diff --git a/python/sglang/srt/layers/attention/trtllm_mha_backend.py b/python/sglang/srt/layers/attention/trtllm_mha_backend.py
index 548c1c389..cefed569a 100644
--- a/python/sglang/srt/layers/attention/trtllm_mha_backend.py
+++ b/python/sglang/srt/layers/attention/trtllm_mha_backend.py
@@ -973,7 +973,10 @@ class TRTLLMHAAttnBackend(FlashInferAttnBackend):
             and not use_fused_qkv
         ):
             q = q.to(torch.float8_e4m3fn)
-        q = q.reshape(-1, layer.tp_q_head_num, layer.head_dim)
+        if self.is_xqa_impl:
+            q = q.contiguous().view(-1, layer.tp_q_head_num, layer.head_dim)
+        else:
+            q = q.reshape(-1, layer.tp_q_head_num, layer.head_dim)
 
         if self.is_nvfp4_kvcache:
             kv_cache, kv_cache_block_scales = self._get_nvfp4_decode_kv_cache(layer)
diff --git a/test/registered/quant/test_llama8b_nvfp4_kv_cache_sm120.py b/test/registered/quant/test_llama8b_nvfp4_kv_cache_sm120.py
index cbbf6349e..62addea32 100644
--- a/test/registered/quant/test_llama8b_nvfp4_kv_cache_sm120.py
+++ b/test/registered/quant/test_llama8b_nvfp4_kv_cache_sm120.py
@@ -9,7 +9,6 @@ register_cuda_ci(
     est_time=300,
     stage="extra-a",
     runner_config="1-gpu-small",
-    disabled="Temporarily disabled due to failing accuracy",
 )
 
 
@@ -31,19 +30,12 @@ class TestLlama8BNVFP4KVCacheSM120(GSM8KMixin, DefaultServerBase):
     other_args = [
         "--quantization",
         "modelopt_fp4",
-        "--fp4-gemm-backend",
-        "auto",
         "--kv-cache-dtype",
         "nvfp4",
         "--prefill-attention-backend",
         "flashinfer",
         "--decode-attention-backend",
         "trtllm_mha",
-        "--page-size",
-        "64",
-        "--mem-fraction-static",
-        "0.87",
-        "--cuda-graph-backend-prefill=disabled",
     ]