Merge pull request #52 from adamkarvonen/update_scr_tpp

update scr_tpp_schema to show top 20 by default
adamkarvonen · Jan 28, 2025 · f551e7b · f551e7b
2 parents 7b2021c + 59320e2
commit f551e7b
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 2 deletions.
diff --git a/sae_bench/evals/scr_and_tpp/eval_output.py b/sae_bench/evals/scr_and_tpp/eval_output.py
@@ -71,6 +71,7 @@ class ScrMetrics(BaseMetrics):
         None,
         title="SCR Metric, Top 20 SAE latents",
         description="SCR Metric (selecting dir1 if inital profession accuracy is lower than initial gender accuracy, else dir2) ablating the top 20 SAE latents",
+        json_schema_extra=DEFAULT_DISPLAY,
     )
     scr_dir2_threshold_20: float | None = Field(
         None,
@@ -324,6 +325,7 @@ class TppMetrics(BaseMetrics):
         None,
         title="TPP Metric, Top 20 SAE latents",
         description="TPP metric when ablating the top 20 SAE latents",
+        json_schema_extra=DEFAULT_DISPLAY,
     )
     tpp_threshold_20_intended_diff_only: float | None = Field(
         None,

diff --git a/sae_bench/evals/scr_and_tpp/eval_output_schema_scr.json b/sae_bench/evals/scr_and_tpp/eval_output_schema_scr.json
@@ -298,7 +298,8 @@
           ],
           "default": null,
           "description": "SCR Metric (selecting dir1 if inital profession accuracy is lower than initial gender accuracy, else dir2) ablating the top 20 SAE latents",
-          "title": "SCR Metric, Top 20 SAE latents"
+          "title": "SCR Metric, Top 20 SAE latents",
+          "ui_default_display": true
         },
         "scr_dir2_threshold_20": {
           "anyOf": [

diff --git a/sae_bench/evals/scr_and_tpp/eval_output_schema_tpp.json b/sae_bench/evals/scr_and_tpp/eval_output_schema_tpp.json
@@ -285,7 +285,8 @@
           ],
           "default": null,
           "description": "TPP metric when ablating the top 20 SAE latents",
-          "title": "TPP Metric, Top 20 SAE latents"
+          "title": "TPP Metric, Top 20 SAE latents",
+          "ui_default_display": true
         },
         "tpp_threshold_20_intended_diff_only": {
           "anyOf": [