Commit Graph

Select branches

Hide Pull Requests

4bit

cerebras-wip

ci-test

copilot/fix-cpu-backend-deadlock

copilot/investigate-failing-tests

experiments/blocking

feature/parallel-decoding-gpt2-example

fix-mul-mat

gg/ci-paths

gg/cmake-rm-unused

gg/cuda-assert-mul-mat-pad

gg/files-remove-old

gg/ggml-restore-api

gg/gguf-spec-diagram

gg/vulkan-apple-crash-repro

gg/yarn-tests

ggml-backend-metal

ggml-cpp

ggml-dev-v0.9.3

ggml-rc-v0.27.0

ggml-rc-v0.9.1

ggml-rc-v0.9.3

ggml-rc-v0.9.4

gpt-2-opt

gq

llama

master

mul-mat-id-batch

release

sync-llama.cpp-24-12-12-0

sync-llama.cpp-25-03-27-2

sync-llama.cpp-25-03-31

sync-llama.cpp-25-05-13

sync-llama.cpp-25-05-19

sync-llama.cpp-25-06-01

sync-llama.cpp-25-07-28

sync-llama.cpp-25-10-11

sync-llama.cpp-26-07-17

sync-llama.cpp-26-08-21

sync-llama.cpp-26-09-09

sync-whisper-25-04-25

sync-whisper-25-04-3

sync-whisper.cpp-25-05-01

sync-whisper.cpp-25-05-20

sync-whisper.cpp-25-06-01

sync-whisper.cpp-25-07-02

sync-whisper.cpp-25-07-12

t5

test

#100

#1003

#1004

#1006

#1007

#1009

#101

#1011

#1013

#1015

#1016

#1017

#1018

#1019

#102

#1020

#1022

#1023

#1024

#1026

#1028

#1029

#1030

#1032

#1033

#1033/merge

#1034

#1037

#1039

#104

#1040

#1043

#1045

#1046

#1047

#1048

#1051

#1052

#1053

#1054

#1056

#1056/merge

#1059

#1060

#1065

#107

#1070

#1073

#1078

#108

#1081

#1084

#1089

#1095

#1095/merge

#1096

#1097

#1098

#1099

#1099/merge

#1101

#1101/merge

#1103

#1105

#1105/merge

#1106

#1107

#111

#1110

#1110/merge

#1112

#1113

#1114

#1115

#1116

#1118

#112

#1121

#1122

#1123

#1125

#1126

#1127

#1128

#1129

#113

#1130

#1132

#1133

#1134

#1135

#1137

#1138

#1139

#114

#1141

#1142

#1144

#1146

#1152

#1157

#1158

#1159

#1159/merge

#116

#1160

#1161

#1162

#1163

#1164

#1165

#1166

#1167

#1168

#1169

#117

#1170

#1173

#1174

#1175

#1176

#1177

#1178

#1179

#1180

#1181

#1182

#1183

#1184

#1185

#1187

#1188

#1189

#1190

#1191

#1192

#1193

#1197

#1197/merge

#1198

#1199

#12

#1200

#1201

#1203

#1204

#1205

#1206

#1207

#1208

#1209

#1213

#1214

#1217

#1221

#1222

#1226

#1227

#1228

#1229

#1231

#1232

#1233

#1237

#1238

#1239

#1240

#1241

#1242

#1243

#1243/merge

#1246

#1247

#1248

#1249

#1250

#1251

#1252

#1253

#1255

#1256

#1257

#1258

#1259

#1260

#1263

#1265

#1266

#1267

#1268

#1269

#127

#1270

#1271

#1272

#1273

#1274

#1276

#1278

#1279

#1280

#1281

#1282

#1283

#1284

#1285

#1286

#1289

#1290

#1291

#1292

#1293

#1294

#1295

#1297

#1299

#130

#1300

#1301

#1304

#1307

#1309

#1310

#1311

#1312

#1313

#1315

#1316

#1317

#1318

#1319

#132

#1320

#1322

#1323

#1327

#1329

#1330

#1331

#1332

#1334

#1336

#1338

#1339

#134

#1340

#1341

#1342

#1344

#1345

#1346

#1347

#1348

#1349

#1350

#1351

#1352

#1353

#1354

#1355

#1356

#1357

#1358

#1359

#1360

#1361

#1362

#1363

#1364

#1368

#1370

#1372

#1374

#1376

#1377

#1379

#1380

#1382

#1383

#1384

#1384/merge

#1385

#1386

#1386/merge

#1387

#139

#1390

#1390/merge

#1391

#1392

#1392/merge

#1394

#1395

#1396

#1397

#1399

#1400

#1402

#1403

#1408

#1409

#1410

#1414

#1415

#1417

#1418

#1419

#142

#1422

#1423

#1424

#1425

#1426

#1428

#1429

#143

#1432

#1434

#1435

#1435/merge

#1439

#1440

#1441

#1442

#1444

#1446

#1449

#145

#1451

#1453

#1454

#1455

#1456

#1457

#1457/merge

#1459

#146

#1461

#1461/merge

#1462

#1463

#1464

#1465

#1465/merge

#1468

#1469

#1471

#1472

#1473

#1474

#1475

#1475/merge

#1476

#1477

#1478

#1479

#1480

#1483

#1484

#1485

#1487

#1488

#1489

#1489/merge

#1490

#1492

#1493

#1494

#1496

#15

#1500

#1502

#1504

#1507

#1508

#1509

#151

#1510

#1512

#1512/merge

#1515

#1516

#1517

#1518

#1518/merge

#1519

#152

#1520

#1521

#1522

#1523

#1526

#1527

#1528

#153

#1531

#1533

#1534

#1536

#1538

#1539

#154

#1540

#1541

#1544

#1547

#1548

#1549

#155

#1550

#1551

#1556

#1558

#1559

#1560

#1561

#1561/merge

#1562

#1562/merge

#1563

#1564

#1567

#1568

#1569

#1570

#1571

#1571/merge

#1572

#1573

#1574

#1574/merge

#1575

#1576

#1577

#1578

#1580

#1581

#1582

#1583

#1584

#1585

#1585/merge

#1586

#1587

#1588

#1589

#159

#1590

#1590/merge

#1591

#1592

#1593

#1594

#1595

#1596

#1597

#1598

#1599

#160

#1600

#1601

#1601/merge

#1602

#1602/merge

#1603

#1603/merge

#1604

#1604/merge

#1606

#1607

#1610

#1612

#1612/merge

#1613

#1613/merge

#1614

#1614/merge

#1616

#1616/merge

#1617

#1618

#1619

#1619/merge

#162

#1620

#1620/merge

#1622

#1623

#1623/merge

#1625

#1625/merge

#1626

#1627

#1631

#1631/merge

#1632

#1633

#1634

#1635

#1636

#1637

#1638

#1639

#164

#1640

#1640/merge

#1641

#1642

#1644

#1645

#1646

#1647

#1647/merge

#165

#1651

#1652

#1653

#1654

#1656

#1656/merge

#1657

#1659

#1665

#1666

#167

#1671

#1671/merge

#1672

#1672/merge

#1673

#1673/merge

#1675

#1676

#173

#174

#175

#176

#177

#179

#181

#184

#186

#190

#191

#194

#197

#198

#199

#201

#203

#204

#206

#207

#212

#213

#218

#219

#221

#222

#224

#228

#229

#231

#234

#237

#239

#244

#246

#247

#252

#254

#255

#256

#26

#261

#262

#263

#264

#267

#269

#27

#270

#274

#275

#277

#279

#296

#298

#300

#301

#302

#305

#307

#309

#311

#313

#314

#315

#316

#317

#319

#321

#323

#325

#326

#328

#333

#335

#336

#337

#338

#339

#340

#342

#343

#345

#346

#349

#35

#350

#351

#352

#354

#358

#359

#363

#364

#365

#367

#368

#370

#372

#373

#374

#375

#377

#38

#380

#383

#384

#386

#389

#39

#391

#392

#393

#394

#395

#398

#40

#400

#401

#402

#403

#405

#407

#409

#410

#411

#414

#415

#416

#418

#422

#424

#426

#428

#433

#435

#438

#439

#442

#443

#447

#449

#450

#457

#458

#46

#461

#462

#464

#466

#47

#470

#476

#477

#479

#481

#482

#483

#485

#486

#487

#489

#490

#491

#493

#494

#495

#496

#498

#504

#505

#506

#508

#509

#51

#511

#513

#514

#521

#522

#523

#524

#525

#527

#528

#531

#543

#547

#548

#550

#551

#552

#553

#555

#557

#558

#560

#562

#564

#57

#570

#570/merge

#571

#572

#573

#574

#575

#576

#577

#58

#580

#580/merge

#581

#581/merge

#582

#583

#585

#586

#587

#591

#592

#595

#596

#597

#598

#599

#601

#602

#606

#607

#608

#611

#613

#616

#619

#62

#620

#621

#625

#626

#627

#629

#631

#632

#634

#636

#638

#639

#640

#641

#643

#645

#646

#649

#650

#654

#657

#657/merge

#658

#659

#661

#661/merge

#662

#664

#665

#667

#669

#671

#673

#674

#676

#677

#678

#679

#681

#682

#683

#685

#686

#687

#688

#689

#69

#690

#691

#692

#693

#694

#696

#699

#70

#701

#702

#706

#708

#709

#712

#713

#715

#719

#725

#726

#727

#728

#729

#73

#730

#731

#736

#737

#74

#741

#742

#743

#746

#747

#748

#75

#750

#753

#761

#763

#764

#765

#768

#77

#770

#773

#775

#777

#78

#780

#781

#782

#783

#784

#785

#789

#792

#798

#801

#801/merge

#804

#805

#806

#808

#813

#813/merge

#814

#817

#818

#821

#822

#823

#826

#827

#829

#831

#831/merge

#837

#838

#839

#84

#840

#842

#843

#844

#847

#849

#850

#853

#854

#855

#855/merge

#856

#857

#858

#859

#860

#864

#864/merge

#865

#867

#867/merge

#870

#874

#874/merge

#876

#879

#880

#884

#885

#890

#893

#895

#896

#897

#898

#899

#900

#901

#902

#904

#905

#906

#908

#909

#91

#912

#913

#914

#917

#918

#919

#920

#921

#921/merge

#922

#924

#927

#928

#929

#932

#933

#934

#936

#937

#939

#940

#940/merge

#943

#946

#947

#948

#949

#95

#951

#952

#955

#96

#960

#961

#963

#965

#966

#967

#969

#971

#972

#973

#974

#976

#978

#979

#98

#980

#981

#982

#983

#984

#987

#988

#992

#993

#994

#996

#997

#998

#999

v0.10.0

v0.10.1

v0.10.2

v0.11.0

v0.11.1

v0.12.0

v0.12.1

v0.13.0

v0.13.1

v0.14.0

v0.15.0

v0.15.1

v0.15.2

v0.15.3

v0.16.0

v0.17.0

v0.18.0

v0.18.1

v0.19.0

v0.20.0

v0.20.1

v0.20.2

v0.21.0

v0.22.0

v0.23.0

v0.24.0

v0.25.0

v0.25.1

v0.25.2

v0.25.3

v0.26.0

v0.27.0

v0.28.0

v0.9.1

v0.9.10

v0.9.11

v0.9.3

v0.9.4

v0.9.5

v0.9.6

v0.9.7

v0.9.8

v0.9.9

Mono Color

• facdc44a67 ggml : bump version to v.28.0 (#1676) master v0.28.0 Georgi Gerganov 2026-10-28 12:24:54 +02:00
• 23261912c6 sync : llama.cpp Georgi Gerganov 2026-10-28 11:55:32 +02:00
• 5e5175f383 cuda : add top-k MoE fusion for 384 experts (llama/31288) Yonghye Kwon 2026-10-28 17:13:01 +09:00
• 3f866f9b4c vulkan: clamp A prefetch to end_k in coopmat1 int8 matmul (llama/30283) Kris 2026-10-28 09:12:22 +01:00
• 06db896e3b HIP/CUDA: remove pragma unroll from loops in conv2d where itteration count cant be infered at compile time (llama/31255) uvos 2026-10-27 19:42:04 +01:00
• 85dcbfe428 rpc : catch std::length_error in recv_msg (llama/31204) Katara 2026-10-27 14:04:55 +01:00
• d1989dd26f vulkan: fix CI llvmpipe by allowing dim 3 broadcast in MUL_MAT (llama/31239) Pascal 2026-10-27 13:39:03 +01:00
• 0981f4fd9e metal : add GGML_OP_TOP_K_MASK (llama/31230) Georgi Gerganov 2026-10-27 12:13:23 +02:00
• 73cdd6f739 ggml : relax scale to non-contiguous views (llama/31146) Georgi GerganovandPascal 2026-10-27 07:55:31 +02:00
• 0e91cde046 opencl: refine mul mat id -1 index (llama/31170) lhez 2026-10-26 22:52:07 -07:00
• 026a2eb357 opencl: initial support for q3_k (llama/31172) lhezandHongqiang Wang 2026-10-26 22:51:22 -07:00
• a93614358a hexagon: add GGML_HEXAGON_VTCM_SIZE (llama/31221) Ehsan Bateni 2026-10-26 19:56:01 -04:00
• 687b158a82 opencl: write KV-cache views in set_tensor so a state restore is not dropped (llama/30471) Hongqiang Wang 2026-10-26 11:39:22 -07:00
• 8abd879a24 hexagon: MUL_MAT_ID row skipping optimizations and improvements (llama/31169) Max Krasnyansky 2026-10-26 08:21:14 -07:00
• 8284b4b228 ggml-cpu : rewrite concat as one row kernel for all types (llama/31179) Pascal 2026-10-26 15:18:45 +01:00
• 7abd2f2641 sync : llama.cpp Georgi Gerganov 2026-10-28 11:55:11 +02:00
• 865631d020 vulkan: rms norm optimization using subgroup reductions (llama/29882) virajwad 2026-10-28 11:54:41 +02:00
• d1ba9f3619 sync : llama.cpp Georgi Gerganov 2026-10-28 11:53:55 +02:00
• 9e4cb3cca4 sycl: template the plain MMVQ launch triples (llama/31013) Titaniumtown 2026-10-26 07:13:16 -07:00
• b51867281f sycl: template the per-quant concat impls (llama/31008) Titaniumtown 2026-10-26 07:12:56 -07:00
• 93133a7da0 sycl: add set_tensor_2d for faster tensor-split loading (llama/30358) Captain-TrippsandJon Staples 2026-10-26 09:12:25 -05:00
• 6afa37e633 sycl : bound reorder temp memory to 32 MiB (llama/31062) KyleHagy 2026-10-26 07:11:57 -07:00
• 4542ff2e37 ggml-cpu : fix unused variables in arm arch (llama/31180) Sigbjørn Skjæret 2026-10-26 11:38:55 +01:00
• 10bcc3f390 ci : fix race in webgpu memset (llama/31162) Evan Huus 2026-10-26 05:02:12 -04:00
• 2be37268e8 metal : stop overflow from happening in large matrix output offsets (llama/30072) feni6andfeni6 2026-10-26 02:13:39 -06:00
• 1b31c05083 deepseekv4.1: follow-up review fixes (llama/31137) Aman GuptaandGeorgi Gerganov 2026-10-26 15:52:02 +08:00
• e9fc24b928 cuda : use std::string instead of VLA for archName (llama/30463) Adrien Gallouët 2026-10-26 00:08:22 +01:00
• 1899b2ef51 CUDA: fuse FFN gate/up matmuls and GLU in MMQ with interleaved gate/up rows (llama/29948) anujj 2026-10-25 14:54:38 +05:30
• 9450649106 ggml-cuda: fuse qwen MRoPE + set-rows (llama/31025) Aman Gupta 2026-10-25 17:15:45 +08:00
• e15995258e vulkan: Add a larger tile size for conv3d (coopmat2 only) (llama/29920) Jeff Bolz 2026-10-25 04:13:48 -05:00
• 8d226fe927 vulkan : one warp tile per subgroup in the small matmul tiles (llama/30106) Ian McKellar 2026-10-25 02:13:14 -07:00
• 592a906abf hexagon: generic dims collapse/reshape for unary, binary and matmul ops (llama/31071) Max Krasnyansky 2026-10-25 02:11:48 -07:00
• 6706d832cb opencl: add topk and lightning indexer (llama/31026) lhez 2026-10-25 02:11:15 -07:00
• 37d700dbcf opencl: add initial q2_k support (llama/30999) lhezandHongqiang Wang 2026-10-25 02:09:56 -07:00
• 4212b0d650 model: DSV4.1 flash (llama/31086) Aman GuptaandDaniel Han 2026-10-25 09:43:51 +08:00
• 06984d35eb hexagon: KV-head team split for multicore HMX flash attention (llama/31055) Ehsan Bateni 2026-10-24 15:19:32 -04:00
• d22f11563d metal : remove mul_mm_id amax rescaling (llama/30924) Georgi Gerganov 2026-10-24 18:40:19 +03:00
• 45b4b384b6 rpc : add bounds check in rdma_recv (llama/30399) Katara 2026-10-24 16:34:46 +01:00
• 843436ec4d llama : deny -sm tensor for arches with moe amax rescaling (llama/31081) Georgi Gerganov 2026-10-24 13:48:32 +03:00
• 568ba0659e ggml : allow MUL_MAT_ID -1 index to skip the computation (llama/26631) Xuan-Son Nguyenandyomaytk 2026-10-24 12:15:36 +02:00
• edc03add96 vulkan: ggml src1 precision support (llama/30656) Ruben Ortlam 2026-10-24 06:05:05 +02:00
• 8807118508 HIP: add byte permute fallback when __builtin_amdgcn_perm is unavailable (llama/30915) Arseniy Obolenskiy 2026-10-24 06:04:05 +02:00
• 1fd478aa01 hexagon: add SGN op (llama/30346) kurquharandMax Krasnyansky 2026-10-23 14:58:44 -07:00
• b3844edf2b hip: fix bf16 WMMA operand vector type for gfx11/gfx12 (llama/30885) WalterCool 2026-10-23 18:31:28 -03:00
• 597082422d ggm-cuda: Enforce DP4A MMQ tile height multiple of warp size (llama/30837) thelittlefireman 2026-10-23 21:37:20 +02:00
• 752716f5d1 CUDA: fix build after chunked CUDA graphs (#30804) (llama/31053) SreeramandJohannes Gäßler 2026-10-23 11:47:20 -07:00
• 8a6e170a85 qwen4exp : build the QSA selection without the kq_mask (llama/30965) Alexey Kopytko 2026-10-24 02:49:08 +09:00
• 310a27aa75 CUDA: chunk graphs to amortize graph launch (llama/30804) Aman Gupta 2026-10-24 00:26:16 +08:00
• b217b632b3 hexagon: add CEIL op (llama/31006) Max Krasnyansky 2026-10-23 07:46:53 -07:00
• 461c21283c ggml : fix time overflow in ggml_time_us (llama/30969) Kartik Gulia 2026-10-23 18:26:59 +05:30
• b44c81b2ff CANN: missing const qualifiers added for consistency (llama/30933) Raul Torres 2026-10-23 08:34:20 +01:00
• 6c7b935e23 sycl : pass device into mul_mat_uses_onednn (llama/31014) Titaniumtown 2026-10-22 23:37:53 -07:00
• 74ecf131c6 sync : llama.cpp Georgi Gerganov 2026-10-28 11:53:08 +02:00
• 9ce0959031 hexagon: Op Fusion refactoring, optimizations and new fusions like BINARY_NX (back-to-back ADDs, etc) (llama/30843) Max Krasnyansky 2026-10-28 11:52:32 +02:00
• 6ea1993f03 sync : llama.cpp Georgi Gerganov 2026-10-28 11:51:06 +02:00
• c64f2ec5f3 sycl: Replace host waits in fattn-mkl with explicit dependencies (llama/30684) Todd Malsbary 2026-10-22 19:43:12 -07:00
• fdd4e61eb9 sycl: Add graph record and replay (llama/28725) Todd Malsbary 2026-10-22 19:39:52 -07:00
• 4b72143a55 opencl: add depthwise conv2d (llama/30900) kurquharandLi He 2026-10-22 10:54:24 -07:00
• ea93f5061c Multi-column on Intel XMX matrix engines refactored (llama/29864) budiony 2026-10-22 16:39:01 +03:00
• e05d3b3fc0 vulkan: replace std::aligned_alloc with posix_memalign for termux compatibility (llama/30828) Bill Sideris 2026-10-22 15:40:13 +03:00
• 3c02b21527 sycl: fix CONV_2D wrong results for non-contiguous input (llama/30363) Michael Taylor 2026-10-22 05:19:58 -03:00
• f708681635 kleidiai : add Q8_0 SVE dotprod GEMV and SVE i8mm GEMM kernels (llama/30808) Charles Xu 2026-10-22 08:56:21 +02:00
• 393cc292eb sycl: merge the fp16 and fp32 dequant switches (llama/30740) Titaniumtown 2026-10-21 22:30:21 -07:00
• 0e0b6be1f8 webgpu : add Q2_0 support (llama/30914) Chanho Lee 2026-10-22 14:30:04 +09:00
• 5eb05063e1 opencl: add F32-to-I32 ARGMAX (llama/30897) kurquharandLi He 2026-10-21 18:02:50 -07:00
• b81bf009ef llama: create new fused op for QSA mask (llama/30640) Aman GuptaandPascal 2026-10-22 01:02:22 +08:00
• ed8639d2ba sycl: q8 fast dequant (llama/30775) 92es 2026-10-21 10:01:03 -07:00
• db7c146a9a ggml-opencl : add IM2COL_3D support (llama/30895) kurquharandLi He 2026-10-21 07:59:55 -07:00
• c47a5a8a39 sync : llama.cpp Georgi Gerganov 2026-10-28 11:49:51 +02:00
• c15985f407 metal : fix memory leaks and cleanup logic in GPU capture path (llama/30890) Niklas Wenzel 2026-10-28 11:49:29 +02:00
• 1d13c8e70c sync : llama.cpp Georgi Gerganov 2026-10-28 11:49:02 +02:00
• 614decbf1e hexagon: neon-vectorized and threaded repack (tiling) (llama/30901) Max Krasnyanskyandhubezhou 2026-10-21 00:48:39 -07:00
• bbbf288f05 opencl: add dp4a GEMM variants to vector load activation (llama/30753) lhez 2026-10-21 00:48:13 -07:00
• 496eff9d50 sycl : fix fa crash for Query head dimension mismatch (neq0 != HSK / D=2048/4096) (llama/30286) Neo Zhang 2026-10-21 15:46:43 +08:00
• ec7bc58b34 sycl: collapse duplicate need_check launch bodies in mmq (llama/30739) Titaniumtown 2026-10-21 00:46:00 -07:00
• 43ae36d0b7 ggml-cpu : compute f32 gelu_quick and geglu_quick without the fp16 table (llama/30737) bosh 2026-10-21 14:34:37 +07:00
• 83df5951d3 llama : scale moe down projection per expert token (llama/30861) Georgi Gerganov 2026-10-21 09:01:00 +03:00
• c60c0ed4a4 Sycl iq3 crash fix (llama/30604) Anant Shrivastava 2026-10-21 10:34:13 +05:30
• d242ac9e57 opencl: fix rope program build error with newer Clang (llama/30893) Ben Ashbaugh 2026-10-20 17:19:08 -07:00
• 30dffe9d02 hexagon: fix VTCM size query on images that need layout out-params (llama/30655) Tarek Dakhran 2026-10-20 19:22:08 +02:00
• 6996d1ec4b sync : llama.cpp Georgi Gerganov 2026-10-28 11:48:20 +02:00
• 87d33b41f4 ggml-cuda : support ssm_conv kernel size 7 (llama/30873) Tarek Dakhran 2026-10-28 11:47:58 +02:00
• 08234b3b92 sync : llama.cpp Georgi Gerganov 2026-10-28 11:47:14 +02:00
• f6c63dda12 metal : optimize argmax and pool_1d (llama/30862) Georgi Gerganov 2026-10-20 16:46:00 +03:00
• aaed894808 opencl: add several gemm variants for small batches and some fixes (llama/30850) lhezandHongqiang Wang 2026-10-20 04:37:28 -07:00
• c2fe0d071a sycl : replace memset due to not always ordered (llama/30812) Neo Zhang 2026-10-20 19:35:48 +08:00
• d4f8391bc5 ggml-cpu : select the NEON fp32->fp16 conversion (llama/30505) Bernard Ladenthin 2026-10-20 13:35:06 +02:00
• fb198b56ac HIP/CUDA: Fix reduction in mmf getting tile widths not devisiable by 64 (llama/30836) uvos 2026-10-20 13:32:38 +02:00
• d5ba1a7f60 ggml-zdnn: introduce flash attention op (llama/30512) Aaron Teo 2026-10-20 19:31:35 +08:00
• d7d5807b32 CUDA: Clear CUDA graphs after legacy memory pool flush (llama/26771) Bálint Bokros 2026-10-20 13:30:50 +02:00
• 7d77885cb2 metal : fix missing MoE kernel variant for 9 experts per token (#30610) (llama/30699) Caladrel 2026-10-20 06:02:05 -05:00
• 2b9ab6bd0d hexagon: update to snapdragon-toolchain v0.8 and enable hexagon v85 (llama/30854) Max Krasnyansky 2026-10-19 23:49:59 -07:00
• 57ea06b838 sycl: fix compiler warnings - missing prototypes, dead code, extra semicolons (llama/30271) Michael Taylor 2026-10-20 01:47:25 -03:00
• 50e4a18eae sycl : fix get_rows on reordered quantized weights (llama/30430) Sam MurdochandSam 2026-10-20 10:47:21 +11:00
• 6dac34084f opencl: add cooperative-k dp4a gemm for q4_0, q4_k and q6_k (llama/30741) lhez 2026-10-19 12:30:39 -07:00
• 90e6a7a31a opencl: add get_rows for i32, cpy for i32 and f32 (llama/30756) lhez 2026-10-19 12:22:34 -07:00
• 96135fa886 metal : fix moe_reduce fusion race when dst aliases its inputs (llama/30795) Georgi Gerganov 2026-10-19 11:55:02 +03:00
• 228ecef1c3 cann : refactor #ifdef blocks to avoid unreachable code after return (llama/25273) Raul Torres 2026-10-19 09:17:54 +01:00
• 8601f6399c CUDA: fix warning for nonexistent extra semicolon (llama/30774) Johannes Gäßler 2026-10-19 10:08:21 +02:00
• aeadd496c5 ggml-cpu: vectorize ggml_v_silu and ggml_v_expf (llama/30791) Aaron Teo 2026-10-19 15:59:49 +08:00