From 173e3495dce60181e889071f13be1703b7b2d4a7 Mon Sep 17 00:00:00 2001
From: Khalil Meftah <khalil.meftah@huggingface.co>
Date: Thu, 23 Apr 2026 13:50:45 +0200
Subject: [PATCH] refactor(wandb): differentiate tags for reward model and
 policy training in cfg_to_group function

---
 src/lerobot/common/wandb_utils.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/src/lerobot/common/wandb_utils.py b/src/lerobot/common/wandb_utils.py
index e3190b6ce..b782cd751 100644
--- a/src/lerobot/common/wandb_utils.py
+++ b/src/lerobot/common/wandb_utils.py
@@ -41,8 +41,12 @@ def cfg_to_group(
             return tag
         return tag[:max_tag_length]
 
+    if cfg.is_reward_model_training:
+        trainable_tag = f"reward_model:{cfg.reward_model.type}"
+    else:
+        trainable_tag = f"policy:{cfg.policy.type}"
     lst = [
-        f"policy:{cfg.policy.type}",
+        trainable_tag,
         f"seed:{cfg.seed}",
     ]
     if cfg.dataset is not None: