v0.8.1

v0.8.1 (2024-05-21)

Fix

• fix: persona_generalization experiment script (#167)

• update persona gen script with datasets

• fix lint

• adding a test to track missing persona datasets

• adding all dataset prompts

• adding test for get_all_prompts

• adding qwen training script

• fixing qwen script

• tweaking qwen script

• update qwen sweep script

• adding llama layer sweep

• fixing llama2 layers

• fixing plot style

• fixing formatting

• adding plotting helper for steerability

• fixing tests

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com>
Co-authored-by: David Chanin <chanindav@gmail.com> (2eb6b62)

Unknown

• update plots (b914302)

• update figures for paper (ada05d8)

• update figures for id steering (35e11cc)

• update paper figures (c765cab)

• add figures for correlating id and ood steering (ffe1515)

• Paper/preprocessing (#170)

• add preprocessing script

• update figures

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (307878f)

• updates to id results (453737d)

• In distribution results (#168)

• add figures

• add figures

• wip: concept erasure

• update

• delete unused notebooks

• update plots

• concept erasure

• fix lint

• ignore type in random sv experiment

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (0ca6521)

• adding qwen formatting support and adding a sweep (#166)

• adding qwen formatting support and adding a sweep

• fixing formatting

• saving progress during sweep (245022f)

• delete unused notebooks (68b68e5)

• add randomly sampled datasts (796bf90)

v0.8.0

v0.8.0 (2024-05-07)

Feature

• feat: more datasets (#164)

• add all xrisk datasets

• make truthfulqa consistent with others

• fix some prompts

• Verify all existing persona prompts

• add persona prompts for Xrisk, sycophancy, tqa

• refactor: make more datasets

• minor

• fix lint

• fix test persona prompt len

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (0f6dff2)

v0.7.0

v0.7.0 (2024-05-05)

Feature

• feat: pipeline generation (9c69185)

Unknown

• lint: run linters (88cf51a)

v0.6.0

v0.6.0 (2024-04-30)

Chore

• chore: add note to explain cluster utils (a13a503)

• chore: add cluster utils (dcd731a)

• chore: modify path to include local dependencies (4a97734)

• chore: update time requirement in qsub (d1268c8)

• chore: delete incorrect test step (1f4c2bb)

• chore: update lockfile (40baf38)

• chore: update ci (#161)

• update ci

• chore: fix typo

• chore: add py312 test

• update ruff settings

• chore: ruff format

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (4482d27)

Feature

• feat: cross steering results db (#162)

• modify jobscript to submit array job

• chore: fix typo

• Add sqlite database for results

• implement multiprocessing to make db

• update db

• add notebook for in-distribution-steerability

• add analysis notebook for intra concept variability

• fix ruff format

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (cd16daa)

Unknown

• categorize persona prompts (6dd631f)

• adding persona analysis notebook (38d3ed4)

v0.5.0

v0.5.0 (2024-04-24)

Chore

• chore: replace hardcoded user id with daniel (b1271a6)

• chore: fix qsub script (5f50c60)

• chore: cleanup dead code (9e615ff)

Feature

• feat: slim eval results and allowing multipler multipliers in cross-steering (#159)

• adding slim eval results and allowing multipler multipliers in cross-steering

• fixing linting

• fixing typing (64d94aa)

• feat: add more persona prompts (#160)

• add more persona prompts

• add persona prompts, tests

• Modify persona_generalization to work with more prompts

• Add script to run persona generalization

• fix: translation constants

• Add test for variables

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (1cdf847)

• feat: steerability metrics (f9e22cc)

Fix

• fix: replace global var (d91a75f)

Unknown

• update test fixture (53d06ab)

• text: add dataset fixture (1b10ef4)

• set layer to 13 (f75473d)

• fixing linting (cdb5494)

• allow customizing metric name in persona plots (d379a6d)

• save persona generalization results individually (de8bbd0)

• Faster eval (#153)

• hopefully improving eval speed

• more device issues

• reformatting (03a6dfc)

• tweaking cross steering (2babe64)

v0.4.0

v0.4.0 (2024-04-12)

Chore

• chore: modify coderabbit config to reduce verbosity (f13fc43)

• chore: ignore png files (5c9ddc8)

Feature

• feat: improve steering experiments utils (#147)

• Add statsmodels

• Add notebook ow/ results on choosing steerability metric

• feat: add saving, loading for SVs

• Finish initial study on aggregation method

• rename

• fix: use train_completion_template

• Update lockfile

• Remove system prompt for config for backwards-compatibility

• feat: improve logging of missing steering configs

• Update notebooks

• chore: remove failing py311 ci run

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (f1487a9)

• feat: add functions to compute logit statistics (#145)

• Add functions to compute logit statistics

• Make logit statistics optional

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (c447864)

Refactor

• refactor: experiments (#141)

• Add concept metrics calculation

• fix: concept metrics

• Add unit test for metrics

• feat: layer-wise steering metrics

• update config fields

• update experiments code

• minor

• refactor: experiments code

• refactor: experiments code

• Test datasets exist before running

• fix: database

• add method to get config, fix delete_table

• changes

• more changes

• Fix bug in experiment path

• Add sweeps

• WIP

• Fix tests

• Fix tests

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (ee5f190)

Unknown

• updating persona generalization (#151)

• updating persona generalization

• temporarily disabling test due to cpu/cuda issue on ci (ef91f2f)

• Add evaluate_generalization.py notebook (4a25847)

• minor fixes (506ff11)

• WIP: Persona cross steering (#150)

• setting up cross-evaluation experiments

• improving progress reporting in experiments

• adding option to normalize steering magnitude to baseline

• tweaking params

• fixing nested progress

• updating persona evals

• passing eval params through persona experiment

• setting up script for persona generalization experiments

• more debugging output

• updating test

• fixing typing

• make datasets as part of experiments script

• fixing eval dataset selection

• fixing eval

• adding cross steering plots

• shorten labels in cross-steering plots

• WIP adding plotting helpers

• refactoring plotting code

• adding more plotting options

• adding more content to plots

• outptting more info in graphs (352df94)

• Add sft training examples (bcf8c2c)

• Experiments (#146)

• Add sweeps

• WIP experimental code

• Update experiment notebook

• Remove pycache

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (037bea3)

• Add notebooks to run experiments (9c662c7)

• Add fucntion to load sweep results (66cb2d7)

v0.3.0

v0.3.0 (2024-03-07)

Feature

• feat: steering experiments (#132)

• Add experimental code

• fix: make_country_capital script

• feat: add code to run steering experiment

• update experiments code

• fix: add --config_path arg

• fix: config yaml parsing

• chore: add more configs

• chore: add even more configs

• refactor: plotting

• feat: add script to run sweep

• fix: do not set completion template by default

• refactor sweeps

• refactor: token concept sweep

• fix: bugbears

• chore: add comments

• fix: steering_index of datasets

• test: steering token index

• updating steering_vectors library version

• evaluate on more layers

• refactor: use steering-vectors code, log instead of print

• chore: fix docstring

• test: training, evaluating steering vectors

• fix: minor

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com>
Co-authored-by: David Chanin <chanindav@gmail.com> (8d1bd7d)

v0.2.0

v0.2.0 (2024-03-06)

Feature

• feat: add 'prompt' string for bats dataset. (83688c1)

Fix

• fix: make_bats function (d789eb7)

Unknown

• Add BATS dataset (#125)

• Add BATS dataset

• test: add test for make_bats

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (dcb7ccd)

v0.1.0

v0.1.0 (2024-03-05)

Chore

• chore: remove pypi publishing from ci (6ddd28d)

Feature

• feat: refactor types (#123)

• Refactor types

• Delete old experimental code

• Refactor datasets

• Improve dataset split parsing; update make_dataset types

• Update preprocessed datasets save dir

• refactor: delete duplicated files

• Fix pyright errors

• Fix tests

• Fix lint

• chore: re-add tqa raw datasets

• feat: add tests for evaluators

• feat: re-add caa-style prompt

• chore: delete unused code

• refactor: migrate SteeringHook to repepo.core

• test: assert pos, neg prompts the same

---

Co-authored-by: Daniel CH Tan <dtch1997@users.noreply.github.com> (e600270)

v0.0.0

v0.0.0 (2024-03-04)

Chore

• chore: add semantic release to ci (56be43f)

Unknown

• Delete dead code (a1da6de)

• translate the main persona datasets from @dtch1997's work (#108) (b0b13a2)

• Translation experiments (#99)

• adding helpers for generating and parsing language from dataset filename

• adding compare_dataset_translations experiment

• adding experiment helpers

• tweaking pirate translation strings

• adding translations for non-persona mwes

• fixing up make mwe helper

• adding a 'ctx' pseudo-style

• Revert "adding a 'ctx' pseudo-style"

This reverts commit a0058c4.

• refactoring to allow using arbitrary dataset variations, insead of the hacky pseudo-language stuff

• fixing using existing results in cross-steering

• adding helpers to calculate jenson-shannon and KL for bernoulli distributions

• using js dist for steering deltas

• adding more tests (46927b0)

• adding translated persona MWE variants (#103)

• adding translated persona MWE variants by pre-pending the generation ctx to each example

• formatting translated_strings (1a42e96)

• adding google translate and re-translating persona datasets (#102)

• adding google translate and re-translating persona datasets

• fixing linting

• removing unused test (312b4ab)

• standardizing dataset naming around language (#100) (8fda2f9)

• Generalization experiments (#96)

• Add functions to do translation

• Add TQA translate

• Fix key name bug

• WIP

• Add script to generate TQA translated datasets

• update expt name and dataset splits

• Add Llama chat formatter

• Minor fixes in caa_repro

• Add options to print output, save steering vectors

• Set default experiment path by train / test datasets

• Add functionality to print examples

• Add script to plot results

• Add title to plotting code

• Fix pdm lock

• Add (very ugly) function to plot multiple results

Very ugly implementation but it works

• Ignore png files

• Enable translated system prompt

• Add new experiments dir

• Add notebook to analyze TQA vectors

• Add script to download datasets

• Add script to download datasets

• WIP translate

• Add code to extract and save steering vectors

• Update experiments

• Add more dataset names

• Improve dataset inspection

• Modify script to extract all SVs

• Changes to notebooks

• Update readme

• WIP

• Fix download datasets

• Enable 4-bit loading

• WIP

• Visualize pairwise cos similarities

• Inspect dataset s dataframe

• Clustering results

• Fix lint errors

• Add script to extract concept vectors

• WIP

• Refactoring

• Refactoring

• Add script to run all experiments

• Fix bug with results suffix

• Uncomment some lines

• Update README, bash script

• Restore original experiments dir

• Fix lint

• Fix lint

• Add more aggregations

• Fix bug in download

• Ignore html files

• Add test for data preprocessing

• Add tests for preprocessing

• fixing black formatting issues

• fixing typing

---

Co-authored-by: dtch1997 <dtch1997@users.noreply.github.com>
Co-authored-by: David Chanin <chanindav@gmail.com> (6680de7)

• Translate mwe and sycophancy (#97)

• importing raw persona MWE datasets from anthropic

• adding translation for mwe persona datasets and translating the first 5

• translating sycophancy datasets

• make_sycophancy_caa parses translations, and adding translations for misc strings

• adding a convenience wrapper to load_translation

• adding a script to make MWE personas datasets

• fix lint formatting

• alternating every 2 samples for MWE, not every 1 (ce83a8d)

• translating TQA into styles and languages with gpt4 (#94)

• translating TQA into styles and languages with gpt4

• dont force ascii, its not 1998

• fixing test mocking

• refactoring translations to make supporting more datasets easier (d6d241a)

• Caa tqa (#91)

• refactoring formatting and benchmarking to support CAA

• adding a basic test for get_normalized_correct_probs()

• fixing tests

• increasing sft loss threshold to make test less flaky

• adding a TQA CAA dataset / experiment

---

Co-authored-by: David Chanin <chanindav@gmail.com> (97b1236)

• Refactoring formatting and benchmarking to support CAA (#87)

• refactoring formatting and benchmarking to support CAA

• adding a basic test for get_normalized_correct_probs()

• fixing tests

• increasing sft loss threshold to make test less flaky (c98f067)

• Merge pull request #88 from dtch1997/openai-translators

Add openai as dependency, and translators notebook (c1fb281)

• Add openai as dependency, and translators notebook

This notebook has a few simple functions for translating inputs and
dataframes using gpt-4. You will need an openai API key to run the code
(obviously). (dadf9f2)

• fixing tests after CAA merge (#85) (6efbfaf)

• Caa experiments 2 (#80)

• Add script to generate CAA datasets

• Add correct CAA datasets

• Add gitignore for experiments

• Modify default template

• Add get_normalized_correct_probs function

• Add script to generate vectors

• Add scripts to prompt w/ SV, plot results

• Add notebook to compare our vs their CAA vectors

• Add instructions to reproduce results

• Add plots

• Add evaluator for normalized correct probs

• Skip failing tests

---

Co-authored-by: dtch1997 <dtch1997@users.noreply.github.com> (cca4b29)

• Fix failing test (b7975b1)

• refactoring prompting/formatting (#77)

• refactoring prompting/formatting

• fixing conflict in tests (dca53ac)

• Merge pull request #79 from dtch1997/swap-in-steering-vecs

swapping in steering-vectors lib (f86d0d2)

• Merge pull request #78 from dtch1997/verify-caa-steering

adding a test to assert our steering is identical to CAA steering (a5ea301)

• swapping in steering-vectors lib (8e86019)

• adding a test to assert our steering is identical to CAA steering (f26b2df)

• Verify our code matches CAA (#76)

• adding a llama chat formater and prompter based on CAA

• testing that our reading vectors match CAA reading vectors

• fixing linting

• fixing test (aa1dd24)

• cleaning up oddities with steering vecs and repe algo (#72) (773db50)

• CAA tweaks / improvements (#70)

• Add bitsandbytes, accelerate

• Hardcode second-last token activation position for steering vectors

• Add notebook diffmerge package for pretty git diffs

• Add note on how to change RepE directions

• Add note on how hooks work

• Add options to decouple reading and control

• fixing tests

---

Co-authored-by: dtch1997 <dtch1997@users.noreply.github.com> (1487639)

• CAA base (#69)

• adding a record_activations() function to make it easy to collect model activations

• replacing repe with our own CAA-esque implementation

• only patch generated tokens

• fix generating start index selection

• fixing pyright error (55980ea)

• Add CAA datasets (#68)

• Add CAA datasets

• Update makefile

• Add test for make_ab_prompt

---

Co-authored-by: dtch1997 <dtch1997@users.noreply.github.com> (c1ae7f1)

• Sft hf trainer (#50)

• Working HF trainer script

• customize wandb logging

• Remove unused keys from SFTDataset

• Add unit test for SFT

• Fix import

• Fix lint

• Fix lint (again)

• Fix test

• Fix benchmark, pipeline logic

Update the train_and_evaluate fn to be consistent
with alg...