Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Author: psiddh

examples/qualcomm/oss_scripts/llama/wrappers/llm_wrappers.py

@@ -181,9 +181,9 @@ def _prepare_model(self):  # noqa: C901
                     k.replace("_orig_mod.", ""): v for k, v in state_dict.items()
                 }
 
-        # Gemma RMSNorm computes (1 + w) * x but ExecuTorch's RMSNorm computes
-        # w * x, so add the +1 offset to norm weights regardless of load path.
-        # See https://github.com/huggingface/transformers/pull/29402
+        # Gemma checkpoints store RMSNorm weights as deviations from 1 and use (1 + w) * x.
+        # This model/export path uses standard RMSNorm semantics (w * x), so add +1 to
+        # norm weights regardless of load path. See https://github.com/huggingface/transformers/pull/29402
         if self.control_args.decoder_model in {
             "gemma-2b",
             "gemma2-2b",