使用NPU直连方式推理DeepSeek-R1-W8A8无法启动
收藏回复举报
使用NPU直连方式推理DeepSeek-R1-W8A8无法启动
t('forum.solved') 已解决
新人帖
发表于2025-02-26 00:08:46
0 查看

使用NPU直连方式推理DeepSeek-R1-W8A8无法启动

  使用两台800I服务器启动MindIE服务,但是没有启动成功,想问一下问题出现的原因与解决的方式,望大佬们指点:)

连接方式

  1. 使用2台800I服务器,NPU对应直连
  2. 2台800I使用一根网线直连
  3. 使用iBMC进行开发
    大概是这样: true

报错日志

  1. 主节点
2025-02-25 22:01:05.175478 37332 LLM log default format: [yyyy-mm-dd hh:mm:ss.uuuuuu][processid] [threadid] [llm] [loglevel][file:line] [error code] msg
[2025-02-25 22:01:05.175525] [37212] [281472602534240] [llm] [INFO] [post_processing_manager.cpp:146] Get post processing manager
[2025-02-25 22:01:05.175568] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471393788256
[2025-02-25 22:01:05.175608] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469465325920
[2025-02-25 22:01:05.175637] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469421023584
[2025-02-25 22:01:05.175663] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469339562336
[2025-02-25 22:01:05.175694] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469331108192
[2025-02-25 22:01:05.175734] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469322654048
[2025-02-25 22:01:05.175767] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469314199904
[2025-02-25 22:01:05.175797] [37212] [281472602534240] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281469305745760
[2025-02-25 22:02:55.084+08:00] [37184] [37185] [mindie-server] [INFO] [master_GRPC_communicator.cpp:764] : [model_backend] Waiting to receive all register msg from all slave nodes.
[2025-02-25 22:02:55.084+08:00] [37184] [37185] [mindie-server] [ERROR] [block_num_waiter.cpp:67] : [MIE04E120507] [model_backend] The received node [192.168.0.101] status is error
[2025-02-25 22:03:00.084+08:00] [37184] [37185] [mindie-server] [ERROR] [block_num_waiter.cpp:67] : [MIE04E120507] [model_backend] The received node [192.168.0.101] status is error
[2025-02-25 22:03:05.084+08:00] [37184] [37185] [mindie-server] [ERROR] [block_num_waiter.cpp:67] : [MIE04E120507] [model_backend] The received node [192.168.0.101] status is error
[2025-02-25 22:03:10.085+08:00] [37184] [37185] [mindie-server] [ERROR] [block_num_waiter.cpp:67] : [MIE04E120507] [model_backend] The received node [192.168.0.101] status is error
[2025-02-25 22:03:15.085+08:00] [37184] [37185] [mindie-server] [ERROR] [block_num_waiter.cpp:67] : [MIE04E120507] [model_backend] The received node [192.168.0.101] status is error
  1. 副节点
[2025-02-25 22:02:55.094+08:00] [24236] [localhost.localdomain] [system] [model_backend] [start backend-master] [success]
[2025-02-25 22:02:55.095460] [24236] [281470815039424] [llm] [INFO] [llm_manager_impl.cpp:549] Multi Nodes inference slave instance need not init GMIS and IBIS.
[2025-02-25 22:02:55.095+08:00] [24236] [27770] [mindie-server] [INFO] [heartbeat_sender.cpp:66] : [model_backend] Create channel to contact receiver
[2025-02-25 22:02:55.095600] [24236] [281470815039424] [llm] [INFO] [llm_manager_impl.cpp:79] LLMRuntime thread start success!
[2025-02-25 22:02:55.095+08:00] [24236] [localhost.localdomain] [system] [server] [start engine(mindieservice_llm_engine)] [success]
[2025-02-25 22:02:55.095+08:00] [24236] [localhost.localdomain] [system] [server] [load modelDeepSeek] [success]
[2025-02-25 22:02:55.095+08:00] [24236] [localhost.localdomain] [system] [server] [start engine] [success]
Multi Nodes infer slave instance need not init TokenizerProcessPool and HttpWrapper
[2025-02-25 22:02:55.096+08:00] [24236] [24237] [mindie-server] [INFO] [endpoint.cpp:105] : [endpoint] Multi Nodes infer slave instance need not init TokenizerProcessPool and HttpWrapper.
[2025-02-25 22:02:55.096+08:00] [24236] [24237] [mindie-server] [INFO] [endpoint.cpp:108] : [endpoint] Start endpoint success
[2025-02-25 22:03:00.096+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:03:05.550+08:00] [24236] [localhost.localdomain] [system] [server] [start endpoint] [success]
Daemon start success!
[2025-02-25 22:03:05.551+08:00] [24236] [24237] [mindie-server] [INFO] [llm_daemon.cpp:147] : [daemon] Daemon start success!
[2025-02-25 22:03:25.098+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:03:50.099+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:04:15.100+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:04:40.100+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:05:05.101+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:05:30.102+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:05:55.104+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:06:20.104+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded
[2025-02-25 22:06:45.105+08:00] [24236] [27770] [mindie-server] [ERROR] [heartbeat_sender.cpp:89] : [MIE04E120514] [model_backend] RPC failed with status code[4], and error message: Deadline Exceeded

我要发帖子