diff --git a/python/sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py b/python/sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py index 9fdadf6ac..68f1d658e 100644 --- a/python/sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py +++ b/python/sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py @@ -107,7 +107,7 @@ class LMCRadixCache(RadixCache): "v_buffer", getattr(self.token_to_kv_pool_allocator._kvcache, "v_buffer"), ), - tp_group=tp_group, + tp_group=tp_group.device_group if tp_group is not None else None, ) self.load_stream = torch.cuda.Stream()