chore: upgrade sgl-kernel v0.1.8.post2 (#7186)

Co-authored-by: zhyncs <me@zhyncs.com>
2025-06-15 09:26:18 +08:00
parent 55561e2553
commit ed89837cf4
4 changed files with 8 additions and 3 deletions
--- a/python/sglang/srt/layers/attention/cutlass_mla_backend.py
+++ b/python/sglang/srt/layers/attention/cutlass_mla_backend.py
@@ -280,6 +280,7 @@ class CutlassMLABackend(FlashInferMLAAttnBackend):
            seq_lens=forward_batch.seq_lens.to(torch.int32),
            page_table=self.forward_metadata.block_kv_indices,
            workspace=self.forward_metadata.workspace,
+            sm_scale=layer.scaling,
            num_kv_splits=1,
        )