[NPU] Use use_dsa to dispatch Ascend DSA attention (#28436)

This commit is contained in:
Peng Xingchen
2026-06-17 15:47:12 +08:00
committed by GitHub
parent b54f8432ad
commit d0e974f40b
@@ -45,12 +45,12 @@ def handle_attention_ascend(attn, forward_batch):
and not forward_batch.forward_mode.is_target_verify() and not forward_batch.forward_mode.is_target_verify()
and not forward_batch.forward_mode.is_draft_extend_v2() and not forward_batch.forward_mode.is_draft_extend_v2()
): ):
if hasattr(attn, "indexer"): if hasattr(attn, "use_dsa") and attn.use_dsa:
return AttnForwardMethod.DSA_NPU return AttnForwardMethod.DSA_NPU
else: else:
return AttnForwardMethod.MHA_NPU return AttnForwardMethod.MHA_NPU
else: else:
if hasattr(attn, "indexer"): if hasattr(attn, "use_dsa") and attn.use_dsa:
return AttnForwardMethod.DSA_NPU return AttnForwardMethod.DSA_NPU
else: else:
return AttnForwardMethod.MLA_NPU return AttnForwardMethod.MLA_NPU