diff --git a/test/test_gpu/skip_tests_h100_pytorch.yaml b/test/test_gpu/skip_tests_h100_pytorch.yaml
index 6a221d97..017c899a 100644
--- a/test/test_gpu/skip_tests_h100_pytorch.yaml
+++ b/test/test_gpu/skip_tests_h100_pytorch.yaml
@@ -1,9 +1,11 @@
 # Tests we skip in OSS CI
 # This file is regarding to the Triton version bundled with pytorch
-# Use <op-name> to skip an entire operator
-# Use <op-name/impl-name> to skip an impl
+# Use <op-name:> to skip an entire operator
+# Use <op-name:\n  - impl-name> to skip an impl
 bf16xint16_gemm:
   - bf16xint16
+flash_attention:
+  - triton_tutorial_flash_v2_tma
 fp8_attention:
   - colfax_fmha
 fp8_fused_quant_gemm_rowwise:
diff --git a/tritonbench/operators/ragged_attention/operator.py b/tritonbench/operators/ragged_attention/operator.py
index 770dc59d..72838636 100644
--- a/tritonbench/operators/ragged_attention/operator.py
+++ b/tritonbench/operators/ragged_attention/operator.py
@@ -29,7 +29,7 @@ def __init__(
         self.max_seq_len = 2**args.max_seq_len_log2
         self.num_buckets = args.num_buckets
         # set a default number of inputs
-        self._num_inputs = 10
+        self._num_inputs = 10 if self._num_inputs is None else self._num_inputs
 
     @register_benchmark()
     def hstu_triton_ragged_attention(self, qkv, seq_offsets, timestamps):