[Feat]support sequence parallelism by pass for VL models (#5632)

2026-02-27 08:27:41 +08:00
parent ed175d6d92
commit 5def28dcd3
22 changed files with 460 additions and 101 deletions
--- a/tests/ut/ops/test_mla.py
+++ b/tests/ut/ops/test_mla.py
@@ -157,7 +157,7 @@ class TestAscendMultiHeadLatentAttention(TestBase):
        hidden_states = torch.randn(3, self.hidden_size)

        mock_forward_context = MagicMock(spec=ForwardContext)
-        mock_forward_context.sp_enabled = False
+        mock_forward_context.flash_comm_v1_enabled = False
        mock_get_forward_context.return_value = mock_forward_context

        mock_mla_forward.return_value = (3, self.hidden_size)