v0.1.2

What's Changed

• [P0] make sure config only has strs by @aryamanarora in #144
• [Minor] Update README.md by @frankaging in #145
• [P1] Add Gemma + minor fixes by @aryamanarora in #149
• Add Llava integration to Pyvene by @ZhengPeterWang in #151
• [Minor] Adding use_cache flag to intervenable model forward call by @AmirZur in #152
• Update visualization code for causal model and rotation matrices by @atticusg in #156
• [Minor] Remove ortho init for DAS by @frankaging in #160
• [Minor] Bump up version by @frankaging in #161

New Contributors

• @AmirZur made their first contribution in #152

Full Changelog: v0.1.1...v0.1.2

v0.1.1: Cleanup requirements

What's Changed

• [P0] removing seaborn dependency by @frankaging in #143

Full Changelog: v0.1.0...v0.1.1

v0.1.0: Fix requirements

What's Changed

• Zen/update0.0.9 by @frankaging in #142

Full Changelog: v0.0.9...v0.1.0

v0.0.9: Minor updates on causal models

What's Changed

• [Minor] Revert CausalModel to accept input/output functions when generating factual/counterfactual datasets by @amirzur2023 in #137
• [P0] Adding back GPT2 and other model supports by @frankaging in #139
• [P0] fix requirements.txt by @aryamanarora in #140
• [Minor] version bump by @frankaging in #141

Full Changelog: v0.0.8...v0.0.9

v0.0.8: Minor updates on model APIs

What's Changed

• [Minor] Add in trainable intervention based on binary mask intervention by @frankaging in #102
• [P1] Gradent interventions tests by @ZhengPeterWang in #103
• [Minor] Adding retry to gradient test cases to avoid failure by @frankaging in #104
• [Minor] Adding support for any two-input lambda interventions; ResNet tutorials (no config) by @frankaging in #105
• [Minor] Enable gradient for base inputs by @frankaging in #107
• [Minor] Adding an exploratory tutorial on voting in LLaMA by @frankaging in #108
• [P1] Added BLIP-ITM model definitions by @KhoomeiK in #106
• [Minor] Fix typo in the voting tutorial; fix a bug in generation by @frankaging in #109
• [Minor] Adding in subspace support for lambda by @frankaging in #113
• Merge main into peterwz by @ZhengPeterWang in #115
• [P1] Update transformers version by @ZhengPeterWang in #116
• Add GPT2 for sequence classification by @atticusg in #117
• [P1] Disable base output by default in fwd() and gen() by @frankaging in #118
• [P2] IOI replication notebooks by @aryamanarora in #119
• [P0] Rename train/evaluate methods so train/eval can set model state by @aryamanarora in #120
• [P1] Add IntervenableModel.named_parameters() by @aryamanarora in #122
• [P0] Fix unseen model type support by @frankaging in #123
• [Minor] Support zero_grad as a module by @frankaging in #124
• [Minor] Allow other field passing in generate besides input_ids by @frankaging in #125
• [Minor] Loading interventions as non-static method by @frankaging in #126
• [Minor] Expose more APIs for loading and saving by @frankaging in #127
• [Minor] Update README.md by @frankaging in #128
• [Minor] Allow sharing interventions across multiple positions by @frankaging in #129
• [Minor] Fix the pyvene101 with correct examples by @frankaging in #130
• Test Cases for Causal Model, MQNLI Intro Notebook by @amirzur2023 in #131
• [Minor] Update README.md with the correct citation link by @frankaging in #132
• [Minor] Update CONTRIBUTING.md by @eltociear in #133
• [Minor] Support MistralModel and MistralForCausalLM by @jiudingsun01 in #135
• [Minor] Accepting labels field for loss calculation by @frankaging in #136

New Contributors

• @KhoomeiK made their first contribution in #106
• @amirzur2023 made their first contribution in #131
• @eltociear made their first contribution in #133

Full Changelog: v0.0.7...v0.0.8

v0.0.7: Major API updates with additional intervention supports

Descriptions:

• Remove redundant keywords Intervenable for singleton classes or types everywhere.
• API supports a wide range of dynamic casting, and dynamic broadcasting.
• Providing additional code examples on intervention sharing and complex causal intervention schemes.
• Update basic 101 tutorials with simple onboarding examples.

What's Changed

• Make self.embed_dim not None. by @ZhengPeterWang in #49
• [Minor] allow dynamic boundary init for boundless das by @frankaging in #50
• [Minor] Minor change to test with num of label by @frankaging in #55
• [Minor] add simple location broadcast for easy interface (#52) by @frankaging in #56
• [Minor] update test name by @frankaging in #58
• [P1] Adding in constant source intervention support with new tests by @frankaging in #59
• [Minor] Add in full causal tracing fig 1 reproduced results by @frankaging in #60
• [Minor] Update to full reproduce result fig1 the ROME paper by @frankaging in #61
• [Minor] incidental changes to the tutorials for updated results by @frankaging in #65
• Zen/causaltracingfull by @frankaging in #66
• [Minor] update broadcast in generation by @frankaging in #69
• Model generation API simplified and cleanup tech debt on redundant variables by @frankaging in #70
• [Bug Fix] Topological order scoring is not accurate by @frankaging in #75
• [Major] Update API namings and parameters by @frankaging in #78
• [Minor] Update a few doc links by @frankaging in #79
• [Minor] Add Viz by @frankaging in #80
• gather_neurons() unit tests by @ZhengPeterWang in #81
• Add preliminary scatter() function tests by @ZhengPeterWang in #82
• Add an extra test on scatter_neurons() testing case with no head. by @ZhengPeterWang in #84
• [Minor] Support backpack and replicate by @frankaging in #85
• [Bug fix] fix backpack init by @frankaging in #87
• More scatter_neurons() tests by @ZhengPeterWang in #88
• [Minor] fix requirements.txt to have a more stable Colab experience (#86) by @frankaging in #89
• Update datagenerators to support tokenizing for LMs by @atticusg in #91
• [Major] Update with string access and code refactory (#83) by @frankaging in #93
• [Minor] adding PR and BUG template by @frankaging in #94
• Model utils tests for GPT-2 by @ZhengPeterWang in #92
• [Minor] Support ITI Paper Results (#68) by @frankaging in #95
• [P1] Adaptive changes with pyvene 101 colab by @ZhengPeterWang in #96
• [Minor] Fix gradient backprop trainables with upstream interventions by @frankaging in #97
• [P0] remove autograd debugging by @aryamanarora in #98
• [P2] Update probing tutorial by @aryamanarora in #99
• [P1] Add intervention util unit tests by @ZhengPeterWang in #100
• [Minor] Bump up pip version by @frankaging in #101

Full Changelog: v0.0.6...v0.0.7

v0.0.6: Minor fix to remove default cache directory

Merge pull request #48 from stanfordnlp/zen/cachepath

Remove default cache directory

v0.0.5: Minor update after library migration for various links

Merge pull request #47 from stanfordnlp/zen/setup

update setup link, tutorial links

v0.0.4: Minor fix on library imports

fix library imports

v0.0.3: Add support for probe training, interchange intervention training

Descriptions:

1. Add generic activation collection intervention to collect activations for probe training. It is integrated with other interventions. You can intervene and then collect.
2. Support interchange intervention training.
3. More tutorials are provided.
4. Add CONTRIBUTING.md for dev rules.
5. Change to relative imports.