diff --git a/docker/rocm.Dockerfile b/docker/rocm.Dockerfile
index 2e3e5ef1b..11df166a9 100644
--- a/docker/rocm.Dockerfile
+++ b/docker/rocm.Dockerfile
@@ -98,7 +98,7 @@ ARG ENABLE_MORI=0
 ARG NIC_BACKEND=none
 
 ARG MORI_REPO="https://github.com/ROCm/mori.git"
-ARG MORI_COMMIT="20920706a9004018dbd87c7387f207d08d0e05af"
+ARG MORI_COMMIT="2f88d06aba75400262ca5c1ca5986cf1fdf4cd82"
 
 # AMD AINIC apt repo settings
 ARG AINIC_VERSION=1.117.5
diff --git a/python/sglang/srt/layers/moe/token_dispatcher/moriep.py b/python/sglang/srt/layers/moe/token_dispatcher/moriep.py
index a0ea4656a..6c189d0f7 100644
--- a/python/sglang/srt/layers/moe/token_dispatcher/moriep.py
+++ b/python/sglang/srt/layers/moe/token_dispatcher/moriep.py
@@ -189,6 +189,8 @@ def init_mori_op(
     world_size = get_moe_expert_parallel_world_size()
     rank = get_moe_expert_parallel_rank()
 
+    gpu_per_node = 8 if world_size >= 8 else world_size
+
     cpu_group = group.cpu_group
     torch._C._distributed_c10d._register_process_group("mori", cpu_group)
     mori.shmem.shmem_torch_process_group_init("mori")
@@ -227,6 +229,7 @@ def init_mori_op(
         warp_num_per_block=warp_num_per_block,
         block_num=block_num,
         kernel_type=kernel_type,
+        gpu_per_node=gpu_per_node,
         rdma_block_num=rdma_block_num,
         num_qp_per_pe=2,
     )