gpu fine-tune create

Create a fine-tuning job

Synopsis

Submits a fine-tuning job. Provide an already-uploaded dataset with --training-file, or pass a local JSONL with --dataset to upload it first. With --wait-timeout>0 the CLI polls the job until it reaches a terminal status (succeeded/failed/cancelled); --wait-timeout=0 (the default) returns immediately so you can track progress with gpu fine-tune get and gpu fine-tune events.

gpu fine-tune create [flags]

Examples

# Upload a local dataset and start a QLoRA job in one step
gpu fine-tune create --model qwen2.5-7b-instruct --dataset ./train.jsonl --method qlora

# Use a previously uploaded file id
gpu fine-tune create --model qwen2.5-7b-instruct --training-file file-abc123

# Tune the LoRA hyperparameters and cap GPU-hour spend
gpu fine-tune create --model qwen2.5-7b-instruct --dataset ./train.jsonl \
  --lora-r 32 --lora-alpha 64 --lora-dropout 0.1 --max-budget 25

# Block until the job finishes (poll up to 2 hours)
gpu fine-tune create --model qwen2.5-7b-instruct --dataset ./train.jsonl --wait-timeout 2h

Options

      --dataset string          Local JSONL dataset to upload before creating the job
      --gpu-preference string   Optional GPU preference hint
  -h, --help                    help for create
      --lora-alpha int          LoRA alpha (default 32)
      --lora-dropout float      LoRA dropout (default 0.05)
      --lora-r int              LoRA rank (r) (default 16)
      --max-budget float        Max GPU-hour budget in USD (0 = no cap)
      --method string           Tuning method: lora | qlora (default "qlora")
      --model string            Tunable base model id (e.g., qwen2.5-7b-instruct) — required
      --training-file string    Already-uploaded training file id
      --wait-timeout duration   How long to poll the job for a terminal status; 0 = don't poll, return immediately

Options inherited from parent commands

      --api-base string   API base URL (override with GPUAI_API_BASE env) (default "https://api.gpu.ai/v1")
      --debug             Enable debug logging to stderr
  -o, --output string     Output format: table|json (default table on TTY, json otherwise)

SEE ALSO

• gpu fine-tune - Manage fine-tuning jobs