Reduce batch size

pomonam · Mar 20, 2024 · f47fc90 · f47fc90
1 parent 11eb42a
commit f47fc90
Show file tree

Hide file tree

Showing 2 changed files with 24 additions and 5 deletions.
diff --git a/examples/cifar/analyze.py b/examples/cifar/analyze.py
@@ -125,7 +125,6 @@ def main():
         per_device_batch_size=None,
         factor_args=factor_args,
         overwrite_output_dir=True,
-        initial_per_device_batch_size_attempt=8192,
     )
     analyzer.compute_pairwise_scores(
         scores_name="pairwise",
@@ -135,7 +134,6 @@ def main():
         train_dataset=train_dataset,
         per_device_query_batch_size=500,
         overwrite_output_dir=True,
-        initial_per_device_train_batch_size_attempt=8192,
     )
     scores = analyzer.load_pairwise_scores("pairwise")
     print(scores)

diff --git a/examples/uci/tutorial.ipynb b/examples/uci/tutorial.ipynb
@@ -1241,7 +1241,7 @@
    },
    "outputs": [],
    "source": [
-    "random_indices = list(range(summed_scores.shape[0]))\n",
+    "random_indices = list(range(len(train_dataset)))\n",
     "shuffle(random_indices)\n",
     "random_removed_loss_lst = []\n",
     "\n",
@@ -1257,10 +1257,23 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 41,
    "id": "9e6fd656-c1e5-42cb-87f4-d7690f473fe9",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Computing pairwise scores (query gradient) [0/1]   0%|                                                                                                       [time left: ?, time spent: 00:00]\n",
+      "Computing pairwise scores (training gradient) [1/1] 100%|███████████████████████████████████████████████████████████████████████████████████████████████ [time left: 00:00, time spent: 00:00]\u001b[A\n",
+      "Computing pairwise scores (query gradient) [1/1] 100%|██████████████████████████████████████████████████████████████████████████████████████████████████ [time left: 00:00, time spent: 00:00]\n",
+      "Computing pairwise scores (query gradient) [0/1]   0%|                                                                                                       [time left: ?, time spent: 00:00]\n",
+      "Computing pairwise scores (training gradient) [1/1] 100%|███████████████████████████████████████████████████████████████████████████████████████████████ [time left: 00:00, time spent: 00:00]\u001b[A\n",
+      "Computing pairwise scores (query gradient) [1/1] 100%|██████████████████████████████████████████████████████████████████████████████████████████████████ [time left: 00:00, time spent: 00:00]\n"
+     ]
+    }
+   ],
    "source": [
     "from kronfluence import FactorArguments\n",
     "\n",
@@ -1317,6 +1330,14 @@
     "plt.ylabel(\"Query Loss\")\n",
     "plt.xlabel(\"Number of Training Samples Removed\")"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f0f7d37b-4d80-480c-ab0b-a9fa0aab2096",
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {