Jobs Create, manage, and monitor distributed computing jobs. Includes job creation, status monitoring, log viewing, and shell access.
Create, manage, and monitor distributed computing jobs.
Includes job creation, status monitoring, log viewing, and shell access.
cosmicac jobs < subcomman d > [options] Create a new GPU container or managed inference job. Use flags for non-interactive mode or omit flags for interactive mode.
cosmicac jobs create [options] Option Description --nameJob name --typeJob type: container, vllm, or parakeet --locationLocation (e.g., us, IN) --tagsComma-separated tags --cost-limitCost limit (optional) --notifyComma-separated push notification events to enable: job_failed,job_degraded,job_recovered,job_restart_storm (or all/none; default: all) --gpuGPU configuration in format "TYPE=COUNT" (e.g., "H100=2"; 16 for multi-node managed inference) --gpu-typeGPU type (alternative to --gpu) --gpu-countGPU count: 1, 2, 4, 8, or 16 (16 is multi-node, inference only) --driverGPU driver (default: CUDA 12.9) --vram-gbVRAM in GB per GPU (optional) --cpu-cores-per-gpuCPU cores per GPU (optional) --mem-gb-per-gpuMemory in GB per GPU (optional) --base-imageBase image for GPU container (e.g., Ubuntu22.04/CUDA12.9) --root-disk-size-gbRoot disk size in GB (default: 250, min: 100) --modelModel ID for inference --runtime-imageRuntime image (default: vllm/vllm-openai:v0.17.1) --data-typeData type: Auto, BF16, FP16, FP32 (default: Auto) --quantisationQuantisation: none, fp8, awq, gptq, compressed-tensors, bitsandbytes (default: none) --tensor-parallelTensor parallel: 1, 2, 4, 8, or none (default: none) --gpu-memory-utilizationGPU memory utilization 0-1 (default: 0.9) --max-concurrent-sequencesMax concurrent sequences 1-512 (default: 256) --max-model-lengthMax model length in tokens (default: 32768) --reasoning-parserReasoning parser: qwen3 or deepseek_r1 (omit for none) --multimodalEnable multimodal support for vision-language models --endpoint-nameEndpoint name for API access --replicaNumber of replicas for vLLM or Parakeet: 1, 2, or 4 (default: 1) --require-auth-headerRequire Authorization header (default: true) --no-auth-headerDisable Authorization header requirement --envEnvironment variables in format "NAME1=VALUE1,NAME2=VALUE2" --chunk-durationAudio chunk duration in seconds (Parakeet only, default: 10) --chunk-overlapAudio chunk overlap in seconds (Parakeet only) --max-file-size-mbMax file size in MB (Parakeet only, min 1024) --configPath to JSON config file --dry-runValidate configuration without creating the job
Delete a specific job using its Job ID.
Prompts for confirmation unless --yes is passed.
cosmicac jobs delete < jobI d > [options] Argument Required Description jobIdYes Job ID (required)
Option Description --yesSkip confirmation prompt
Retrieve and display detailed information about a specific job.
cosmicac jobs detail < jobI d > Argument Required Description jobIdYes Job ID (required)
Retrieve and display a list of jobs.
Retrieve Loki history or stream live logs for a job (same monitor API as the UI).
cosmicac jobs logs < jobI d > [options] Argument Required Description jobIdYes Job ID (required)
Option Description --typeLog type: application, system (default: application) --replica-idFilter by replica ID --levelMin history level: all, info, warn, error (default: all) --searchSubstring filter on history message text --periodHistory window: 15m, 1h, 6h, 24h, 7d (default: 1h) --startHistory start in ISO 8601 (overrides --period start) --endHistory end in ISO 8601 (overrides --period end) --limitMax history entries (default 200 dumped / 1000 in TTY, max 1000) --followStream live logs over WebSocket --liveAlias for --follow --rawOutput raw JSON
Commands for managing individual job replicas.
Restart or delete specific replicas of a distributed job, including queued, starting, and pending replicas.
cosmicac jobs replica < subcomman d > [options] Subcommand Description delete Delete a job replica restart Restart a job replica
Delete a specific replica of a job using its Job ID and replica ID.
Queued, starting, and pending replicas can be deleted. Prompts for confirmation unless --yes is passed.
cosmicac jobs replica delete < jobI d > < replicaI d > [options] Argument Required Description jobIdYes Job ID (required) replicaIdYes Replica ID (required)
Option Description --yesSkip confirmation prompt
Restart a specific replica of a job using its Job ID and replica ID.
Queued, starting, and pending replicas can be restarted.
cosmicac jobs replica restart < jobI d > < replicaI d > Argument Required Description jobIdYes Job ID (required) replicaIdYes Replica ID (required)
Restart all replicas of a specific job using its Job ID.
cosmicac jobs restart < jobI d > Argument Required Description jobIdYes Job ID (required)
Start an interactive shell session for a running GPU container job.
cosmicac jobs shell < jobI d > < containerI d > [options] Argument Required Description jobIdYes Job ID (required) containerIdYes Container ID (required)
Option Description --timeoutSeconds to wait for a shell peer (default: 30)
