diff --git a/custom_ops/gpu_ops/w4afp8_gemm/w4afp8_gemm.cu b/custom_ops/gpu_ops/w4afp8_gemm/w4afp8_gemm.cu
index 4ff79ad854..d0a6bfeb18 100644
--- a/custom_ops/gpu_ops/w4afp8_gemm/w4afp8_gemm.cu
+++ b/custom_ops/gpu_ops/w4afp8_gemm/w4afp8_gemm.cu
@@ -77,16 +77,17 @@ void DisPatchW4AFp8Gemm(
             max_tokens,
             stream)
     } else {
-        GEMM_SWITCH_FP16(
-            M, K, batch_size, token_padding_size, kBlockN, TailN,
-            weight,
-            input,
-            out,
-            weight_scale,
-            input_row_sum,
-            tokens,
-            max_tokens,
-            stream)
+        PD_THROW("Only supported dtype in ['FP16'].");
+        // GEMM_SWITCH_FP16(
+        //     M, K, batch_size, token_padding_size, kBlockN, TailN,
+        //     weight,
+        //     input,
+        //     out,
+        //     weight_scale,
+        //     input_row_sum,
+        //     tokens,
+        //     max_tokens,
+        //     stream)
     }
 }
 
diff --git a/custom_ops/utils/auto_gen_w4afp8_gemm_kernel.py b/custom_ops/utils/auto_gen_w4afp8_gemm_kernel.py
index 87b06fa747..1798263a27 100644
--- a/custom_ops/utils/auto_gen_w4afp8_gemm_kernel.py
+++ b/custom_ops/utils/auto_gen_w4afp8_gemm_kernel.py
@@ -85,12 +85,12 @@
 
 gemm_case = [
     [8192, 3584, 8, 0],  # eb45T ffn1
-    [8192, 3584, 8, 2048],  # eb45T ffn1
-    [7168, 8192, 8, 0],  # eb45T ffn2
-    [7168, 8192, 8, 2048],  # eb45T ffn2
+    # [8192, 3584, 8, 2048],  # eb45T ffn1
+    # [7168, 8192, 8, 0],  # eb45T ffn2
+    # [7168, 8192, 8, 2048],  # eb45T ffn2
 ]
 
-dtype = ["BF16", "FP16"]
+dtype = ["BF16"]
 
 
 def get_cutlass_type(type):