Skip to content

CLI Command Reference Manual

This document provides a comprehensive command reference for the pm-compute command-line utility.

1. Global Option

You can run any command under a different user session than the active session by prepending the global --user option:

pm-compute --user <USERNAME> <COMMAND>

2. Configuration & Status Commands

config set

Configure CLI settings like the central API endpoint. 

pm-compute config set --api-url <URL> [--insecure]

  • --api-url: The HTTP/HTTPS address of the Controller (e.g., https://api.computeai.local).
  • --insecure: Skip SSL/TLS certificate validation (use only in self-signed laboratory clusters).

config view

Print the current active API URL configuration. 

pm-compute config view

health

Test if the API Controller is reachable and operational. 

pm-compute health

version

Print the active CLI tool version. 

pm-compute version

3. Account & Session Authentication Commands

signup

Create a new user account. 

pm-compute signup --username <NAME> --password <PASS> --email <EMAIL>

login

Authenticate with the Controller and save session tokens. 

pm-compute login [--username <NAME>] [--password <PASS>]

logout

Log out and invalidate tokens for the current terminal. 

pm-compute logout

whoami

Show active username and security role. 

pm-compute whoami

sessions

List all logged-in usernames on the local machine. 

pm-compute sessions

switch

Switch to another logged-in user profile. 

pm-compute switch <USERNAME> [--password <PASS>]

4. Dataset & File Workspace Commands

file upload

Upload files (data, scripts, weights) into your workspace. 

pm-compute file upload <FILEPATH> [--type <CATEGORY>] [--extract]

  • --type: File category. Supported: dataset (default), code, model, checkpoint.
  • --extract: Boolean flag. If specified, automatically extracts zip or tar archives in the workspace after uploading.

file list

List uploaded files. 

pm-compute file list [--type <CATEGORY>]

  • --type: Filter list by file category.

file download

Download a workspace file by its Database File ID. 

pm-compute file download <FILE_ID> [-o <OUTPUT_PATH>]

file delete

Delete a file by File ID. 

pm-compute file delete <FILE_ID>

file browse

Browse workspace folder contents recursively. 

pm-compute file browse [<RELATIVE_PATH>]

file download-path

Download a workspace file by its relative path. 

pm-compute file download-path <RELATIVE_PATH> [-o <OUTPUT_PATH>]

5. Job & Training Execution Commands

train

Submit a Python training script directly from your workstation to run on a worker node. The CLI automatically reads your local Python file and sends it in the job submission payload (no prior file upload step is required). 

pm-compute train <SCRIPT_PATH> --framework <FRAMEWORK> [--resource <RESOURCE>] --name <RUN_NAME> [--distributed] [--num-nodes <N>] [--checkpoint-interval <EPOCHS>] [--follow]

  • <SCRIPT_PATH>: Local path to the Python training script on your workstation (e.g. ./train.py).
  • --framework: Target framework. Options: pytorch or tensorflow (required).
  • --resource: Compute resource type. Options: cpu or gpu (default cpu).
  • --name: Display name for the training job.
  • --distributed: Enable distributed multi-node execution across the cluster.
  • --num-nodes: Number of worker nodes to allocate (required when --distributed is specified).
  • --checkpoint-interval: Epoch frequency to automatically detect and register saved checkpoints (default 5).
  • --follow, -f: Automatically stream stdout/stderr logs in the terminal until completion.

jobs

List submitted training runs. 

pm-compute jobs [--status <STATUS>]

  • --status: Filter by status: submitted, running, completed, failed, or cancelled.

status

Print full metadata details, metrics, and state of a training run. 

pm-compute status <JOB_ID> [--metrics]

  • --metrics: Include detailed epoch-by-epoch evaluation metric history.

logs

Stream training job terminal stdout/stderr logs. 

pm-compute logs <JOB_ID> [--follow]

checkpoints

List saved checkpoint weights for a training job. 

pm-compute checkpoints <JOB_ID>

resume

Relaunch a failed or cancelled training job from its last saved checkpoint. For details on how the platform manages checkpoints and injects resume epoch environment variables, see the Runtime Variables & Contracts Guide. 

pm-compute resume <JOB_ID> [--follow]

cancel

Stop execution of a running training job. 

pm-compute cancel <JOB_ID>

delete

Delete a training job record. 

pm-compute delete <JOB_ID>

6. Model serving & Deployment Commands

upload-model

Register a model weights file in the Model Registry. 

pm-compute upload-model <FILEPATH> --name <MODEL_NAME> --version <VERSION> --input-shape <INPUT_SHAPE> --output-shape <OUTPUT_SHAPE> [--input-dtype <DTYPE>] [--output-dtype <DTYPE>] [--input-names <NAMES>] [--output-names <NAMES>]

  • <FILEPATH>: Local weights file path.
  • --name: Target model name.
  • --version: Version integer (e.g. 1).
  • --input-shape: JSON array representing input shape (e.g. "[[-1, 1000]]").
  • --output-shape: JSON array representing output shape (e.g. "[[-1, 2]]").
  • --input-dtype: JSON array of data types (default ["FP32"]).
  • --output-dtype: JSON array of data types (default ["FP32"]).

models

List registered models. 

pm-compute models

delete-model

Delete a registered model from registry by name and version. 

pm-compute delete-model <NAME> <VERSION>

deploy

Deploy a registered model as a serving endpoint. 

pm-compute deploy <MODEL_ID>

deployments

List serving deployments and endpoint configurations. 

pm-compute deployments

deployment-status

Show detailed live status, configurations, and health for a deployment. 

pm-compute deployment-status <DEPLOYMENT_ID>

start-deployment

Re-activate a stopped serving deployment. 

pm-compute start-deployment <DEPLOYMENT_ID>

stop-deployment

Pause serving for a deployment to release resources. 

pm-compute stop-deployment <DEPLOYMENT_ID>

delete-deployment

Tear down and delete a serving deployment. 

pm-compute delete-deployment <DEPLOYMENT_ID>

keys create

Create an inference authentication key for a deployment. 

pm-compute keys create <DEPLOYMENT_ID> [--name <KEY_NAME>]

keys list

List active inference keys for a deployment. 

pm-compute keys list <DEPLOYMENT_ID>

keys revoke

Revoke an API inference key. 

pm-compute keys revoke <KEY_ID>

7. Cluster Monitoring Commands

monitoring

Print resource history metrics. 

pm-compute monitoring [--hours <HOURS>] [--limit <MAX_RECORDS>]

dashboard

Display active jobs, resource limits, and cluster success metrics in a summary console view. 

pm-compute dashboard

8. Administrative Commands (Admin Only)

These commands are restricted to users with admin security roles. For administrative setup guides, onboarding policies, and user quota details, see the Platform Administration Guide.

update-slots

Update the maximum concurrent job slots/capacity of a worker resource node (admin only). 

pm-compute update-slots <RESOURCE_ID> <SLOTS>

  • <RESOURCE_ID>: UUID of the target worker resource node.
  • <SLOTS>: Positive integer representing new maximum concurrent jobs.

remove-resource

Deregister a worker node. 

pm-compute remove-resource --name <VM_NAME> [--soft] [--force]

  • --soft: Temporarily disable the worker without removing it.
  • --force: Force remove even if jobs are active.

resources

List active worker nodes, resource slots, and CPU/GPU details. 

pm-compute resources

user list

List platform user accounts and limits. 

pm-compute user list

user create

Create a new platform user with limits. 

pm-compute user create <USERNAME> [--password <PASS>] [--email <EMAIL>] [--role <user/admin>] [--max-jobs <LIMIT>] [--max-gpu <LIMIT>]

user password

Reset password for a user. 

pm-compute user password <USERNAME> [--new-password <PASS>]

user delete

Delete a user account. 

pm-compute user delete <USERNAME>

mlflow config

Show the configuration of the active Experiment Tracking server. 

pm-compute mlflow config

mlflow set

Configure active Experiment Tracking server parameters. 

pm-compute mlflow set <TRACKING_URI> [--auth] [--username <USER>] [--password <PASS>]

mlflow test

Test the connection between the Controller and the Experiment Tracking server. 

pm-compute mlflow test