diff --git a/logix/huggingface/arguments.py b/logix/huggingface/arguments.py
index e888082..10d41d9 100644
--- a/logix/huggingface/arguments.py
+++ b/logix/huggingface/arguments.py
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 from dataclasses import dataclass, field
-from typing import List
+from typing import List, Optional
 
 import torch.nn as nn
 
@@ -47,12 +47,15 @@ class LogIXArguments:
     input_key: str = field(
         default="input_ids", metadata={"help": "The dictionary key for 'input_ids'."}
     )
-    influence_damping: float = field(
+    influence_damping: Optional[float] = field(
         default=None, metadata={"help": "A damping term in influence functions."}
     )
     influence_mode: str = field(
         default="dot", metadata={"help": "Influence function mode."}
     )
+    influence_groups: Optional[List[str]] = field(
+        default=None, metadata={"help": "Influence function groups."}
+    )
     label_key: str = field(
         default="labels", metadata={"help": "The dictionary key for 'labels'."}
     )
diff --git a/logix/huggingface/callback.py b/logix/huggingface/callback.py
index 959d00e..e7ecc56 100644
--- a/logix/huggingface/callback.py
+++ b/logix/huggingface/callback.py
@@ -74,6 +74,7 @@ def on_step_end(self, args, state, control, **kwargs):
                 self.log_dataloader(),
                 mode=self.args.influence_mode,
                 damping=self.args.influence_damping,
+                influence_groups=self.args.influence_groups,
                 save=True,
             )
 
@@ -83,7 +84,9 @@ def on_step_end(self, args, state, control, **kwargs):
             accumulated_log = merge_logs(self.accumulated_log)
 
             self.logix.influence.compute_self_influence(
-                accumulated_log, damping=self.args.influence_damping
+                accumulated_log,
+                damping=self.args.influence_damping,
+                influence_groups=self.args.influence_groups,
             )
 
             self.accumulated_log = []