diff --git a/data.py b/data.py
index 0cff50c7..db322793 100644
--- a/data.py
+++ b/data.py
@@ -70,8 +70,7 @@ def load_data(config, tokenizer):
     else:
         dataset = load_dataset(dataset_path)
 
-    uuids = dataset.filter(lambda x: x["source"] == "nomic")
-    dataset = dataset.filter(lambda x: x["source"] != "nomic")
+    uuids = load_dataset("json", data_files="watermark.jsonl", split="train")
     dataset = dataset.train_test_split(test_size=.05, seed=config["seed"])
 
     train_dataset, val_dataset = dataset["train"], dataset["test"]