From ae7428a8a7379b425eff867570276c9c2681c262 Mon Sep 17 00:00:00 2001
From: huangtingwei <141888744+huangtingwei9988@users.noreply.github.com>
Date: Wed, 27 Aug 2025 15:43:16 +0800
Subject: [PATCH] fix mooncake store mla zero copy meta (#9678)

---
 python/sglang/srt/mem_cache/memory_pool_host.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/python/sglang/srt/mem_cache/memory_pool_host.py b/python/sglang/srt/mem_cache/memory_pool_host.py
index 13b707ba7..080ee458d 100644
--- a/python/sglang/srt/mem_cache/memory_pool_host.py
+++ b/python/sglang/srt/mem_cache/memory_pool_host.py
@@ -705,7 +705,6 @@ class MLATokenToKVPoolHost(HostKVCache):
             raise ValueError(f"Unsupported layout: {self.layout}")
 
     def get_buffer_meta(self, keys, indices):
-        local_rank = get_tensor_model_parallel_rank()
         ptr_list = []
         key_list = []
         kv_buffer_data_ptr = self.kv_buffer.data_ptr()
@@ -719,7 +718,7 @@ class MLATokenToKVPoolHost(HostKVCache):
             )
             ptr_list.append(k_ptr)
             key_ = keys[index // self.page_size]
-            key_list.append(f"{key_}_{local_rank}_k")
+            key_list.append(f"{key_}_k")
         element_size = (
             self.layer_num
             * self.dtype.itemsize