mlx-examples

zhangyiss/mlx-examples

Fork 0

mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-08-27 19:21:50 +08:00

Commit Graph

Select branches

Hide Pull Requests

awq

awq-tq

dist-eval

distributed-layers

fix_whisper_temp

fix_wwdc_notebook_models

flux-dist-improv

flux-qlora

load-gguf

main

openlm

packed-quants

#1

#10

#1001

#1003

#1004

#1006

#1009

#1012

#1013

#1014

#1015

#1016

#102

#1020

#1023

#1024

#1026

#1027

#1028

#103

#1035

#1036

#1037

#1038

#1040

#1045

#1047

#1048

#1048

#1049

#105

#1050

#1054

#1055

#106

#1061

#1062

#1063

#1065

#1068

#1069

#107

#1070

#1072

#1074

#1075

#1078

#1079

#108

#1080

#1081

#1082

#1085

#1089

#1090

#1092

#1093

#1094

#1096

#1099

#110

#1100

#1103

#1105

#1113

#1114

#1115

#1117

#1118

#1119

#112

#1121

#1122

#1125

#1128

#1129

#113

#1131

#1132

#1133

#1135

#1137

#1140

#1141

#1143

#1144

#1145

#1146

#1148

#1149

#115

#1152

#1153

#1154

#1155

#1156

#1157

#1158

#1159

#116

#1163

#1164

#1166

#1173

#1174

#1176

#1178

#1180

#1189

#1191

#1192

#1193

#1194

#1196

#1197

#1199

#12

#1200

#1202

#1204

#1205

#1206

#1208

#1209

#121

#1210

#1211

#1212

#1213

#1214

#1215

#1216

#1217

#1220

#1222

#1225

#1228

#1229

#123

#1230

#1230

#1231

#1231

#1233

#1234

#1235

#1240

#1241

#1242

#1246

#1249

#1250

#1251

#1253

#1257

#1259

#1260

#1263

#1265

#1267

#1270

#1271

#1272

#1273

#1276

#1277

#1278

#1279

#1280

#1283

#1287

#129

#1291

#1294

#1295

#1297

#1298

#1299

#1300

#1301

#1302

#1304

#1305

#1308

#1311

#1312

#1314

#1315

#1316

#1319

#1321

#1322

#1323

#1324

#1325

#1326

#133

#1330

#1331

#1332

#1336

#1338

#1339

#134

#1340

#1342

#1346

#1348

#1349

#1350

#1353

#1358

#1359

#136

#1364

#1365

#1367

#1370

#1371

#1375

#1375

#1377

#1377

#1383

#1383

#1385

#1385

#1386

#1386

#14

#140

#141

#144

#145

#147

#149

#151

#152

#153

#154

#157

#158

#159

#160

#161

#162

#163

#164

#165

#166

#167

#169

#171

#172

#173

#174

#176

#177

#178

#180

#183

#184

#186

#187

#189

#19

#190

#191

#192

#193

#195

#197

#198

#2

#200

#201

#202

#203

#205

#208

#211

#213

#214

#215

#219

#22

#221

#222

#227

#229

#23

#231

#234

#235

#237

#238

#24

#240

#241

#242

#243

#245

#248

#250

#251

#252

#253

#254

#255

#257

#260

#263

#264

#265

#266

#269

#27

#270

#271

#272

#274

#275

#276

#278

#282

#284

#285

#287

#290

#291

#292

#293

#294

#295

#30

#300

#301

#302

#303

#306

#307

#308

#309

#310

#311

#312

#315

#318

#319

#32

#320

#321

#325

#326

#33

#331

#333

#335

#337

#338

#340

#342

#343

#347

#350

#351

#352

#353

#354

#358

#36

#360

#361

#363

#364

#365

#366

#369

#37

#373

#375

#377

#378

#379

#380

#386

#387

#388

#389

#391

#392

#393

#396

#397

#398

#399

#40

#401

#405

#408

#409

#41

#41

#411

#413

#414

#415

#417

#418

#419

#42

#420

#421

#424

#426

#427

#429

#43

#431

#432

#433

#439

#441

#443

#445

#446

#449

#45

#450

#451

#453

#455

#457

#458

#461

#462

#466

#467

#468

#469

#47

#470

#471

#472

#474

#475

#479

#48

#482

#483

#483

#486

#489

#491

#494

#495

#496

#497

#498

#5

#50

#501

#502

#503

#505

#506

#509

#51

#510

#514

#515

#516

#518

#519

#52

#520

#521

#522

#523

#528

#53

#530

#531

#534

#536

#539

#541

#544

#545

#546

#547

#548

#549

#55

#552

#555

#558

#562

#563

#565

#566

#569

#570

#571

#572

#573

#574

#577

#578

#580

#581

#585

#589

#590

#591

#592

#595

#596

#599

#602

#603

#604

#608

#609

#610

#611

#613

#614

#62

#621

#623

#628

#632

#633

#634

#636

#639

#64

#640

#643

#644

#645

#648

#650

#651

#654

#657

#66

#661

#665

#666

#667

#668

#670

#673

#674

#675

#676

#679

#680

#681

#682

#683

#684

#685

#687

#688

#69

#690

#691

#693

#694

#697

#698

#7

#701

#702

#703

#705

#707

#708

#711

#712

#715

#716

#717

#719

#720

#721

#729

#73

#731

#735

#736

#739

#74

#740

#743

#744

#746

#749

#75

#752

#753

#758

#76

#760

#763

#766

#77

#770

#773

#775

#778

#779

#78

#782

#789

#79

#790

#792

#793

#794

#797

#798

#8

#80

#800

#802

#803

#806

#807

#810

#813

#817

#818

#82

#821

#822

#824

#825

#827

#828

#830

#831

#833

#835

#837

#838

#839

#84

#840

#85

#851

#852

#853

#855

#856

#857

#86

#863

#867

#87

#871

#877

#879

#88

#882

#885

#886

#888

#889

#89

#890

#891

#895

#898

#899

#90

#902

#903

#904

#905

#906

#907

#911

#913

#914

#915

#915

#920

#923

#926

#93

#931

#932

#935

#936

#937

#94

#940

#942

#945

#946

#948

#949

#954

#955

#956

#957

#96

#960

#961

#962

#963

#965

#969

#97

#971

#971

#973

#979

#98

#981

#983

#984

#989

#99

#990

#991

#993

#995

#996

#998

35ecc17042 fix Goekdeniz-Guelmez 2025-02-11 17:07:08 +0100
e80bf95182 fix Goekdeniz-Guelmez 2025-02-11 09:26:43 +0100
e96afe9e9f updates Goekdeniz-Guelmez 2025-02-11 09:09:28 +0100
bded1a8fcd

fix looping in whisper (#1273) Awni Hannun 2025-02-10 13:04:35 -0800
3dd1a228e3 fix looping in whisper Awni Hannun 2025-02-10 12:57:18 -0800
88ca747e9e nits Goekdeniz-Guelmez 2025-02-10 19:46:19 +0100
b7bc811507 nits Goekdeniz-Guelmez 2025-02-10 19:45:19 +0100
9e28bedc81 Fix comment Matt Clayton 2025-02-10 12:32:00 -0500
4df23e961c Re-work for minimal changes, add test Matt Clayton 2025-02-10 12:30:12 -0500
fff5daeb85 Cleanup Matt Clayton 2025-02-10 12:02:14 -0500
93591970cf Add from_draft field in GenerationResponse Matt Clayton 2025-02-10 12:00:11 -0500
e5aa2c3b5d nits Goekdeniz-Guelmez 2025-02-10 17:51:14 +0100
f88e897019 removing helper functions Goekdeniz-Guelmez 2025-02-10 16:07:28 +0100
09275facd5

Update requirements.txt - adding torch Mike 2025-02-10 12:50:41 +0000
3716946aae Merge branch 'adding-dpo-training' of https://github.com/Goekdeniz-Guelmez/mlx-examples into adding-dpo-training Goekdeniz-Guelmez 2025-02-10 10:56:57 +0100
aa932e83e3 nits Goekdeniz-Guelmez 2025-02-10 10:56:50 +0100
16a3c7613f

Merge branch 'main' into adding-dpo-training Gökdeniz Gülmez 2025-02-10 10:55:39 +0100
d9da35f458 nits Goekdeniz-Guelmez 2025-02-10 10:52:32 +0100
575ece6ef0

Merge branch 'main' into adding-orpo-training Gökdeniz Gülmez 2025-02-10 10:51:01 +0100
0dac286539

Merge branch 'main' into adding-GRPO-training Gökdeniz Gülmez 2025-02-10 10:43:22 +0100
61e1063dec hunyuan finetune Awni Hannun 2025-02-09 21:30:35 -0800
5865899c81

Completion only fine-tuning of instruction models with collections of HF datasets (#1103) Chime Ogbuji 2025-02-09 23:12:34 -0500
eda597bdef simplify Awni Hannun 2025-02-09 19:37:11 -0800
bb2c8bcf96 more nits Awni Hannun 2025-02-09 17:58:15 -0800
6e9542a934 put offset in prompt, simplify Awni Hannun 2025-02-09 17:31:23 -0800
1ced1b00ca

rm temp argument (#1267) Sri Harsha Pamu 2025-02-09 11:39:11 -0800
8b21611e51 rm temp argument Sri Harsha Pamu 2025-02-09 09:54:26 -0800
6ace6dc6b2 simplify collections Awni Hannun 2025-02-09 08:32:18 -0800
00712522ba rebase loss calculation Goekdeniz-Guelmez 2025-02-09 17:13:05 +0100
a527cdb39b fix: prevent gradients from flowing through the reference model's logits Goekdeniz-Guelmez 2025-02-09 17:02:58 +0100
b9748e9ee4 Generalize the get_item method to all CompletionDatasets Chime Ogbuji 2024-12-09 11:46:00 -0500
7989d0a874 Move response template to LoRA configuration Chime Ogbuji 2024-12-08 12:32:52 -0500
95e1f22812 Incorporate use of response template for completion masking Follow example of trl's DataCollatorForCompletionOnlyLM to use response template to identify beginning of completion/continuation tokens for the purpose of masking out the other tokens during loss calculation Chime Ogbuji 2024-12-08 11:26:50 -0500
cb87f6f22c Add response template (or token) argument For use in calculating mask for everything up to the after the response prompt (i.e., the continuation/completion) Chime Ogbuji 2024-12-08 11:24:19 -0500
6df285ef6c Synch use of special tokens with iterate_batches Chime Ogbuji 2024-12-06 07:20:01 -0500
f989401881 Default for hf_datasets configuration Chime Ogbuji 2024-12-06 07:07:06 -0500
5ce58e4b6a Update documentation Chime Ogbuji 2024-11-10 10:10:04 -0500
3f08dfc762 Don't dupe BOS Chime Ogbuji 2024-11-10 10:08:44 -0500
69282ab7fc Minor fix Chime Ogbuji 2024-11-06 12:58:00 -0500
4890870053 Add ability to fetch raw prompt and completion text from completion datasets Chime Ogbuji 2024-11-06 12:53:54 -0500
a5b866cf73 Fix index calculation Chime Ogbuji 2024-11-03 20:36:55 -0500
a4a86ad898 Fix iteration over HF dataset collection Chime Ogbuji 2024-11-03 20:30:47 -0500
78c33e5037 Fix keyword argument invokation Chime Ogbuji 2024-11-03 20:26:15 -0500
387c45efa2 Fixes to references to hf_datasets Chime Ogbuji 2024-11-03 20:04:15 -0500
214c79be9c Fixes to config format in documentattion Chime Ogbuji 2024-11-03 20:00:35 -0500
8ec802f468 Updates to LoRA documentation Chime Ogbuji 2024-11-03 19:41:09 -0500
14a75f3f03 Generalize HF datasets to a collection of HF dataasets via datasets, adds support for custom chat HF datasets (#1088), and fixes (#1087) Chime Ogbuji 2024-11-03 19:11:54 -0500
3496cbea46 Add input masking for fine-tuning in documentation Chime Ogbuji 2024-11-10 09:54:32 -0500
71d9f8cc38 Fix Chime Ogbuji 2024-11-06 20:29:12 -0500
02abeeade4 Update sublist search and calculation of input id length Chime Ogbuji 2024-11-06 13:57:59 -0500
30fd5af843 Fix variable reference Chime Ogbuji 2024-11-06 12:33:49 -0500
27cd361d76 Updates CL lora tuner with input masking that uses default_loss (and iterate_batches) by default. Chime Ogbuji 2024-11-05 19:10:01 -0500
84fc1bde48 Minor documentation update Chime Ogbuji 2024-11-05 15:25:45 -0500
79a042768f Replace iterate_input_masked_batches with iterate_delineated_batches, an updated attempt to better sync with iterate_batches logic Chime Ogbuji 2024-11-05 15:17:23 -0500
604be3cec9 Add input_masked loss calculation and batching w/ padding Chime Ogbuji 2024-06-07 12:35:07 -0400
54179901b5 fix Goekdeniz-Guelmez 2025-02-09 15:41:47 +0100
39e9469059 freeze ref model Goekdeniz-Guelmez 2025-02-09 15:30:51 +0100
9ba6146a76 fix Goekdeniz-Guelmez 2025-02-09 14:32:50 +0100
f58c7de901

Some improvements to speedup alignment computation in MLX Whisper (#1259) Awni Hannun 2025-02-08 15:47:00 -0800
1503bd4f55

support hunyuan 7b (#1263) Awni Hannun 2025-02-08 15:46:47 -0800
31611b62d7

Add IBM granite model (#1265) Awni Hannun 2025-02-08 15:46:15 -0800
b522102b80 add thinking option Awni Hannun 2025-02-08 08:50:43 -0800
cc9af4f19b add granite Awni Hannun 2025-02-08 08:34:50 -0800
3fc1b2effb support hunyuan 7b Awni Hannun 2025-02-08 08:08:23 -0800
5bffb74bc4 add logits processor to spec gen Awni Hannun 2025-02-07 20:38:45 -0800
6120a5f376

Faster DSv2/3 expert score computation (#1257) Awni Hannun 2025-02-07 10:24:57 -0800
f3960c2b4b fix alignment Awni Hannun 2025-02-07 09:30:20 -0800
9414429309 some improvements to speedup alignment computation in MLX Whisper Awni Hannun 2025-02-07 09:07:24 -0800
52c41b5b5a

Fix prompt cache for models without chat template (#1250) Awni Hannun 2025-02-06 11:10:58 -0800
c9672041ec

Merge branch 'ml-explore:main' into adding-orpo-training Gökdeniz Gülmez 2025-02-06 19:31:31 +0100
747c08e202

Chore: pre-commit bump (#1253) Nripesh Niketan 2025-02-06 17:06:31 +0000
db2340b85e

Merge branch 'main' into pre-commit_bump_6/2/25 Nripesh Niketan 2025-02-06 16:59:52 +0000
6084086f85 Chore: pre-commit bump Nripesh Niketan 2025-02-06 16:59:01 +0000
65b792d7c0 fix lazy load distributed-layers Awni Hannun 2025-01-14 13:14:48 -0800
617f9289b9 Make the chat distributed Angelos Katharopoulos 2024-11-05 13:09:34 -0800
026362e0f8 Remove async eval and add sequential load Angelos Katharopoulos 2024-11-05 13:04:07 -0800
a0ce0594f6 Temporarily remove async_eval Angelos Katharopoulos 2024-08-29 10:33:43 -0700
d77840207c Start distributed inference for llama models Angelos Katharopoulos 2024-07-15 13:24:50 -0700
594b4356cb

Merge branch 'ml-explore:main' into adding-orpo-training Gökdeniz Gülmez 2025-02-06 08:16:22 +0100
67106711ce

Merge branch 'ml-explore:main' into adding-dpo-training Gökdeniz Gülmez 2025-02-06 08:16:10 +0100
94dcd0f63e

Merge branch 'ml-explore:main' into adding-GRPO-training Gökdeniz Gülmez 2025-02-06 08:15:58 +0100
ec50a869b0 init grpo cshang 2025-02-05 18:50:14 -0800
087adcfacb fix prompt cache with no chat template Awni Hannun 2025-02-05 06:37:31 -0800
ef1483420a fix prompt cache with no chat template Awni Hannun 2025-02-05 06:32:35 -0800
550c17f0de compile and use put along axis in deep seek routing function Awni Hannun 2025-01-31 09:54:04 -0800
6e6ba07b54 fix deepseek sharding (#1242) Awni Hannun 2025-02-03 16:59:50 -0800
bcfa55d882 updates Goekdeniz-Guelmez 2025-02-05 15:02:12 +0100
0a19522ec4 updates Goekdeniz-Guelmez 2025-02-05 14:38:09 +0100
35a2d99cf9 smoll fix Goekdeniz-Guelmez 2025-02-05 11:30:21 +0100
a33cad84b4 udpates Goekdeniz-Guelmez 2025-02-05 09:48:00 +0100
d84ad0cf86 fix testing Goekdeniz-Guelmez 2025-02-05 08:53:30 +0100
2a8e6f6e44 udpate Goekdeniz-Guelmez 2025-02-05 08:47:03 +0100
0a09a93454 fix cache handling Goekdeniz-Guelmez 2025-02-05 08:44:06 +0100
e2e5478da5

READMEs: fix typo in link, minor update. (#1246) Pedro Cuenca 2025-02-04 20:52:32 +0100
a0cd8c2900 READMEs: fix typo in link, minor update. Pedro Cuenca 2025-02-04 20:47:28 +0100
56712664f6 nice metric printing in testing Goekdeniz-Guelmez 2025-02-04 11:21:52 +0100
b1c1e1353e nice printing the test metrics Goekdeniz-Guelmez 2025-02-04 11:19:59 +0100
069431bd65 adding test_ppl in testing Goekdeniz-Guelmez 2025-02-04 11:18:09 +0100
43940ec673 fix Test Goekdeniz-Guelmez 2025-02-04 11:13:07 +0100
1beefd58a0 add create_dataset Goekdeniz-Guelmez 2025-02-04 11:06:57 +0100