Deployments

Roll back a deployment to a previous revision

Changed on

Re-applies the spec of a previous revision. Pods running the target revision will be retained. Other pods will be drained and restarted with the target revision. Only the 50 most recent revisions per deployment can be rolled back to; older targets return 404.

post/v1/deployments/{id}/rollback

Request

  • Base URL
    https://api.together.ai
  • URL
    https://api.together.ai/v1/deployments/{id}/rollback
  • Auth
    HTTP bearer

Path parameters

idstring required

Deployment ID or name.

Request body

revision_identifierstring required

Revision number or revision ID to roll back to.

Response

Deployment after rollback

argsstring[]

Args are the arguments passed to the container's command

capacity_type'stable' | 'preemptible'

Controls how replicas above reserved capacity behave. stable replicas stay running after scale-up; preemptible replicas may be evicted during capacity contention.

commandstring[]

Command is the entrypoint command run in the container

cpunumber

CPU is the amount of CPU resource allocated to each replica in cores (fractional value is allowed)

created_atstring date-time

CreatedAt is the ISO8601 timestamp when this deployment was created

descriptionstring

Description provides a human-readable explanation of the deployment's purpose or content

desired_replicasinteger

DesiredReplicas is the number of replicas that the orchestrator is targeting

gpu_countinteger

GPUCount is the number of GPUs allocated to each replica in this deployment

gpu_type'h100-80gb' | 'h100-40gb-mig' | 'h200-140gb' | 'b200-192gb'

GPUType specifies the type of GPU requested (if any) for this deployment

health_check_pathstring

HealthCheckPath is the HTTP path used for health checks of the application

idstring

ID is the unique identifier of the deployment

imagestring

Image specifies the container image used for this deployment

max_replicasinteger

MaxReplicas is the maximum number of replicas to run for this deployment

memorynumber

Memory is the amount of memory allocated to each replica in GiB (fractional value is allowed)

min_replicasinteger

MinReplicas is the minimum number of replicas to run for this deployment

namestring

Name is the name of the deployment

object'deployment'

The object type, which is always deployment.

portinteger

Port is the container port that the deployment exposes

ready_replicasinteger

ReadyReplicas is the current number of replicas that are in the Ready state

replica_eventsobject

ReplicaEvents is a mapping of replica names or IDs to their status events

status'Updating' | 'Scaling' | 'Ready' | 'Failed' | 'ScaledToZero'
storageinteger

Storage is the amount of storage (in MB or units as defined by the platform) allocated to each replica

termination_grace_period_secondsinteger

TerminationGracePeriodSeconds is the time in seconds to wait for graceful shutdown before forcefully terminating the replica

updated_atstring date-time

UpdatedAt is the ISO8601 timestamp when this deployment was last updated

Example response

{
  "autoscaling": {
    "metric": "HTTPTotalRequests",
    "target": 100,
    "time_interval_minutes": 10
  },
  "model_mounts": [
    {
      "model_id": "ml_CbJNwQC2ZqCU2iFT3mrCh",
      "mount_path": "/models",
      "revision_id": "rv_8kQ2mN4pL7xR9tV1wY3zA"
    }
  ]
}

Changes