Sae-lens

Fix

* fix: Update README.md (290) ([`1d1ac1e`](https://github.com/jbloomAus/SAELens/commit/1d1ac1e4c811f6062c1c4eb777d790a86957a082))

Fix

* fix: neuronpedia oai v5 sae ids ([`ffec7ed`](https://github.com/jbloomAus/SAELens/commit/ffec7ed2ee015846839c3ba128f48fe7edbe53b6))

Unknown

* Merge pull request 291 from jbloomAus/fix_np_oai_ids

fix: neuronpedia oai v5 sae ids ([`253191e`](https://github.com/jbloomAus/SAELens/commit/253191ee98e2a16637fe3f70183e81b8bd0b7705))

* fix the ids for att&34; ([`9e81c1c`](https://github.com/jbloomAus/SAELens/commit/9e81c1c216a599fdc80ac061895457e5eddce8fe))

Feature

* feat: Add SAE Bench SAEs (285)

* add ignore tokens in evals

* remove accidental hard coding

* fix mse

* extract sae filtering code

* add sae_bench saes

* use from pretrained no processing by default

* use open web text by default

* add estimate of number of SAEs print statements

* add unit tests

* type fix ([`680c52b`](https://github.com/jbloomAus/SAELens/commit/680c52b95abfb039687f5328c52b1b41f8c3e05d))

Fix

* fix: add OAI mid SAEs for neuronpedia ([`a9cb852`](https://github.com/jbloomAus/SAELens/commit/a9cb852ae89256a8912e260e9c1dcaf7eaa8bef4))

* fix: Gemma Scope 9b IT ids for Neuronpedia ([`dcafbff`](https://github.com/jbloomAus/SAELens/commit/dcafbffc15bdb46b77b7277f751cb915833bedf6))

Unknown

* Merge pull request 282 from jbloomAus/oai_mid_fix

fix: add OAI mid SAEs for neuronpedia ([`643f0c5`](https://github.com/jbloomAus/SAELens/commit/643f0c5cdf0d441a38d4c4279e42828fd6896aae))

* Merge pull request 281 from jbloomAus/fix_np_gemmascope_ids

fix: Gemma Scope 9b IT ids for Neuronpedia ([`e354918`](https://github.com/jbloomAus/SAELens/commit/e354918e24f2e4df2a3a0b8a49833774801484d2))

Fix

* fix: more gemma scope canonical ids + a few canonical ids were off. (280)

* fix model name in config for it models

* add 9b non-standard sizes

* fix att 131k canonical that were off

* fix mlp 131k canonical that were off ([`aa3e733`](https://github.com/jbloomAus/SAELens/commit/aa3e733a352052b7b27b48b9617cea99099b4c0a))

Fix

* fix: centre writing weights defaults (279)

* add model_from_pretrained_kwargs to SAE config

* default to using no processing when training SAEs

* add centre writing weights true as config override for some SAEs

* add warning about from pretrained kwargs

* fix saving of config by trainer

* fix: test ([`7c0d1f7`](https://github.com/jbloomAus/SAELens/commit/7c0d1f70e5169b1209fccf2280ebeb82b5e5917b))