Commit Graph

Select branches

Hide Pull Requests

ByronHsu-patch-1

add_chunk_size

austin362667/fix_distill_normalization_term

austin362667/fix_flex_tests

austin362667/llama_flex_attn

bhe/fix_tvd_test_for_interl_amd

byhsu/ack

byhsu/ce-reduction

byhsu/convergence_test_reset

byhsu/e2e-bm

byhsu/mlsys-benchmark

byhsu/modal-ci

byhsu/model-ci-poke

byhsu/poke-test-4

byhsu/rebuild-image

ci

fix/rope-dead-constexpr-batch-size-dynamic-shapes

fix_rocm

fix_unit_tests

gh-pages

gh-pages-backup

h1_test_branch

hpguo/lce_add_entropy_loss

hpguo/ruff_style_check

jaszhu/add_skicit_learn_version

jaszhu/ping_version

jkolehm/mq

jsd-opt

layer-norm-opt

main

megatron-vp-flce

nemotron-relu-squared-patch

pr-1279

pr-144

pr1340

pramodith/autotune_chunk_size

qwen2_5_vl

release-0_5_11

revert-364-patch-4

rocm-install

shisahni/add_hf_token

shisahni/cl_module

shisahni/dpo_format

shisahni/grpo_changes_save

shisahni/precision

shisahni/preference_refactor

shisahni/stash_dec9

shisahni/temp_branch

sshimizu/ci-workflow

sshimizu/h100-benchmarks

sshimizu/test

tcc/backend

tcc/bench_flce

tcc/bench_vis

tcc/format

tcc/geglu-casting

tcc/helion-flce

tcc/qwen3vl

tcc/ruff-ci

tcc/test-ci

vaibhav/cutile-flce-chunking-b200

vaibhav/cutile-flce-chunking-triton-policy

vaibhav/flce-2-chunk-size

vaibhav/flce-3-cutile-cublas

vaibhavjindal/experimental

vaibhavjindal/swiglu-experimental

vjindal/testing

yudai-remove-contact

yudai/geglu

#1

#10

#1000

#1001

#1002

#1003

#1004

#1005

#1006

#1007

#1008

#1009

#1010

#1014

#1016

#1017

#1018

#1019

#102

#1020

#1023

#1024

#1025

#1026

#1027

#1028

#103

#1030

#1031

#1032

#1033

#1034

#1035

#1037

#1038

#1039

#1040

#1041

#1042

#1043

#1045

#1046

#1047

#1049

#105

#1050

#1052

#1053

#1054

#1055

#1056

#1058

#1060

#1061

#1062

#1063

#1065

#1068

#1068

#1069

#107

#1070

#1071

#1072

#1073

#1074

#1076

#1076

#1077

#1077

#1078

#1079

#108

#1080

#1083

#1084

#1084

#1085

#1085

#1086

#1087

#1088

#1089

#109

#1090

#1091

#1092

#1093

#1094

#1095

#1096

#1097

#1098

#1099

#11

#110

#1100

#1101

#1102

#1103

#1104

#1105

#1106

#1107

#1108

#1109

#111

#1111

#1112

#1113

#1114

#1115

#1116

#1118

#1120

#1121

#1122

#1122

#1123

#1124

#1125

#1125

#1126

#1126

#1127

#1128

#1128

#1129

#1130

#1131

#1132

#1133

#1134

#1136

#1137

#1138

#1139

#114

#1140

#1141

#1142

#1143

#1144

#1145

#1146

#1147

#1148

#1149

#115

#1150

#1151

#1153

#1154

#1157

#116

#1160

#1161

#1162

#1163

#1164

#1165

#1166

#1167

#1168

#1169

#117

#1170

#1171

#1172

#1173

#1174

#1175

#1175

#1176

#1177

#1178

#1179

#1180

#1180

#1181

#1182

#1183

#1184

#1185

#1187

#1188

#1189

#1190

#1191

#1192

#1193

#1194

#1195

#1196

#1198

#1199

#12

#120

#1201

#1202

#1203

#1204

#1206

#1207

#1208

#1209

#121

#1210

#1211

#1212

#1214

#1216

#1217

#1218

#1219

#122

#1220

#1221

#1222

#1223

#1225

#1226

#1227

#1228

#123

#1230

#1230

#1231

#1233

#1234

#1234

#1235

#1236

#1237

#1237

#1238

#1238

#1239

#124

#1240

#1241

#1242

#1243

#1244

#1245

#1247

#1248

#1249

#125

#1250

#1251

#1251

#1252

#1252

#1253

#1253

#1254

#1256

#1257

#1258

#1259

#1260

#1261

#1261

#1262

#1263

#1264

#1264

#1265

#1266

#1267

#1268

#1269

#1270

#1271

#1273

#1274

#1275

#1276

#1276

#1277

#1278

#1279

#128

#1280

#1282

#1283

#1284

#1284

#1285

#1286

#1286

#1287

#1288

#1288

#1289

#1289

#1290

#1291

#1292

#1292

#1293

#1293

#1294

#1294

#1297

#1297

#1298

#1299

#13

#130

#1300

#1300

#1301

#1304

#1305

#1306

#1307

#1309

#1310

#1311

#1311

#1312

#1312

#1313

#1314

#1316

#1317

#1317

#1318

#1319

#1319

#132

#1320

#1320

#1321

#1322

#1324

#1324

#1325

#1325

#1326

#1328

#1328

#1329

#1331

#1331

#1332

#1333

#1334

#1336

#1336

#1337

#1338

#1338

#1339

#134

#1340

#1341

#1342

#1343

#1344

#1345

#1345

#1346

#1348

#1349

#135

#1350

#1351

#1351

#1352

#1352

#1353

#1353

#1354

#1355

#1356

#1357

#1358

#1359

#136

#1360

#1360

#1361

#1362

#1363

#1363

#1364

#1365

#1366

#1367

#1368

#1369

#1370

#1371

#1372

#1372

#1374

#1376

#1377

#1379

#1379

#1380

#1381

#1382

#1384

#1385

#1386

#1387

#1387

#1388

#1389

#1390

#1391

#1391

#1392

#1394

#1395

#1396

#1396

#1397

#1398

#1399

#14

#140

#1400

#1400

#1401

#1401

#1402

#1402

#1403

#1403

#1404

#1405

#1405

#1406

#1407

#1408

#1409

#141

#1410

#1411

#1412

#1413

#1414

#1415

#1416

#1417

#1418

#1419

#142

#1420

#1421

#1422

#1423

#1424

#1425

#1425

#1426

#1427

#1428

#1429

#1429

#143

#1430

#1431

#1433

#1434

#1435

#1436

#1437

#1438

#1439

#1439

#144

#1440

#1441

#1442

#1442

#1443

#1444

#1445

#1446

#1447

#1447

#1448

#1449

#1449

#145

#1450

#1451

#1452

#1454

#1455

#1455

#1456

#1456

#1457

#1458

#1459

#1459

#1460

#1460

#1461

#1462

#1463

#1464

#1465

#1466

#1467

#1468

#1469

#1470

#1471

#1472

#1473

#1474

#1475

#1476

#1477

#1478

#1479

#148

#1480

#1480

#1482

#1484

#1486

#1487

#1487

#1488

#1489

#149

#1490

#1491

#1492

#1493

#1494

#1495

#1497

#1498

#1499

#15

#150

#1500

#1501

#1502

#1503

#1504

#1505

#1506

#1507

#1509

#1510

#1511

#1513

#1514

#1515

#1516

#1517

#1517

#1518

#1518

#1519

#1519

#1520

#1520

#1521

#1521

#1522

#1523

#1523

#1524

#1524

#1525

#1525

#1527

#1527

#1528

#1528

#153

#154

#155

#157

#158

#159

#16

#161

#162

#163

#164

#166

#167

#168

#169

#17

#170

#171

#172

#173

#175

#178

#18

#180

#181

#182

#183

#185

#186

#187

#19

#190

#191

#193

#194

#195

#196

#198

#199

#2

#20

#200

#201

#202

#203

#207

#209

#21

#210

#212

#213

#214

#216

#217

#218

#219

#22

#220

#221

#222

#223

#224

#225

#226

#228

#229

#23

#230

#232

#234

#237

#238

#239

#24

#240

#244

#246

#247

#25

#251

#253

#255

#256

#26

#260

#261

#262

#263

#264

#265

#267

#269

#27

#270

#273

#274

#275

#276

#28

#280

#282

#283

#284

#286

#287

#288

#289

#29

#290

#291

#292

#293

#294

#295

#296

#297

#298

#299

#3

#30

#300

#302

#304

#306

#307

#308

#31

#310

#316

#317

#318

#32

#320

#321

#322

#324

#325

#326

#327

#327

#33

#330

#331

#333

#336

#339

#34

#340

#344

#345

#346

#347

#348

#349

#35

#350

#352

#353

#354

#356

#357

#358

#359

#36

#360

#361

#362

#363

#364

#365

#366

#367

#37

#372

#373

#374

#375

#376

#377

#378

#379

#38

#380

#381

#382

#384

#385

#386

#387

#388

#389

#39

#392

#393

#394

#395

#395

#396

#397

#399

#4

#40

#400

#402

#403

#403

#405

#406

#407

#408

#41

#410

#412

#413

#415

#416

#417

#419

#42

#420

#422

#423

#424

#425

#427

#428

#429

#43

#430

#431

#432

#433

#434

#435

#436

#437

#438

#44

#440

#442

#443

#444

#446

#448

#449

#45

#450

#452

#453

#454

#455

#456

#457

#459

#46

#460

#462

#463

#464

#465

#466

#467

#468

#468

#469

#47

#470

#471

#472

#473

#475

#477

#479

#481

#482

#483

#484

#485

#486

#489

#49

#490

#491

#492

#493

#494

#495

#496

#497

#498

#499

#5

#50

#502

#503

#504

#506

#508

#510

#511

#516

#519

#52

#521

#522

#523

#524

#526

#529

#53

#530

#531

#532

#534

#535

#536

#539

#54

#540

#541

#546

#547

#549

#55

#550

#551

#552

#553

#554

#556

#558

#559

#560

#561

#562

#563

#564

#566

#567

#570

#571

#572

#574

#575

#576

#577

#578

#579

#58

#580

#581

#582

#583

#583

#585

#587

#588

#589

#59

#590

#591

#592

#595

#596

#597

#598

#599

#6

#60

#602

#605

#606

#608

#609

#610

#611

#612

#613

#614

#616

#617

#618

#619

#621

#627

#628

#629

#630

#632

#633

#634

#635

#639

#643

#645

#646

#647

#648

#649

#651

#652

#653

#654

#655

#658

#659

#660

#661

#662

#663

#665

#666

#667

#668

#669

#672

#673

#674

#676

#68

#680

#683

#684

#685

#686

#687

#689

#69

#691

#692

#693

#695

#696

#697

#698

#699

#7

#700

#702

#704

#706

#707

#708

#710

#711

#712

#714

#716

#717

#718

#719

#720

#721

#721

#722

#724

#725

#726

#727

#728

#730

#731

#732

#735

#736

#737

#738

#740

#741

#743

#744

#745

#747

#748

#749

#75

#751

#752

#753

#754

#755

#756

#758

#759

#76

#760

#762

#763

#765

#77

#770

#771

#772

#773

#774

#775

#778

#779

#78

#780

#782

#783

#784

#785

#787

#788

#79

#790

#791

#792

#794

#795

#797

#798

#799

#8

#80

#800

#804

#806

#807

#808

#809

#811

#812

#815

#816

#821

#822

#823

#824

#825

#828

#829

#830

#831

#832

#833

#834

#835

#836

#837

#838

#839

#84

#840

#841

#842

#843

#845

#846

#847

#849

#85

#850

#850

#851

#853

#856

#858

#859

#86

#860

#861

#862

#863

#864

#867

#869

#870

#874

#876

#877

#878

#879

#88

#881

#882

#884

#886

#887

#888

#889

#890

#892

#894

#895

#898

#899

#9

#90

#900

#901

#903

#904

#906

#907

#908

#909

#91

#910

#911

#912

#913

#914

#917

#918

#919

#92

#921

#922

#924

#926

#927

#928

#928

#929

#93

#931

#932

#933

#934

#935

#938

#939

#94

#940

#945

#946

#947

#948

#949

#95

#950

#952

#953

#955

#957

#961

#962

#963

#966

#967

#970

#971

#973

#974

#976

#977

#980

#981

#982

#983

#984

#985

#986

#987

#988

#989

#990

#991

#992

#993

#994

#995

#996

#997

#998

#999

v0.0.1

v0.1.0

v0.1.1

v0.2.0

v0.2.1

v0.3.0

v0.3.1

v0.4.0

v0.4.1

v0.4.2

v0.5.0

v0.5.1

v0.5.10

v0.5.2

v0.5.3

v0.5.4

v0.5.5

v0.5.6

v0.5.7

v0.5.8

v0.5.9

v0.6.0

v0.6.1

v0.6.2

v0.6.3

v0.6.4

v0.6.5

v0.7.0

v0.8.0

v0.8.1

v0.8.2

v0.8.3

v0.8.4

Mono Color

• 23ba1065a7 Update README.md (#1522) main Vaibhav Jindal 2026-10-11 17:26:02 +00:00
• d5f281789a fix(test): stop passing removed device kwarg to Llama4 rotary embed… (#1516) Vaibhav JindalandCopilot 2026-10-07 23:20:35 +00:00
• 98f6371a9c ci: run only Checkstyle in the merge queue; make AMD/Intel manual-only (#1515) Vaibhav JindalandCopilot 2026-10-07 21:54:52 +00:00
• 74777f44fb fix(test): drop unsupported device kwarg from config-based RoPE con… (#1514) Vaibhav JindalandCopilot 2026-10-07 20:52:57 +00:00
• b5cdbf7e34 perf(norm): skip dW/dB calculation when parameters are frozen (#1067) (#1440) 4 people 2026-10-07 02:49:41 +00:00
• 590a608574 fix(monkey_patch): let the norm flags cover the whole vision tower (#1443) 3 people 2026-10-07 00:47:52 +00:00
• fc330e197e perf(distillation): skip the hard (CE) loss when it is neither weighted nor returned (#1489) 3 people 2026-10-07 00:11:39 +00:00
• 280444a998 fix(test): use even head_dim in rope functional-correctness weird shape (#1513) Vaibhav JindalandCopilot 2026-10-06 22:03:57 +00:00
• 2024b7d83b test(cutedsl): gap-op dispatcher tests + fused_add_rms_norm 3-D backward + shape fallbacks (#1457) Arup DeandCopilot 2026-10-06 20:22:43 +00:00
• 29eb8ca223 Fix Falcon-H1 output scaling in LCE forward (#1477) cherry77-cloudandArup De 2026-10-06 18:39:46 +00:00
• c508f2b4aa fix(grpo): respect independent token and mask strides (#1461) LOGO127andArup De 2026-10-06 18:39:33 +00:00
• e2039eb8c7 fix(cutedsl): fuse Gemma RMSNorm weight offset in FP32 (#1462) luca huangandArup De 2026-10-06 03:56:30 +00:00
• 80fb156fce fix(monkey_patch): patch Q/K norms on existing Qwen models (#1470) luca huangandArup De 2026-10-06 03:48:09 +00:00
• 9384e4096f bench(flce): add configurable chunk-memory sweep (#1479) Fuxiang DuandArup De 2026-10-06 03:39:48 +00:00
• ae47958e39 fix: respect gradient column strides in fused linear CE (#1500) 3 people 2026-10-06 02:59:39 +00:00
• 395d6166e8 fix(kl_div, tvd): make torch.compile work (NameError, then fp64 loop-carried scalar) (#1509) 3 people 2026-10-06 02:34:33 +00:00
• 57d5ff4cdd Restore benchmarks after gh-deploy gh-pages github-actions[bot] 2026-10-05 23:46:22 +00:00
• 067b31c4de Deployed 5b0f210 with MkDocs version: 1.6.1 github-actions[bot] 2026-10-05 23:46:20 +00:00
• 5b0f210cc4 [Megatron] Rebind CE call sites so apply_liger_kernel_to_megatron(cross_entropy=True) takes effect (#1486) LazuriteandVaibhav Jindal 2026-10-05 23:45:07 +00:00
• 488acbd9c0 Fix llama4 class-level SwiGLU patch for configs with dense layers (#1507) EdsonandVaibhav Jindal 2026-10-05 21:26:33 +00:00
• 77624a9aa7 Fix qwen2_vl / qwen2_5_vl class-level RMSNorm patch on transformers >= 5.2 (#1511) EdsonandVaibhav Jindal 2026-10-05 21:21:28 +00:00
• 8e698102ae fix(glm4v): version-aware hidden_size/vocab_size for Glm4vConfig (#1214) Stefan WangandVaibhav Jindal 2026-10-05 20:59:16 +00:00
• e3e6daa889 Fix sequence-level vLLM correction in VESPO loss (#1493) cherry77-cloudandVaibhav Jindal 2026-10-05 19:19:01 +00:00
• 8343b3cf7b [Feature] Support apply_liger_kernel_to_deepseek_v2 (#1501) Yulong CaoandVaibhav Jindal 2026-10-05 18:54:58 +00:00
• 7d88a77860 fix(grpo): correct sequence-level KL and masked loss gradients (#1491) cherry77-cloud 2026-10-05 18:50:33 +00:00
• d2965de5de fix(ppo): warn instead of raising when ref_per_token_logps and ref_input are both passed (#1506) Alexandre KohlerandVaibhav Jindal 2026-10-05 18:49:29 +00:00
• 91e51dfd9c Allow manual publishing of Liger Kernel releases (#1504) 3 people 2026-10-04 21:14:13 +00:00
• b297821787 feat(ops): support partial RoPE in Triton kernel (#1451) Piyush UmateandJari Kolehmainen 2026-10-02 18:18:49 +00:00
• 492bfeb791 Fix CUDA 13 Blackwell MoE with uniform 1SM UMMA (#1505) 3 people 2026-10-01 17:01:18 +00:00
• 43293d84d7 Bump shared Liger and LCK release version to 0.8.4 (#1503) v0.8.4 3 people 2026-09-30 20:22:07 +00:00
• 8866e4ee79 Warn and skip native configuration on unsupported devices (#1502) 3 people 2026-09-30 18:16:02 +00:00
• 0043f43309 Add shared LCK configure API for FLSCE and MoE (#1499) 3 people 2026-09-30 16:15:42 +00:00
• 6ad077c363 Cast tl.program_id to int64 in llama4_rope and qwen2vl_mrope (#1342) 3 people 2026-09-28 07:01:06 +00:00
• ed1f856d5b fix: request float32 logits from the fused JSD projection GEMM (#1433) 3 people 2026-09-28 06:41:19 +00:00
• 5b828eef2b Fix JSD underflow for low-probability vocabulary entries (#1397) Michaël DefferrardandVaibhav Jindal 2026-09-28 06:35:32 +00:00
• b5bcace9b1 Fix group norm backward for inputs with more than three dimensions (#1376) Truong VuandVaibhav Jindal 2026-09-28 06:29:41 +00:00
• 559d33e7b1 Fix TVD gradient at p == q (#1374) Truong VuandVaibhav Jindal 2026-09-28 06:27:51 +00:00
• 39adfeefab Fix AttnRes silently dropping blocks past 32 (#1344) Aditya Singh 2026-09-28 06:24:50 +00:00
• f01ffb1f5d docs: fix input/target shapes in DPO and ORPO forward docstrings (#1406) Aryan DhawanandVaibhav Jindal 2026-09-28 06:20:53 +00:00
• 6b2b09ce4f Add requires-python >=3.10 to pyproject.toml (#1332) 3 people 2026-09-28 06:18:03 +00:00
• 2f53aed67f Restore benchmarks after gh-deploy github-actions[bot] 2026-09-28 06:16:18 +00:00
• b53a1a98cf Deployed c8c9999 with MkDocs version: 1.6.1 github-actions[bot] 2026-09-28 06:16:16 +00:00
• c8c999932f docs: sync README triton floor with setup.py and fix contributing install command (#1301) Kobi HikriandVaibhav Jindal 2026-09-28 06:15:04 +00:00
• 0c9e8ad056 Preserve FP32 LayerNorm statistics for backward (#1498) cherry77-cloud 2026-09-27 23:37:47 +00:00
• e064108ef6 Add fused CE + TV distance kernel (#1384) 3 people 2026-09-27 10:53:57 +00:00
• 72e9159d86 Load unsupported model types instead of raising KeyError in AutoLigerKernelForCausalLM (#1404) Nilesh PatilandVaibhav Jindal 2026-09-27 09:48:48 +00:00
• 998ac696fa fix(llama4): honour skip_logits in lce_forward (#1445) 3 people 2026-09-27 09:47:20 +00:00
• 946a3693c1 Fix softcap padding in cross entropy (#1490) cherry77-cloudandVaibhav Jindal 2026-09-27 09:10:52 +00:00
• 70fe0333fe fix(triton): widen softmax/sparsemax row index to int64 (#1497) znzrybandVaibhav Jindal 2026-09-27 08:44:24 +00:00
• d76315bd77 fix(kto): sum chosen/rejected rewards per sample when chunk_size > 1 (#1495) Alexandre Kohler 2026-09-27 08:33:53 +00:00
• be2a4eb14c Add LCK CUDA-selectable NVSHMEM extras; retain CUDA 12.9 releases (#1492) 3 people 2026-09-25 16:27:26 +00:00
• 40a9d8a61b [Feature] Support apply_liger_kernel_to_deepseek_v3 (#1424) Yulong Cao 2026-09-25 02:26:05 +00:00
• 55bc58a842 fix(triton): widen blocked RMSNorm row index (#1482) 3 people 2026-09-24 23:50:30 +00:00
• 9358939c4d Fix ignored-token masking in cosine distillation (#1478) cherry77-cloudandVaibhav Jindal 2026-09-24 21:47:51 +00:00
• aed0dfbb58 refactor(kl_div): route fused-linear KL inner kernel through the dispatcher (#1476) Yulong CaoandJari Kolehmainen 2026-09-24 21:43:17 +00:00
• ec10acbbe6 perf(ppo): skip dW/dB calculation when lm_head parameters are frozen (#1488) Daniel LentonandJari Kolehmainen 2026-09-24 21:36:46 +00:00
• fbdea0ff3b fix(npu): Replace Ascend A5 comments with 950 (#1484) Rain 2026-09-24 18:21:47 +00:00
• 09f7febf1a fix(npu): Ascend A3/A5 UB tile cap and A5 workarounds for group norm, GRPO, fused MoE (#1464) Rain 2026-09-21 13:29:41 +00:00
• c5b9f31d3e Fix CUTEDSL FLCE functional wrapper arguments (#1475) 3 people 2026-09-18 19:47:57 +00:00
• 19d92e7309 Publish LCK with PyPI Trusted Publishing (#1474) 3 people 2026-09-17 01:25:14 +00:00
• 3899871085 Use a dedicated PyPI credential for LCK (#1473) 3 people 2026-09-16 22:36:26 +00:00
• bad6c3ec42 Keep Triton FLCE C=1 by default and use C=8 for GPT-OSS (#1472) v0.8.3 3 people 2026-09-16 18:36:55 +00:00
• d5df610cbf Bump Liger packages to 0.8.3 (#1471) 3 people 2026-09-16 16:17:49 +00:00
• cb6462b60d Refresh LigerCute README figures and results (#1469) 3 people 2026-09-16 00:03:17 +00:00
• b53fc76341 Fix remaining LCK container build dependencies (#1467) 3 people 2026-09-16 00:01:59 +00:00
• de77856a01 perf(flce): reuse logits buffers and write input gradients directly (#1468) Vaibhav JindalandCopilot 2026-09-15 23:50:07 +00:00
• dc1512630d Fix Python setup in LCK container builds (#1466) 3 people 2026-09-15 22:34:54 +00:00
• d72addc575 Publish multi-architecture Liger Cute wheel on releases (#1465) 3 people 2026-09-15 22:25:19 +00:00
• f2ac965f61 Add SM100 tensor-parallel fused scaled cross entropy (#1458) 3 people 2026-09-14 22:18:34 +00:00
• ebc9afba0d Merge branch 'main' into vaibhav/cutile-flce-chunking-triton-policy vaibhav/cutile-flce-chunking-triton-policy Vaibhav Jindal 2026-09-14 13:48:13 -07:00
• 5786fd4c76 Specialize SM90 MoE communication and fix backward synchronization (#1463) 3 people 2026-09-14 16:16:11 +00:00
• bcb5962dc0 [NPU] Fix Ascend JSD UT failures and register LayerNorm dispatch (#1452) sunyi0505 2026-09-14 12:57:46 +00:00
• a415b179d6 Restore benchmarks after gh-deploy github-actions[bot] 2026-09-14 02:35:56 +00:00
• 651283adc8 Deployed f6385e9 with MkDocs version: 1.6.1 github-actions[bot] 2026-09-14 02:35:54 +00:00
• f6385e909b feat: add FusedLinearKLDivLoss (fused linear + KL divergence for distillation) (#1423) Yulong Cao 2026-09-14 02:34:33 +00:00
• c64878355b Add DTensor support for Ascend SwiGLU (#1448) xvxuopop 2026-09-13 16:47:03 +00:00
• 5f1f7b9d2e refactor(cutile): match Triton FLCE allocation baseline Vaibhav JindalandCopilot 2026-09-10 23:44:00 +00:00
• c426a53a53 fix(cutile): align FLCE accumulation policy with Triton Vaibhav JindalandCopilot 2026-09-10 22:44:42 +00:00
• c600ff5131 test(cutile): use architecture-neutral FLCE test filename Vaibhav JindalandCopilot 2026-09-10 20:53:45 +00:00
• ecd0780285 feat(cutile): chunk FLCE with Triton sizing and cuBLAS GEMMs Vaibhav JindalandCopilot 2026-09-10 19:29:26 +00:00
• 95b01e9402 perf(flce): accumulate low-precision Triton dW with addmm (#1454) Vaibhav JindalandCopilot 2026-09-10 18:30:11 +00:00
• 18692d4f9c feat(cutile): chunk FLCE with cuBLAS gradient GEMMs vaibhav/flce-3-cutile-cublas Vaibhav JindalandCopilot 2026-09-10 08:17:00 +00:00
• c7908aa662 feat(flce): expose explicit token chunk sizes vaibhav/flce-2-chunk-size Vaibhav JindalandCopilot 2026-09-10 07:58:46 +00:00
• d41fe63f1f perf(flce): accumulate low-precision Triton dW with addmm Vaibhav JindalandCopilot 2026-09-10 07:30:00 +00:00
• 3eddb174e0 perf(flce): accumulate low-precision Triton dW with addmm vaibhav/cutile-flce-chunking-b200 Vaibhav JindalandCopilot 2026-09-10 06:36:46 +00:00
• 96233e3d9b feat(cutile): cuTile backends for RMSNorm, LayerNorm, Softmax, JSD, FusedLinearJSD (#1420) Arup DeandCopilot 2026-09-10 02:57:03 +00:00
• e4aad6c0c0 Add chunked cuTile FLCE and matched backend comparison Vaibhav JindalandCopilot 2026-09-10 01:00:09 +00:00
• d2b5474de7 Add SM100 fused linear SwiGLU (#1340) 4 people 2026-09-09 23:11:17 +00:00
• 1aac00f870 feat(cutedsl): fused SwiGLU reads stacked/separate weights without pre-packing pr1340 Vaibhav JindalandCopilot 2026-09-09 23:03:08 +00:00
• ad2436e80e fix(cutedsl): use the current stream for RMSNorm backward (#1450) luca-888 2026-09-09 15:55:13 +00:00
• 8c4d822408 [NPU] Optimize Ascend CE/FLCE, RMSNorm, and RoPE kernels (#1426) sunyi0505 2026-09-09 12:17:05 +00:00
• 77206f839d perf(preference): skip dW/dB calculation when parameters are frozen (#1441) Piyush UmateandVaibhav Jindal 2026-09-09 02:03:32 +00:00
• 2b4f1a413e fix(monkey_patch): report the Liger name on patched gemma3/gemma4/exaone4 MLPs (#1444) David 2026-09-08 22:08:49 +00:00
• 64594266bb feat(cutedsl): CuTeDSL backends for CrossEntropy, FusedLinearCE, FusedLinearJSD (#1419) Arup DeandCopilot 2026-09-03 15:55:23 +00:00
• 1297f4729c Add SM100 fused linear SwiGLU (rebased onto main) Stella DeyandCopilot 2026-09-02 21:50:58 +00:00
• 0a5e9a55f8 Update tests.py to use B200 instead of B200! (#1437) Vaibhav Jindal 2026-09-02 17:45:50 +00:00
• 71293bebcc fix(mlp): correct SwiGLU backward_dI on Blackwell (SM100/B200) (#1435) 3 people 2026-09-02 00:37:28 +00:00
• 1621f42d7e ci(nvi): add cuTile tests on B200 to the daily run (#1436) 3 people 2026-09-01 23:00:26 +00:00
• 0245181a11 Use native CUTLASS + NVSHMEM package terminology (#1434) 3 people 2026-09-01 21:54:38 +00:00
• cc7f68fc9c [CUDA] Add tensor-parallel fused scaled cross entropy frontend (#1428) 3 people 2026-09-01 21:22:46 +00:00