2.2RC1 多机启动失败 Invalid DP number per node: 0
收藏回复举报
2.2RC1 多机启动失败 Invalid DP number per node: 0
t('forum.solved') 已解决
发表于2025-11-27 10:39:59
0 查看
{
    "server_count": "2",
    "server_list": [
        {
            "device": [
                {
                    "device_id": "0",
                    "device_ip": "172.20.9.79",
                    "rank_id": "0",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "1",
                    "device_ip": "172.20.9.80",
                    "rank_id": "1",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "2",
                    "device_ip": "172.20.9.81",
                    "rank_id": "2",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "3",
                    "device_ip": "172.20.9.82",
                    "rank_id": "3",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "4",
                    "device_ip": "172.20.9.83",
                    "rank_id": "4",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "5",
                    "device_ip": "172.20.9.84",
                    "rank_id": "5",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "6",
                    "device_ip": "172.20.9.85",
                    "rank_id": "6",
                    "netmask": "255.255.0.0"
                },
                {
                    "device_id": "7",
                    "device_ip": "172.20.9.86",
                    "rank_id": "7",
                    "netmask": "255.255.0.0"
                }
            ],
            "server_id": "172.30.8.36",
            "container_ip": "172.30.8.36"

RANK TABLE 如上 

mindie config如下

{
    "BackendConfig": {
        "ModelDeployConfig": {
            "ModelConfig": [
                {
                    "backendType": "atb",
                    "cpuMemSize": 5,
                    "modelInstanceType": "Standard",
                    "modelName": "models",
                    "modelWeightPath": "/root/bentoml/models/qwen3-235b-a22b-w8a8/edktiagjospzp6xp",
                    "npuMemSize": -1,
                    "trustRemoteCode": true,
                    "worldSize": 8
                }
            ],
            "maxInputTokenLen": 2048,
            "maxSeqLen": 2560,
            "truncation": false
        },
        "ScheduleConfig": {
            "cacheBlockSize": 128,
            "decodePolicyType": 0,
            "decodeTimeMsPerReq": 50,
            "maxBatchSize": 200,
            "maxIterTimes": 512,
            "maxPreemptCount": 0,
            "maxPrefillBatchSize": 50,
            "maxPrefillTokens": 8192,
            "maxQueueDelayMicroseconds": 5000,
            "prefillPolicyType": 0,
            "prefillTimeMsPerReq": 150,
            "supportSelectBatch": false,
            "templateName": "Standard_LLM",
            "templateType": "Standard"
        },
        "backendName": "mindieservice_llm_engine",
        "interNodeKmcKsfMaster": "tools/pmt/master/ksfa",
        "interNodeKmcKsfStandby": "tools/pmt/standby/ksfb",
        "interNodeTLSEnabled": false,
        "interNodeTlsCaFiles": [
            "ca.pem"
        ],
        "interNodeTlsCaPath": "security/grpc/ca/",
        "interNodeTlsCert": "security/grpc/certs/server.pem",
        "interNodeTlsCrlFiles": [
            "server_crl.pem"
        ],
        "interNodeTlsCrlPath": "security/grpc/certs/",
        "interNodeTlsPk": "security/grpc/keys/server.key.pem",
        "interNodeTlsPkPwd": "security/grpc/pass/mindie_server_key_pwd.txt",
        "modelInstanceNumber": 1,
        "multiNodesInferEnabled": true,
        "multiNodesInferPort": 1120,
        "npuDeviceIds": [
            [
                0,
                1,
                2,
                3,
                4,
                5,
                6,
                7
            ]
        ],
        "tokenizerProcessNumber": 8
    },
    "LogConfig": {
        "dynamicLogLevel": "",
        "dynamicLogLevelValidHours": 2,
        "dynamicLogLevelValidTime": ""
    },
    "ServerConfig": {
        "allowAllZeroIpListening": true,
        "distDPServerEnabled": false,
        "e2eTimeout": 600,
        "fullTextEnabled": false,
        "httpsEnabled": false,
        "inferMode": "standard",
        "interCommPk": "security/grpc/keys/server.key.pem",
        "interCommPkPwd": "security/grpc/pass/key_pwd.txt",
        "interCommPort": 1121,
        "interCommTLSEnabled": true,
        "interCommTlsCaFiles": [
            "ca.pem"
        ],
        "interCommTlsCaPath": "security/grpc/ca/",
        "interCommTlsCert": "security/grpc/certs/server.pem",
        "interCommTlsCrlFiles": [
            "server_crl.pem"
        ],
        "interCommTlsCrlPath": "security/grpc/certs/",
        "ipAddress": "0.0.0.0",
        "kmcKsfMaster": "tools/pmt/master/ksfa",
        "kmcKsfStandby": "tools/pmt/standby/ksfb",
        "managementIpAddress": "127.0.0.2",
        "managementPort": 1026,
        "managementTlsCaFile": [
            "management_ca.pem"
        ],
        "managementTlsCert": "security/certs/management/server.pem",
        "managementTlsCrlFiles": [
            "server_crl.pem"
        ],
        "managementTlsCrlPath": "security/management/certs/",
        "managementTlsPk": "security/keys/management/server.key.pem",
        "managementTlsPkPwd": "security/pass/management/key_pwd.txt",
        "maxLinkNum": 1000,
        "metricsPort": 1027,
        "openAiSupport": "vllm",
        "port": 3001,
        "tlsCaFile": [
            "ca.pem"
        ],
        "tlsCaPath": "security/ca/",
        "tlsCert": "security/certs/server.pem",
        "tlsCrlFiles": [
            "server_crl.pem"
        ],
        "tlsCrlPath": "security/certs/",
        "tlsPk": "security/keys/server.key.pem",
        "tlsPkPwd": "security/pass/key_pwd.txt",
        "tokenTimeout": 600
    },
    "Version": "1.0.0"
}

使用的性能调优参数

ATB_LLM_ENABLE_AUTO_TRANSPOSE=0

ATB_WORKSPACE_MEM_ALLOC_GLOBAL=1

MINDIE_LOG_TO_STDOUT=1

ATB_WORKSPACE_MEM_ALLOC_ALG_TYPE=3

PYTORCH_NPU_ALLOC_CONF=expandable_segments:True

MINDIE_ASYNC_SCHEDULING_ENABLE=1

NPU_MEMORY_FRACTION=0.96

CPU_AFFINITY_CONF=1

TASK_QUEUE_ENABLE=2

报错截图:

cke_164034.png

后续尝试 ModelConfig": [ { "backendType": "atb", "cpuMemSize": 5, "modelInstanceType": "Standard", "modelName": "models", "modelWeightPath": "/root/bentoml/models/qwen3-235b-a22b-w8a8/edktiagjospzp6xp", "npuMemSize": -1, "trustRemoteCode": true, "worldSize": 8, "dp": 2, "tp": 8, "moe_tp": 1, "moe_ep": 16}

遇到的问题 两个节点的报错也不一致

cke_17851.png

cke_17852.png

本帖最后由 匿名用户2025/11/28 13:08:20 编辑

我要发帖子