[https://nvbugs/5474119][fix] Cherry-pick #8809 (#8847)

dongfengy · web-flow · commit f5575a91467d · 2025-11-02T16:44:06.000-08:00
Signed-off-by: Dongfeng Yu &lt;dongfengy@nvidia.com&gt;
diff --git a/tests/unittest/_torch/thop/serial/test_moe.py b/tests/unittest/_torch/thop/serial/test_moe.py
@@ -1779,7 +1779,7 @@ def test_moe_fp8_per_tensor_scale(num_tokens, expert_info, hidden_size,
     reason="The kernel only supports Blackwell. Current SM is %d." %
     getSMVersion(),
 )
-@pytest.mark.parametrize("num_tokens", [1, 1024])
+@pytest.mark.parametrize("num_tokens", [1, 256, 1024])
 @pytest.mark.parametrize("hidden_size", [512])
 @pytest.mark.parametrize("intermediate_size", [512])
 @pytest.mark.parametrize(

Original file line number	Diff line number	Diff line change
`@@ -1779,7 +1779,7 @@ def test_moe_fp8_per_tensor_scale(num_tokens, expert_info, hidden_size,`
`1779`	`1779`	`reason="The kernel only supports Blackwell. Current SM is %d." %`
`1780`	`1780`	`getSMVersion(),`
`1781`	`1781`	`)`
`1782`		`-@pytest.mark.parametrize("num_tokens", [1, 1024])`
	`1782`	`+@pytest.mark.parametrize("num_tokens", [1, 256, 1024])`
`1783`	`1783`	`@pytest.mark.parametrize("hidden_size", [512])`
`1784`	`1784`	`@pytest.mark.parametrize("intermediate_size", [512])`
`1785`	`1785`	`@pytest.mark.parametrize(`