华为计算微信公众号
昇腾AI开发者公众号
华为计算微博
华为计算今日头条
如题,8卡910B(硬件id为d801)部署Deepseek-R1-Distill-Llama-70B,在benchmark时性能很差,基本只有1-2token/s,用npu-smi查看占用率只有个位数。如图:
config:
{ "Version": "1.0.0", "LogConfig" : { "logLevel" : "Info", "logFileSize" : 20, "logFileNum" : 20, "logPath" : "logs/mindservice.log" }, "ServerConfig" : { "ipAddress" : "127.0.0.1", "port" : 1025, "managementPort" : 1026, "metricsPort" : 1027, "maxLinkNum" : 1000, "httpsEnabled" : false, "fullTextEnabled" : false, "tlsCaPath" : "security/ca/", "tlsCaFile" : ["ca.pem"], "tlsCert" : "security/certs/server.pem", "tlsPk" : "security/keys/server.key.pem", "tlsPkPwd" : "security/pass/key_pwd.txt", "tlsCrl" : "security/certs/server_crl.pem", "managementTlsCaFile" : ["management_ca.pem"], "managementTlsCert" : "security/certs/management/server.pem", "managementTlsPk" : "security/keys/management/server.key.pem", "managementTlsPkPwd" : "security/pass/management/key_pwd.txt", "managementTlsCrl" : "security/certs/management/server_crl.pem", "kmcKsfMaster" : "tools/pmt/master/ksfa", "kmcKsfStandby" : "tools/pmt/standby/ksfb", "inferMode" : "standard", "interCommTLSEnabled": false, "interCommPort": 1121, "interCommTlsCaFile" : "security/grpc/ca/ca.pem", "interCommTlsCert" : "security/grpc/certs/server.pem", "interCommPk" : "security/grpc/keys/server.key.pem", "interCommPkPwd" : "security/grpc/pass/key_pwd.txt", "interCommTlsCrl" : "security/certs/server_crl.pem" }, "BackendConfig": { "backendName" : "mindieservice_llm_engine", "modelInstanceNumber" : 1, "npuDeviceIds" : [[0,1,2,3,4,5,6,7]], "tokenizerProcessNumber" : 8, "multiNodesInferEnabled": false, "multiNodesInferPort": 1120, "interNodeTLSEnabled": true, "interNodeTlsCaFile": "security/grpc/ca/ca.pem", "interNodeTlsCert": "security/grpc/certs/server.pem", "interNodeTlsPk": "security/grpc/keys/server.key.pem", "interNodeTlsPkPwd": "security/grpc/pass/mindie_server_key_pwd.txt", "interNodeTlsCrl" : "security/grpc/certs/server_crl.pem", "interNodeKmcKsfMaster": "tools/pmt/master/ksfa", "interNodeKmcKsfStandby": "tools/pmt/standby/ksfb", "ModelDeployConfig": { "maxSeqLen" : 2560, "maxInputTokenLen" : 2048, "truncation" : false, "ModelConfig" : [ { "modelInstanceType": "Standard", "modelName" : "ds-llama-70b", "modelWeightPath" : "/path-to-weights/DeepSeek-R1-Distill-Llama-70B", "worldSize" : 8, "cpuMemSize" : 5, "npuMemSize" : 12, "backendType": "atb" } ] }, "ScheduleConfig": { "templateType": "Standard", "templateName" : "Standard_LLM", "cacheBlockSize" : 128, "maxPrefillBatchSize" : 300, "maxPrefillTokens" : 8192, "prefillTimeMsPerReq" : 150, "prefillPolicyType" : 0, "decodeTimeMsPerReq" : 50, "decodePolicyType" : 0, "maxBatchSize" : 600, "maxIterTimes" : 409600, "maxPreemptCount" : 0, "supportSelectBatch" : false, "maxQueueDelayMicroseconds" : 5000 } } }
模型已经从bf16改为float16,性能异常和这个有关吗?
我要发帖子
如题,8卡910B(硬件id为d801)部署Deepseek-R1-Distill-Llama-70B,在benchmark时性能很差,基本只有1-2token/s,用npu-smi查看占用率只有个位数。如图:
config:
{ "Version": "1.0.0", "LogConfig" : { "logLevel" : "Info", "logFileSize" : 20, "logFileNum" : 20, "logPath" : "logs/mindservice.log" }, "ServerConfig" : { "ipAddress" : "127.0.0.1", "port" : 1025, "managementPort" : 1026, "metricsPort" : 1027, "maxLinkNum" : 1000, "httpsEnabled" : false, "fullTextEnabled" : false, "tlsCaPath" : "security/ca/", "tlsCaFile" : ["ca.pem"], "tlsCert" : "security/certs/server.pem", "tlsPk" : "security/keys/server.key.pem", "tlsPkPwd" : "security/pass/key_pwd.txt", "tlsCrl" : "security/certs/server_crl.pem", "managementTlsCaFile" : ["management_ca.pem"], "managementTlsCert" : "security/certs/management/server.pem", "managementTlsPk" : "security/keys/management/server.key.pem", "managementTlsPkPwd" : "security/pass/management/key_pwd.txt", "managementTlsCrl" : "security/certs/management/server_crl.pem", "kmcKsfMaster" : "tools/pmt/master/ksfa", "kmcKsfStandby" : "tools/pmt/standby/ksfb", "inferMode" : "standard", "interCommTLSEnabled": false, "interCommPort": 1121, "interCommTlsCaFile" : "security/grpc/ca/ca.pem", "interCommTlsCert" : "security/grpc/certs/server.pem", "interCommPk" : "security/grpc/keys/server.key.pem", "interCommPkPwd" : "security/grpc/pass/key_pwd.txt", "interCommTlsCrl" : "security/certs/server_crl.pem" }, "BackendConfig": { "backendName" : "mindieservice_llm_engine", "modelInstanceNumber" : 1, "npuDeviceIds" : [[0,1,2,3,4,5,6,7]], "tokenizerProcessNumber" : 8, "multiNodesInferEnabled": false, "multiNodesInferPort": 1120, "interNodeTLSEnabled": true, "interNodeTlsCaFile": "security/grpc/ca/ca.pem", "interNodeTlsCert": "security/grpc/certs/server.pem", "interNodeTlsPk": "security/grpc/keys/server.key.pem", "interNodeTlsPkPwd": "security/grpc/pass/mindie_server_key_pwd.txt", "interNodeTlsCrl" : "security/grpc/certs/server_crl.pem", "interNodeKmcKsfMaster": "tools/pmt/master/ksfa", "interNodeKmcKsfStandby": "tools/pmt/standby/ksfb", "ModelDeployConfig": { "maxSeqLen" : 2560, "maxInputTokenLen" : 2048, "truncation" : false, "ModelConfig" : [ { "modelInstanceType": "Standard", "modelName" : "ds-llama-70b", "modelWeightPath" : "/path-to-weights/DeepSeek-R1-Distill-Llama-70B", "worldSize" : 8, "cpuMemSize" : 5, "npuMemSize" : 12, "backendType": "atb" } ] }, "ScheduleConfig": { "templateType": "Standard", "templateName" : "Standard_LLM", "cacheBlockSize" : 128, "maxPrefillBatchSize" : 300, "maxPrefillTokens" : 8192, "prefillTimeMsPerReq" : 150, "prefillPolicyType" : 0, "decodeTimeMsPerReq" : 50, "decodePolicyType" : 0, "maxBatchSize" : 600, "maxIterTimes" : 409600, "maxPreemptCount" : 0, "supportSelectBatch" : false, "maxQueueDelayMicroseconds" : 5000 } } }模型已经从bf16改为float16,性能异常和这个有关吗?