Jobs

Create, manage, and monitor distributed computing jobs. Includes job creation, status monitoring, log viewing, and shell access.

Command

cosmicac jobs <subcommand> [options]

Subcommands

Subcommand	Description
create	Create a job
delete	Delete a job
detail	Get job details
list	List of jobs
logs	Query inference logs
replica	Job replica management
restart	Restart a job
shell	Start a shell for a job container

`jobs create`

Create a new GPU container or managed inference job. Use flags for non-interactive mode or omit flags for interactive mode.

Usage

cosmicac jobs create [options]

Options

Option	Description
`--name`	Job name
`--type`	Job type: container, vllm, or parakeet
`--location`	Location (e.g., us, IN)
`--tags`	Comma-separated tags
`--cost-limit`	Cost limit (optional)
`--gpu`	GPU configuration in format "TYPE=COUNT" (e.g., "H100=2")
`--gpu-type`	GPU type (alternative to --gpu)
`--gpu-count`	GPU count (alternative to --gpu)
`--driver`	GPU driver (default: CUDA 12.9)
`--vram-gb`	VRAM in GB per GPU (optional)
`--cpu-cores-per-gpu`	CPU cores per GPU (optional)
`--mem-gb-per-gpu`	Memory in GB per GPU (optional)
`--base-image`	Base image for GPU container (e.g., Ubuntu22.04/CUDA12.9)
`--root-disk-size-gb`	Root disk size in GB (default: 250)
`--model`	Model ID for inference
`--runtime-image`	Runtime image (default: vllm-openai-0.8.5)
`--data-type`	Data type: Auto, BF16, FP16, FP32 (default: Auto)
`--quantisation`	Quantisation: none, fp8, awq, gptq, compressed-tensors, bitsandbytes (default: fp8)
`--tensor-parallel`	Tensor parallel count (default: 1)
`--gpu-memory-utilization`	GPU memory utilization 0-1 (default: 0.9)
`--max-concurrent-sequences`	Max concurrent sequences (default: 256)
`--max-model-length`	Max model length in tokens (default: 32768)
`--reasoning-parser`	Reasoning parser: default, qwen3, deepseek_r1 (default: default)
`--multimodal`	Enable multimodal support for vision-language models
`--endpoint-name`	Endpoint name for API access
`--replica`	Number of replicas (default: 1)
`--require-auth-header`	Require Authorization header (default: true)
`--no-auth-header`	Disable Authorization header requirement
`--env`	Environment variables in format "NAME1=VALUE1,NAME2=VALUE2"
`--chunk-duration`	Audio chunk duration in seconds (Parakeet only)
`--chunk-overlap`	Audio chunk overlap in seconds (Parakeet only)
`--max-file-size-mb`	Max file size in MB (Parakeet only)
`--config`	Path to JSON config file
`--dry-run`	Validate configuration without creating the job

`jobs delete`

Delete a specific job using its Job ID.

Usage

cosmicac jobs delete <jobId>

Arguments

Argument	Required	Description
`jobId`	Yes	Job ID (required)

`jobs detail`

Retrieve and display detailed information about a specific job.

Usage

cosmicac jobs detail <jobId> [options]

Arguments

Argument	Required	Description
`jobId`	Yes	Job ID (required)

Options

Option	Description
`--overwrite-cache`, `-o`	Overwrite cache (optional)

`jobs list`

Retrieve and display a list of jobs.

Usage

cosmicac jobs list

`jobs logs`

Retrieve paginated logs for an inference instance

Usage

cosmicac jobs logs [options]

Options

Option	Description
`--instance-id`	Instance ID (required)
`--start-date`	Start date in ISO 8601 format (e.g., 2026-06-11T00:00:00.000Z)
`--end-date`	End date in ISO 8601 format (e.g., 2026-06-11T02:00:00.000Z)
`--search`	Substring filter on message text
`--level`	Min severity: 1=DEBUG, 2=INFO, 3=WARNING, 4=ERROR, 5=FATAL
`--source`	Log source: 1=system/setup logs, 2=application logs
`--limit`	Page size (default 100, max 1000)
`--page`	Page number (ignored when cursor is set)
`--cursor`	Log record ID for cursor pagination
`--follow`	Continuously stream new logs (tail mode)
`--raw`	Output raw JSON format

`jobs replica`

Commands for managing individual job replicas. Restart or delete specific replicas of a distributed job.

Usage

cosmicac jobs replica <subcommand> [options]

Subcommands

Subcommand	Description
delete	Delete a job replica
restart	Restart a job replica

`jobs replica delete`

Delete a specific replica of a job using its Job ID and replica ID.

Usage

cosmicac jobs replica delete <jobId> <replicaId>

Arguments

Argument	Required	Description
`jobId`	Yes	Job ID (required)
`replicaId`	Yes	Replica ID (required)

`jobs replica restart`

Restart a specific replica of a job using its Job ID and replica ID.

Usage

cosmicac jobs replica restart <jobId> <replicaId>

Arguments

Argument	Required	Description
`jobId`	Yes	Job ID (required)
`replicaId`	Yes	Replica ID (required)

`jobs restart`

Restart all replicas of a specific job using its Job ID.

Usage

cosmicac jobs restart <jobId>

Arguments

Argument	Required	Description
`jobId`	Yes	Job ID (required)

`jobs shell`

Start an interactive shell session for a running GPU container job.

Usage

cosmicac jobs shell <jobId> <containerId>

Arguments

Argument	Required	Description
`jobId`	Yes	Job ID (required)
`containerId`	Yes	Container ID (required)

On this page