diff --git a/vllm/v1/attention/backends/flashinfer.py b/vllm/v1/attention/backends/flashinfer.py
index 9892c360d3d6..1391dc5999a6 100755
--- a/vllm/v1/attention/backends/flashinfer.py
+++ b/vllm/v1/attention/backends/flashinfer.py
@@ -416,7 +416,7 @@ class TRTLLMPrefill:
 
     max_q_len: int
     """
-    The maximum query length *among prefill requests*. 
+    The maximum query length *among prefill requests*.
     """
 
     max_seq_len: int
@@ -1386,7 +1386,9 @@ def forward(
             else:
                 assert isinstance(attn_metadata.prefill, TRTLLMPrefill)
                 # prefill_query may be non-contiguous
-                prefill_query = prefill_query.contiguous()
+                # Use .reshape() to guarantee canonical strides for
+                # is_strictly_contiguous assertion
+                prefill_query = prefill_query.contiguous().reshape(prefill_query.shape)
                 workspace_buffer = _get_trtllm_gen_workspace_buffer()
                 block_tables_prefill = attn_metadata.prefill.block_tables
                 seq_lens_prefill = attn_metadata.prefill.seq_lens
@@ -1497,7 +1499,7 @@ def forward(
             else:
                 # decode_query may be non-contiguous
                 assert isinstance(attn_metadata.decode, TRTLLMDecode)
-                decode_query = decode_query.contiguous()
+                decode_query = decode_query.contiguous().reshape(decode_query.shape)
                 workspace_buffer = _get_trtllm_gen_workspace_buffer()
                 block_tables_decode = attn_metadata.decode.block_tables
                 seq_lens_decode = attn_metadata.decode.seq_lens