Commit Graph

Select branches

Hide Pull Requests

ara

auto-reproduce

e2e-tests

master

reproducibility-fixes

revert-308-anrp/loradetails

revert-46-abliteration-matmul-optimization

#101

#103

#106

#107

#108

#11

#110

#116

#117

#119

#12

#120

#126

#136

#137

#138

#138

#14

#148

#148

#15

#150

#151

#151

#154

#154

#157

#16

#160

#162

#165

#166

#166

#169

#17

#170

#171

#173

#174

#175

#176

#180

#182

#182

#183

#187

#188

#191

#193

#194

#196

#196

#20

#204

#206

#207

#209

#209

#210

#210

#211

#211

#212

#214

#215

#22

#223

#224

#225

#226

#229

#230

#231

#239

#241

#244

#244

#247

#254

#255

#257

#257

#259

#26

#260

#261

#262

#267

#268

#269

#270

#271

#272

#273

#276

#28

#284

#285

#286

#287

#29

#293

#294

#296

#301

#302

#303

#305

#306

#307

#307

#308

#309

#31

#312

#313

#313

#314

#314

#317

#319

#32

#321

#323

#324

#325

#326

#327

#328

#33

#332

#332

#334

#337

#338

#340

#344

#347

#347

#348

#349

#349

#35

#350

#352

#353

#354

#355

#356

#357

#358

#359

#361

#362

#362

#363

#364

#366

#366

#367

#368

#369

#372

#374

#376

#377

#378

#378

#381

#382

#383

#384

#384

#385

#386

#387

#388

#388

#389

#390

#390

#391

#391

#392

#393

#393

#394

#395

#395

#396

#396

#397

#398

#42

#43

#44

#45

#46

#5

#50

#51

#52

#53

#53

#55

#56

#60

#66

#67

#69

#71

#72

#73

#74

#76

#78

#79

#83

#86

#9

v1.0.1

v1.1.0

v1.2.0

v1.3.0

v1.4.0

Mono Color

• 94775d4148 chore: update dependencies Philipp Emanuel Weidmann 2026-03-15 09:31:32 +05:30
• 515a7b9eb5 fix: prevent div-by-zero in evaluator when base_refusals is 0 (#225) cpagac 2026-03-13 00:51:23 -05:00
• e26da5e0e6 fix: display all abliterable components across layers (#215) erm14254 2026-03-11 09:40:37 +01:00
• 4a6304c361 feat(ara): add abliteration method to model card Philipp Emanuel Weidmann 2026-03-10 11:16:09 +05:30
• c76416fe03 feat(ara): expand parameter ranges Philipp Emanuel Weidmann 2026-03-10 10:27:32 +05:30
• 2bb203ee47 fix(ara): store captured I/O tensors on CPU for multi-GPU robustness (#214) joninco 2026-03-07 09:10:45 -05:00
• d79a443e6f feat(ara): fix issues on some multi-GPU setups Philipp Emanuel Weidmann 2026-03-07 18:24:31 +05:30
• ec0367226d style: fix formatting and naming Philipp Emanuel Weidmann 2026-03-06 13:18:08 +05:30
• 5e3c04c802 feat: add Qwen3.5 MoE hybrid layer support (#187) Matthias Stegner 2026-03-06 08:33:57 +01:00
• 0bb9521fbe feat(ara): optimize all parameters Philipp Emanuel Weidmann 2026-03-05 08:55:58 +05:30
• 992fb3a4b3 feat(ara): change weights to match those used for the gpt-oss-20b demo Philipp Emanuel Weidmann 2026-03-04 16:48:25 +05:30
• 304c14adc7 feat(ara): replace weight parameters with balance parameter Philipp Emanuel Weidmann 2026-03-04 11:59:46 +05:30
• 56e57adf36 feat(ara): improve steering term Philipp Emanuel Weidmann 2026-03-04 11:06:01 +05:30
• bd1fa0ade4 feat(ara): remove tie_to_original_matrix term Philipp Emanuel Weidmann 2026-03-04 09:13:38 +05:30
• 3c5d6920bf feat(ara): fix memory leak Philipp Emanuel Weidmann 2026-03-02 18:50:27 +05:30
• b8f4a9c985 feat(ara): implement optimization for ARA parameters Philipp Emanuel Weidmann 2026-03-02 14:36:57 +05:30
• 154241f8a2 feat(ara): implement matrix optimization Philipp Emanuel Weidmann 2026-02-27 14:16:27 +05:30
• 303ba9d978 fix: recheck prefix after inserting predefined (#194) Spiky Moth 2026-02-27 03:37:33 +01:00
• ea7c59a55a feat(ara): add methods for obtaining module I/O Philipp Emanuel Weidmann 2026-02-25 17:44:46 +05:30
• cb4ef3fdfc docs: add Trendshift badge to README Philipp Emanuel Weidmann 2026-02-20 13:00:19 +05:30
• 4c80c4beb9 fix: report VRAM usage across all GPUs instead of only the default device (#169) cpagac 2026-02-17 01:23:41 -06:00
• 3a115e280c fix: produce card for local models with existing readme (#157) Spiky Moth 2026-02-15 14:40:10 +01:00
• 27097bfe8e build: bump version to 1.2.0 v1.2.0 Philipp Emanuel Weidmann 2026-02-14 18:11:42 +05:30
• 025ab3a881 fix: disable LoRA export for now Philipp Emanuel Weidmann 2026-02-14 16:56:12 +05:30
• 1179013999 docs: update README Philipp Emanuel Weidmann 2026-02-14 16:32:08 +05:30
• fe7bc1bae3 docs: update README Philipp Emanuel Weidmann 2026-02-14 10:47:28 +05:30
• e70a1a85e8 fix: don't load checkpoint when evaluating a second model Philipp Emanuel Weidmann 2026-02-14 10:02:17 +05:30
• e7f8be98b7 fix: only export tokenizer when exporting full model Philipp Emanuel Weidmann 2026-02-14 09:18:22 +05:30
• 6017bcd347 fix: use compatible release specifiers for non-dev dependencies Philipp Emanuel Weidmann 2026-02-13 12:27:57 +05:30
• dd0b3a2f69 docs: update README Philipp Emanuel Weidmann 2026-02-11 11:09:17 +05:30
• b873598b77 docs: improve settings documentation Philipp Emanuel Weidmann 2026-02-11 10:19:05 +05:30
• 10ceb3098e chore: update copyright notice Philipp Emanuel Weidmann 2026-02-11 09:46:36 +05:30
• 745b582414 ci: upgrade GitHub Actions to latest versions (#137) Salman Chishti 2026-02-08 11:19:04 +00:00
• d0e9462fb8 ci: upgrade GitHub Actions for Node 24 compatibility (#136) Salman Chishti 2026-02-08 11:18:12 +00:00
• f68a887a7b fix: improve code quality, improve UX, fix small bugs Philipp Emanuel Weidmann 2026-02-08 13:32:00 +05:30
• 2690655a83 feat: print memory usage during run Philipp Emanuel Weidmann 2026-02-02 21:18:01 +05:30
• 3525b1ac22 Implement Magnitude-Preserving Orthogonal Ablation (#52) Spiky Moth 2026-02-02 12:35:19 +01:00
• 42f5a9b553 fix: Use file instead of symlink lock (for windows) (#116) anrp 2026-01-25 14:04:01 +00:00
• 451db0b76e fix: specify study name (#119) anrp 2026-01-25 13:18:23 +00:00
• ebc22c299e feat: Allow study progress to be saved & resumed (#106) anrp 2026-01-23 14:19:37 +00:00
• d5c834c51d fix: Allow abliterating VL models (#108) anrp 2026-01-23 14:04:31 +00:00
• c86f49035e feat: Refactor save machinery and always allow user to save LoRA (#110) anrp 2026-01-20 13:23:47 +00:00
• 85a6ec5ecb fix: Include kernels (allows MXFP4 to be loaded in MXFP4 instead of upcasting) (#107) anrp 2026-01-16 12:00:24 +00:00
• 632b1da622 feat: add config file for slop reduction Philipp Emanuel Weidmann 2026-01-11 18:51:26 +05:30
• 1cfd09d7f3 ci: add style guide for Gemini Philipp Emanuel Weidmann 2026-01-09 14:58:56 +05:30
• 09be09e12e fix: restore classification of empty responses as refusals Philipp Emanuel Weidmann 2026-01-02 16:50:02 +05:30
• 039f6222d2 feat: allow overriding the system prompt per dataset Philipp Emanuel Weidmann 2025-12-31 14:26:44 +05:30
• c4b2ea0c42 feat: allow injecting prefixes and suffixes into prompts Philipp Emanuel Weidmann 2025-12-31 12:00:44 +05:30
• 02a5237a02 feat: add option to print prompt/response pairs Philipp Emanuel Weidmann 2025-12-27 14:48:29 +05:30
• cf8cf6f349 fix: address remaining ty complaint Philipp Emanuel Weidmann 2025-12-22 11:12:45 +05:30
• 2141e110fb ci: treat ty warnings as errors Philipp Emanuel Weidmann 2025-12-22 10:57:36 +05:30
• 39101137ef ci: add type checking Philipp Emanuel Weidmann 2025-12-22 10:48:42 +05:30
• 064bed9a9f fix: resolve issues raised by ty Philipp Emanuel Weidmann 2025-12-22 10:24:55 +05:30
• 8d44b65670 feat: add continuous optimization option(latest changes updated) (#76) _Vinayyyy_ 2025-12-20 18:57:57 +05:30
• 5ddef6fd2f feat: add more CoT templates Philipp Emanuel Weidmann 2025-12-20 17:12:46 +05:30
• 92d0c0d551 feat: enumerate all available GPUs on startup (#86) michaelh 2025-12-16 13:12:15 +01:00
• 243f821d93 feat: Add 4-bit loading + LoRA support for low VRAM optimization (#60) michaelh 2025-12-14 15:49:09 +01:00
• 9d1734855d feat: avoid excessive low divergence iteration (#73) Spiky Moth 2025-12-14 09:56:48 +01:00
• 740aab61ba feat: add max_memory parameter to limit memory usage (#83) George 2025-12-11 17:27:40 +02:00
• d9f2b0407a build: bump version to 1.1.0 v1.1.0 Philipp Emanuel Weidmann 2025-12-10 16:54:03 +05:30
• ca783db6c9 docs: update README Philipp Emanuel Weidmann 2025-12-10 16:30:35 +05:30
• 6acccac994 feat: add progress bars for plotting operations Philipp Emanuel Weidmann 2025-12-10 13:07:34 +05:30
• ac154a55a0 fix: suppress CoT output for thinking models Philipp Emanuel Weidmann 2025-12-09 11:54:08 +05:30
• 15781a8a0c fix: skip common response prefix for thinking models Philipp Emanuel Weidmann 2025-12-09 08:25:10 +05:30
• 24c3aeb442 feat: turn boolean settings into CLI flags Philipp Emanuel Weidmann 2025-12-07 11:37:07 +05:30
• ffbde3ac2a fix: follow up after recent PRs Philipp Emanuel Weidmann 2025-12-07 10:26:16 +05:30
• 932d737edf feat: add silhouette coefficient to residual geometry output Philipp Emanuel Weidmann 2025-12-07 08:48:38 +05:30
• 1f5e977f4f Revert "perf: optimize abliteration matrix op (#46)" (#74) Philipp Emanuel Weidmann 2025-12-07 06:30:37 +05:30
• 49294d2abf Revert "perf: optimize abliteration matrix op (#46)" revert-46-abliteration-matmul-optimization Philipp Emanuel Weidmann 2025-12-07 06:28:19 +05:30
• da27ba8054 fix: always left-pad inputs, and avoid optimizing for empty responses Philipp Emanuel Weidmann 2025-12-06 06:31:09 +05:30
• baf5b0b0d1 feat: add geometric median to residual geometry output Philipp Emanuel Weidmann 2025-12-05 20:15:50 +05:30
• eeb28b28c1 feat: add option to plot residual vectors Philipp Emanuel Weidmann 2025-12-04 14:22:29 +05:30
• d836fb2da9 ci: add PR title lint (#66) red40maxxer 2025-12-02 22:55:48 -05:00
• 60bd531fde perf: optimize abliteration matrix op (#46) red40maxxer 2025-12-01 21:43:43 -05:00
• 1f74ac2888 Guard against refusals in broken English (#45) Spiky Moth 2025-11-26 06:59:08 +01:00
• 63fc0e7d5a feat: Add bfloat16 to default dtypes list (#44) _Vinayyyy_ 2025-11-25 12:22:52 +05:30
• 1efc4ee9e1 Featuring Notebook (Colab/Kaggle) Compatibility (#42) _Vinayyyy_ 2025-11-24 19:46:39 +05:30
• 452b35e7b7 Add trust_remote_code configuration option (#31) Nikolai Kolodziej 2025-11-24 01:57:44 +01:00
• b79b8b1475 Improve support for loading local datasets (#33) Spiky Moth 2025-11-23 06:45:34 +01:00
• 83cbf0612a Add option to print refusal geometry Philipp Emanuel Weidmann 2025-11-22 13:18:54 +05:30
• c35f3031f8 Allow stopping the optimization process early with Ctrl+C Philipp Emanuel Weidmann 2025-11-21 10:11:00 +05:30
• 2e1bb4b655 Use PYTORCH_ALLOC_CONF instead of deprecated PYTORCH_CUDA_ALLOC_CONF (#32) Nikolai Kolodziej 2025-11-21 02:57:28 +01:00
• af02bc6ece Fix support for MXFP4 quantized models with Triton tensors (#28) Anthony Eufemio 2025-11-19 22:13:06 -10:00
• 22a4a5b5b5 Add citation information to README Philipp Emanuel Weidmann 2025-11-19 12:14:17 +05:30
• 694edf18d3 Follow up after recent PRs Philipp Emanuel Weidmann 2025-11-19 11:19:47 +05:30
• c9c022a143 Fix linting issues Philipp Emanuel Weidmann 2025-11-19 10:16:58 +05:30
• 9905d9517f Fix formatting issues Philipp Emanuel Weidmann 2025-11-19 10:04:43 +05:30
• f06e939791 Add Ruff as a dev dependency Philipp Emanuel Weidmann 2025-11-19 09:59:18 +05:30
• f3b9826ca4 Add CI workflow Philipp Emanuel Weidmann 2025-11-19 09:45:54 +05:30
• 13bb7b24d6 Fix KeyError when HuggingFace user profile fields are missing (#20) Richard Young, PhD 2025-11-18 16:02:50 -08:00
• c8b6663b93 Fix multi-GPU support and memory management (#17) Nikolai Kolodziej 2025-11-19 00:39:12 +01:00
• 61fdf72b42 Add support for Granite MoE Hybrid in model.py by including down projections for shared MLP and MoE experts (#14) Ooze 2025-11-18 06:02:58 +03:00
• 7bad84b4f1 perf: clear residuals after computing direction (#15) red40maxxer 2025-11-17 11:48:22 -05:00
• 09730bad70 MPS support (#5) Matt Barnson 2025-11-17 05:12:01 -08:00
• b3545e4b1e Fix retrieving package version v1.0.1 Philipp Emanuel Weidmann 2025-11-16 17:35:13 +05:30
• 3f346b6150 Change package name Philipp Emanuel Weidmann 2025-11-16 17:01:50 +05:30
• 1a59d226c1 Fix spacing after images in README Philipp Emanuel Weidmann 2025-11-16 16:06:08 +05:30
• 12ecf50033 Add README Philipp Emanuel Weidmann 2025-11-16 15:19:27 +05:30
• ea699dce46 Improve appearance of selection menus Philipp Emanuel Weidmann 2025-11-16 11:32:58 +05:30
• 8a1aceff11 Switch to multi-objective optimization Philipp Emanuel Weidmann 2025-11-14 18:04:23 +05:30