903123

ark: add sparse preprocess and benchmarking paths

yiliu30 committed 74 days ago

2d6df5e0

ark: split sparse sdpa kernel stack from dense path

yiliu30 committed 74 days ago

d98f792c

feat: add return_lse support for SDPA and SAGEV1 kernel — enable sequence-parallel (Ring Attention) merge (#1991)

luoyu-intel committed 75 days ago

Verified b481cebb

Enable auto-round-bot for CI comments (#1993)

chensuyue committed 76 days ago

Verified b63a89f5

Implement AutoScheme support for model-free packing with validation a… (#1990)

xin3he committed 76 days ago

Verified cd10607c

Update CUDA UT to cu130 (#1992)

XuehaoSun committed 76 days ago

Verified 52564823

[AutoScheme] fix lm_head no_grad issue and support MOE model with shared mix_score (#1971)

xin3he committed 76 days ago

Verified e6b35d13

fix: use spawn instead of fork for dataset subprocess on macOS (#1980)

ramkrishs committed 77 days ago

Verified 1a7ed495

fix few bugs cause less accuracy of gguf format (#1960)

n1ck-guo committed 77 days ago

Verified e4404be7

Minor fix for auto-round-lib CI workflow (#1983)

chensuyue committed 77 days ago

Verified c945a182

fix: promote None-initialized kwarg to list in forward_capture (#1979)

ramkrishs committed 77 days ago

Verified 0622d472

fix: avoid silent weight loss when OffloadManager fails to save a block (#1982)

xin3he committed 77 days ago

Verified 64378412

rename the base rotation (deterministic/random hadamard) directory name. (#1956)

lkk12014402 committed 77 days ago

Verified 45dcb198

Add llmc ut back (#1974)

yiliu30 committed 78 days ago

Verified d763ae04

Fix gguf cuda ut (#1975)

n1ck-guo committed 78 days ago

Verified 1c6f4e96

Older