updated loading in attribution patching demo to use transformer bridge

.github/workflows/checks.yml

@@ -223,8 +223,8 @@ jobs:
       fail-fast: false
       matrix:
         notebook:
+          - "Attribution_Patching_Demo"
           - "Activation_Patching_in_TL_Demo"
-          # - "Attribution_Patching_Demo"
           - "ARENA_Content"
           - "BERT"
           - "Exploratory_Analysis_Demo"

demos/Activation_Patching_in_TL_Demo.ipynb

@@ -562,6 +562,8 @@
     }
    ],
    "source": [
+    "# 2,160 forward passes of GPT-2; too slow for CI without GPU\n",
+    "# NBVAL_SKIP\n",
     "ALL_HEAD_LABELS = [f\"L{i}H{j}\" for i in range(model.cfg.n_layers) for j in range(model.cfg.n_heads)]\n",
     "if DO_SLOW_RUNS:\n",
     "    attn_head_out_act_patch_results = patching.get_act_patch_attn_head_out_by_pos(model, corrupted_tokens, clean_cache, ioi_metric)\n",
@@ -1283,4 +1285,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 2
-}
+}