From 9f739ac1ff0b2dbaba667bd5f365de3e7fddf9c4 Mon Sep 17 00:00:00 2001
From: Khalil Meftah <khalil.meftah@huggingface.co>
Date: Thu, 23 Apr 2026 13:59:58 +0200
Subject: [PATCH] fix(train): raise ValueError for PEFT usage in reward model
 training

---
 src/lerobot/scripts/lerobot_train.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/lerobot/scripts/lerobot_train.py b/src/lerobot/scripts/lerobot_train.py
index be99b70bb..c8322f238 100644
--- a/src/lerobot/scripts/lerobot_train.py
+++ b/src/lerobot/scripts/lerobot_train.py
@@ -275,6 +275,8 @@ def train(cfg: TrainPipelineConfig, accelerator: "Accelerator | None" = None):
         )
 
     if cfg.peft is not None:
+        if cfg.is_reward_model_training:
+            raise ValueError("PEFT is only supported for policy training. ")
         logging.info("Using PEFT! Wrapping model.")
         peft_cli_overrides = dataclasses.asdict(cfg.peft)
         policy = policy.wrap_with_peft(peft_cli_overrides=peft_cli_overrides)