[BugFix] fix prefixcache performance and accuracy on ascend (#13573)

2025-12-08 02:16:20 +08:00
parent f124539a01
commit 948b6acee8
5 changed files with 289 additions and 28 deletions
--- a/python/sglang/srt/managers/cache_controller.py
+++ b/python/sglang/srt/managers/cache_controller.py
@@ -503,7 +503,7 @@ class HiCacheController:
            elif self.mem_pool_host.layout == "page_first_direct":
                return host_indices, device_indices.cpu()
        elif self.io_backend == "kernel_ascend":
-            return host_indices, device_indices
+            return host_indices, device_indices.cpu()
        else:
            raise ValueError(f"Unsupported io backend")