[AMD][Fix] Qwen3.5: guard zero-grid launch in fused_qk_gemma_rmsnorm(_with_gate) (HIP invalid configuration on idle DP rank) (#31794)

python/sglang/srt/layers/sampler.py

@@ -45,6 +45,7 @@ if is_musa():
         top_p_renorm_prob,
     )
 
+_is_hip = is_hip()
 _use_aiter = get_bool_env_var("SGLANG_USE_AITER") and is_hip()
 if _use_aiter:
     from aiter import greedy_sample as _aiter_greedy_sample
@@ -119,6 +120,9 @@ class Sampler(nn.Module):
         """
         logits = logits_output.next_token_logits
 
+        if _is_hip and logits.shape[0] == 0:
+            return torch.empty((0,), dtype=torch.int64, device=logits.device)
+
         # Preprocess logits (custom processors and NaN handling)
         logits = self._preprocess_logits(logits, sampling_info)
         return_sampling_mask = any(sampling_info.return_sampling_masks or [])