diff --git a/src/axolotl/prompt_strategies/dpo/user_defined.py b/src/axolotl/prompt_strategies/dpo/user_defined.py
index 5ba488ca3d..754b674102 100644
--- a/src/axolotl/prompt_strategies/dpo/user_defined.py
+++ b/src/axolotl/prompt_strategies/dpo/user_defined.py
@@ -4,7 +4,9 @@
 
 
 def default(cfg, dataset_idx=0, **kwargs):
-    ds_cfg = cfg["datasets"][dataset_idx]
+    ds_cfg = cfg["datasets"][dataset_idx]["type"]
+    if not isinstance(ds_cfg, dict):
+        raise ValueError(f"User-defined dataset type must be a dictionary. Got: {ds_cfg}")
     field_prompt = ds_cfg.get("field_prompt", "prompt")
     field_system = ds_cfg.get("field_system", "system")
     field_chosen = ds_cfg.get("field_chosen", "chosen")
diff --git a/src/axolotl/utils/data.py b/src/axolotl/utils/data.py
index 12cff999f6..ad3a5cb2d8 100644
--- a/src/axolotl/utils/data.py
+++ b/src/axolotl/utils/data.py
@@ -937,6 +937,8 @@ def load_split(dataset_cfgs, _cfg):
         for i, data_set in enumerate(split_datasets):
             _type = dataset_cfgs[i]["type"]
             if _type:
+                if isinstance(_type, DictDefault):
+                    _type = "user_defined.default"
                 ds_transform_fn = load_dpo(_type, _cfg, dataset_idx=i)
                 split_datasets[i] = data_set.map(
                     ds_transform_fn,