Fix attention ops interface error

Signed-off-by: Xiao Yu <[email protected]>

Author: amd-xiaoyu12

vllm/_custom_ops.py

@@ -117,8 +117,9 @@ def paged_attention_rocm(
     k_scale: torch.Tensor,
     v_scale: torch.Tensor,
     fp8_out_scale: Optional[torch.Tensor] = None,
-    mfma_type: str = envs.is_set("VLLM_USE_FP8_MFMA") ? "fp8" : "f16",
+    mfma_type: str = "fp8" if envs.is_set("VLLM_USE_FP8_MFMA") else "f16",
 ) -> None:
+    print(f"\nmfma_type: {mfma_type}")
     torch.ops._rocm_C.paged_attention(out, exp_sum, max_logits, tmp_out, query,
                                       key_cache, value_cache, num_kv_heads,
                                       scale, block_tables, seq_lens,