From 1e0b33d1369a1f70ed4705c77021e38cc3d1a994 Mon Sep 17 00:00:00 2001 From: degenfabian Date: Tue, 19 Aug 2025 02:48:19 +0200 Subject: [PATCH 1/2] updating loading in santa coder demo to use transformer bridge --- demos/Santa_Coder.ipynb | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-) diff --git a/demos/Santa_Coder.ipynb b/demos/Santa_Coder.ipynb index 0c95abd1d..e607da867 100644 --- a/demos/Santa_Coder.ipynb +++ b/demos/Santa_Coder.ipynb @@ -54,7 +54,7 @@ }, { "cell_type": "code", - "execution_count": 2, + "execution_count": null, "id": "da9f5a40", "metadata": {}, "outputs": [ @@ -97,11 +97,7 @@ "\n", "import transformer_lens\n", "import transformer_lens.utils as utils\n", - "from transformer_lens.hook_points import (\n", - " HookedRootModule,\n", - " HookPoint,\n", - ") # Hooking utilities\n", - "from transformer_lens import HookedTransformer, HookedTransformerConfig, FactoredMatrix, ActivationCache\n", + "from transformer_lens.model_bridge import TransformerBridge\n", "\n", "torch.set_grad_enabled(False)\n", "\n", @@ -132,7 +128,7 @@ }, { "cell_type": "code", - "execution_count": 4, + "execution_count": null, "id": "1f7ac1e1", "metadata": {}, "outputs": [ @@ -154,7 +150,8 @@ "source": [ "# Disable folding norms and folding norms and biases so that intermediate value\n", "# in between transformer blocks can be compared\n", - "bloom = HookedTransformer.from_pretrained(\"bloom-560m\",fold_ln=False, fold_value_biases=False, center_writing_weights=False)" + "bloom = TransformerBridge.boot_transformers(\"bloom-560m\",fold_ln=False, fold_value_biases=False, center_writing_weights=False)\n", + "bloom.enable_compatibility_mode()" ] }, { From 9ea175f858e27b451a04d6d28f61aa8f5a946ca6 Mon Sep 17 00:00:00 2001 From: degenfabian Date: Tue, 19 Aug 2025 18:00:41 +0200 Subject: [PATCH 2/2] add santa coder demo to CI --- .github/workflows/checks.yml | 1 + 1 file changed, 1 insertion(+) diff --git a/.github/workflows/checks.yml b/.github/workflows/checks.yml index 4de51026c..19cc733e3 100644 --- a/.github/workflows/checks.yml +++ b/.github/workflows/checks.yml @@ -159,6 +159,7 @@ jobs: # - "No_Position_Experiment" - "Othello_GPT" - "Patchscopes_Generation_Demo" + - "Santa_Coder" # - "T5" steps: - uses: actions/checkout@v3