diff --git a/python/sglang/srt/layers/layernorm.py b/python/sglang/srt/layers/layernorm.py index 9f4c77a76..34b75835d 100644 --- a/python/sglang/srt/layers/layernorm.py +++ b/python/sglang/srt/layers/layernorm.py @@ -106,6 +106,8 @@ class RMSNorm(MultiPlatformOp): residual: Optional[torch.Tensor] = None, **kwargs, ) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]: + if x.numel() == 0: + return x if self.variance_size_override is not None: return self.forward_native(x, residual, **kwargs) if is_batch_invariant_mode_enabled():