v1.11.0

Thanks to @obalcells and @andyrdt Llama-2 models should now have 1e-4 atol logit errors rather than 1e0 errors!

We also now force PyTorch2 to be >= 2.1.1 thanks to a PyTorch issue on MPS @jettjaniak pointed out. Thanks all!

What's Changed

• Fix Grokking Notebook by @ArthurConmy in #450
• Fixed current CI issues with accuracy failing for Pythia model by @bryce13950 in #451
• Fixing Llama2 numerical errors by @obalcells in #456
• Pin PyTorch2 to be at least 2.1.1 by @ArthurConmy in #457

New Contributors

• @obalcells made their first contribution in #456

Full Changelog: v1.10.0...v1.11.0

v1.10.0

What's Changed

• Test main demo by @bryce13950 in #441
• Fix contributing docs typo by @alan-cooney in #444
• New model: bigcode/santacoder by @ojh31 in #445
• Added support for bloom-560m model by @SeuperHakkerJa in #434

New Contributors

• @ojh31 made their first contribution in #445
• @SeuperHakkerJa made their first contribution in #434

Full Changelog: v1.9.1...v1.10.0

v1.9.1

What's Changed

• Organise & fix README by @alan-cooney in #430
• add survey link by @jbloomAus in #440
• Relax CUDA requirements by @alan-cooney in #442

Full Changelog: v1.9.0...v1.9.1

v1.9.0

What's Changed

• Fix Exploratory Analysis Demo by @alan-cooney in #429
• Sync readme with docs by @alan-cooney in #428
• Move cspell conf to its own file by @alan-cooney in #431
• Improve ActivationCache docs by @alan-cooney in #432
• Improve getting started docs for Activation Cache by @alan-cooney in #433
• Add docs hot reloading instructions for contributors by @alan-cooney in #436
• Make unit & acceptance tests run in parallel by @alan-cooney in #435
• Update GitHub CD Actions by @alan-cooney in #437

Full Changelog: v1.8.1...v1.8.2

v1.8.1

What's Changed

• Fix all Sphinx warnings by @alan-cooney in #422

Full Changelog: v1.8.0...v1.8.1

v1.8.0

What's Changed

• Add status icons to the readme by @alan-cooney in #415
• Improve attention masking and key-value caching. by @UFO-101 in #386
• Remove Python 3.7 support by @alan-cooney in #423
• Update docs to show prepend_bos and padding_side are optional by @UFO-101 in #418
• Add Python 3.11 Support by @alan-cooney in #425
• Return the current residual when stop_at_layer is not None. by @UFO-101 in #420
• Add contributing instructions by @alan-cooney in #426

Full Changelog: v1.7.0...v1.8.0

v1.7.0

What's Changed

• Add start_at_layer parameter to HookedTransformer by @UFO-101 in #382
• fix: Set IGNORE value in mask to -torch.inf by @connor-henderson in #366
• bug fix attention inf by @jbloomAus in #389
• Bugfix attn left padding by @jbloomAus in #390
• Add developer tooling defaults & VS Code extensions by @alan-cooney in #394
• Fix llama tokenization issue by using tokenizer initialized with add_bos_token=True by @soheeyang in #379
• Move towards consistent commenting/docstring style by @alan-cooney in #395
• Speed up docs generation by @alan-cooney in #396
• Add hook_attn_in by @ArthurConmy in #336
• Added explicit dtype option by @neelnanda-io in #388
• Support PyTorch 2 with Poetry by @alan-cooney in #397
• Remove .venv from formatting checks by @alan-cooney in #401
• Add auto-organise imports on save in VsCode by @alan-cooney in #410
• Fix SolidGoldMagikarp tokenization test by @ArthurConmy in #408
• Add a summary docstring to the sub-modules by @alan-cooney in #407
• Remove unused root directory files by @alan-cooney in #406
• Simplify Docs Generation by @alan-cooney in #398
• Fix GitHub pages deploy by @alan-cooney in #412
• Fix optional types in HookedTransformer by @alan-cooney in #403
• Improve Hooked Transformer Docs by @alan-cooney in #400
• Improve API Docs Organization by @alan-cooney in #399
• Add docstring testing by @alan-cooney in #402
• Fix tokenization in utils.test_prompt by @Felhof in #334

New Contributors

• @UFO-101 made their first contribution in #382
• @connor-henderson made their first contribution in #366

Full Changelog: v1.6.1...v1.7.0

v1.6.1

What's Changed

• Add support for left padding by @soheeyang in #344
• Added gated MLP Hooks by @neelnanda-io in #374
• added support for pythia 160m seeds by @will-hath in #377
• Remove lru caching of weights by @ArthurConmy in #381
• Implement hook_mlp_in for parallel attention/MLP models by @ArthurConmy in #380

New Contributors

• @will-hath made their first contribution in #377

Full Changelog: v1.6.0...v1.6.1

v1.6.0

What's Changed

• Fix FactoredMatrix bug by @callummcdougall in #367
• Fix to automatically infer add_special_tokens for tokenizer by @soheeyang in #370

Full Changelog: v1.5.0...v1.6.0

(Release requested by @callummcdougall for bugfix),

v1.5.0

What's Changed

• Fix generate() by adding greedy decoding code for do_sample=False by @soheeyang in #358
• Updated readme by @neelnanda-io in #360
• Fix bug in rotary embedding for models other than llama and gpt-neo by @soheeyang in #365
• Switch to beartype by @dkamm in #325

Full Changelog: v1.4.0...v1.5.0