提交 c984d49d 编写于 作者: X xulongteng

fix trian image

上级 39d65815
...@@ -26,14 +26,14 @@ spec: ...@@ -26,14 +26,14 @@ spec:
type: "" type: ""
name: seqdata name: seqdata
containers: containers:
- image: sivanzcw/edldemo:v1 - image: wangjiawei1993/edldemo:v12
command: command:
- paddle_k8s - paddle_k8s
- start_fluid - start_fluid
imagePullPolicy: IfNotPresent imagePullPolicy: IfNotPresent
name: pserver name: pserver
volumeMounts: volumeMounts:
- mountPath: /mnt/seqdata - mountPath: /data
name: seqdata name: seqdata
resources: resources:
limits: limits:
...@@ -88,7 +88,7 @@ spec: ...@@ -88,7 +88,7 @@ spec:
- name: ENTRY - name: ENTRY
value: cd /workspace/ctr && python train.py --is_local 0 --cloud_train 1 value: cd /workspace/ctr && python train.py --is_local 0 --cloud_train 1
- name: PADDLE_PORT - name: PADDLE_PORT
value: "30236" value: "30240"
- name: LD_LIBRARY_PATH - name: LD_LIBRARY_PATH
value: /usr/local/lib:/usr/local/nvidia/lib64:/usr/local/rdma/lib64:/usr/lib64/mlnx_ofed/valgrind value: /usr/local/lib:/usr/local/nvidia/lib64:/usr/local/rdma/lib64:/usr/lib64/mlnx_ofed/valgrind
- name: PADDLE_TRAINING_ROLE - name: PADDLE_TRAINING_ROLE
...@@ -106,22 +106,24 @@ spec: ...@@ -106,22 +106,24 @@ spec:
labels: labels:
paddle-job: fluid-ctr paddle-job: fluid-ctr
spec: spec:
nodeSelector:
nodeType: model
imagePullSecrets: imagePullSecrets:
- name: default-secret - name: default-secret
volumes: volumes:
- hostPath: - hostPath:
path: /home/work/ path: /home/work/data
type: "" type: ""
name: seqdata name: seqdata
containers: containers:
- image: sivanzcw/edldemo:v1 - image: wangjiawei1993/edldemo:v12
command: command:
- paddle_k8s - paddle_k8s
- start_fluid - start_fluid
imagePullPolicy: IfNotPresent imagePullPolicy: IfNotPresent
name: trainer name: trainer
volumeMounts: volumeMounts:
- mountPath: /mnt/seqdata - mountPath: /data
name: seqdata name: seqdata
resources: resources:
limits: limits:
...@@ -171,7 +173,7 @@ spec: ...@@ -171,7 +173,7 @@ spec:
- name: FLAGS_rpc_deadline - name: FLAGS_rpc_deadline
value: "36000000" value: "36000000"
- name: PADDLE_PORT - name: PADDLE_PORT
value: "30236" value: "30240"
- name: PADDLE_PSERVERS_NUM - name: PADDLE_PSERVERS_NUM
value: "2" value: "2"
- name: PADDLE_TRAINERS_NUM - name: PADDLE_TRAINERS_NUM
...@@ -183,5 +185,5 @@ spec: ...@@ -183,5 +185,5 @@ spec:
- name: LD_LIBRARY_PATH - name: LD_LIBRARY_PATH
value: /usr/local/lib:/usr/local/nvidia/lib64:/usr/local/rdma/lib64:/usr/lib64/mlnx_ofed/valgrind value: /usr/local/lib:/usr/local/nvidia/lib64:/usr/local/rdma/lib64:/usr/lib64/mlnx_ofed/valgrind
- name: ENTRY - name: ENTRY
value: cd /workspace/ctr && python train.py --is_local 0 --cloud_train 1 value: (/postprocess &) && cd /workspace/ctr && python train.py --is_local 0 --cloud_train 1
restartPolicy: OnFailure restartPolicy: OnFailure
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册