GETTING STARTED · QUICKSTART

Quickstart

Submit your first LoRA fine-tuning job, watch it train, and download the adapter. Once you have a dataset ready this takes about fifteen minutes end to end.

1 · Install the CLI

Requires Python 3.10+.

$ pip install ownllm
✓ installed ownllm

2 · Sign in

$ ownllm login
# opens a browser tab to complete device-flow auth
✓ logged in as you@company.com

The token is written to ~/.ownllm_cli/token.json with 0600 permissions. Sign out any time with ownllm logout.

3 · Prepare a dataset

JSONL, one example per line. For instruction-tuning, each line has instruction and output fields:

{"instruction": "Summarize this renewal clause...", "output": "In one sentence..."}
{"instruction": "Draft a reply to...", "output": "Thanks for..."}

4 · Submit a fine-tuning job

$ ownllm finetune --data ./dataset.jsonl \
    --task instruct \
    --quality balanced \
    --model 7b
→ uploading dataset · estimating cost
→ estimated cost ≈ 3.01 credits · confirm? [y/N]

The CLI prints a token-level cost estimate before the job starts. If you’d rather not confirm interactively, pass --dry-run to just see the estimate, or --no-wait to submit without blocking.

5 · Watch it train

$ ownllm status
job_8c3f… · RUNNING · epoch 2/3 · eval loss 1.58

For a live tail: ownllm logs <job-id> --follow. Or open the job in the web app for a training-loss chart and phase timeline.

6 · Download the adapter

$ ownllm download <job-id>
✓ adapter saved to ./adapters/<job-id>/

The download contains the LoRA weights in safetensors format plus a config.json. Serve it on any vLLM or TGI stack — or wait for hosted inference, which lands after alpha.