example of using kfto to do a finetuneing

services/finetuning/examples/kfto_job.yaml

@@ -0,0 +1,54 @@
+apiVersion: "kubeflow.org/v1"
+kind: PyTorchJob
+metadata:
+  name: tsfmfinetuning-job
+  namespace: default
+  annotations:
+    sidecar.istio.io/inject: "false"
+spec:
+  pytorchReplicaSpecs:
+    Master:
+      replicas: 1
+      restartPolicy: Never
+      template:
+        spec:
+          containers:
+            - name: pytorch
+              image: localhost:5001/tsfmfinetuning:latest
+              imagePullPolicy: Always
+              volumeMounts:
+                - name: volv
+                  mountPath: /data
+              resources:
+                requests:
+                  memory: 8Gi
+                  cpu: 1000m
+                limits:
+                  memory: 12Gi
+                  cpu: 1500m
+              env:
+              - name: TSFM_ALLOW_LOAD_FROM_HF_HUB
+                value: "0"
+              - name: TSFM_MODEL_DIR
+                value: "/data"
+              command:
+                - "python"
+                - "tsfmfinetuning/ftmain.py"
+                - "--payload"
+                - "/data/ftpayload.json"
+                - "--target_dir"
+                - "/data/finetuned-models"
+                - "--model_name"
+                - "finetuned_from_kfto"
+                - "--config_file"
+                - "/data/default_config.yml"
+                - "--model_arch"
+                - "ttm"
+                - "--task_type"
+                - "forecasting"
+          volumes:
+          - name: volv
+            persistentVolumeClaim:
+              claimName: local-path-pvc
+
+