How It Works
Start a fine-tuning run in minutes. Upload your dataset, validate before payment, and download clean training artifacts when it's done.
Built for students and individual builders who want to fine-tune quickly and get downloadable artifacts without infrastructure overhead.
Plans are simple: Launch for smaller runs. Orbit for larger datasets and longer retention.
Six-step flow
Choose a plan (Launch or Orbit)
Upload your dataset
Review eligibility + final price (before you pay)
If your dataset doesn't pass validation, you won't be charged.
Confirm and pay (one-time per run)
Track progress with a secure link
Download your results (Launch: 4 days, Orbit: 7 days)
Start here
You'll upload first, then see eligibility and final price before paying.
Eligibility check before you pay
If your dataset doesn't pass validation, you won't be charged.
Common reasons validation fails
- File doesn't match Dataset Requirements (format/structure).
- Missing required input/output structure.
- Dataset doesn't meet the minimum size/quality checks for the selected plan.
- Per-line or total-size limits exceeded for the selected plan.
Data lifecycle and retention
Upload → Validate → Train → Deliver → Auto-Delete
Downloads stay available for 4 days (Launch) and 7 days (Orbit).
Need more time? Extend retention by 30 days.
We'll email you before your download window expires.
Trust & privacy
- No passwords. Secure sign-in links.
- We use your email only for status updates and your download link.
- Your files are auto-deleted after the retention window.
- Clear retention windows (Launch: 4 days, Orbit: 7 days).
What you get
Included with every run
- Downloadable adapter package
- Run summary (what completed, key run details)
- Basic run logs
Optional add-ons
- Detailed logs pack (extra troubleshooting visibility)
- Retention extension (30 days)
- Priority processing (faster queue start)
- Training focus summary (what your training emphasized)
- Usage guidelines (how to use results, key limitations)
Redacted package preview (included)
run_8h2k31/
├── artifacts/
│ ├── adapter.safetensors
│ └── adapter-config.json
├── summary/
│ └── run-summary.json
└── logs/
└── basic-logs.txtOptional add-ons (downloadables)
add-ons/ ├── detailed-logs.zip ├── training-focus-summary.pdf └── usage-guidelines.pdf
Some add-ons change service behavior and don't appear as files (Priority processing, Retention extension).
Preview shows an example structure. Exact filenames may vary.
You'll upload first, then see eligibility and final price before paying.
What we don't do
- No hosted inference endpoints.
- No subscriptions (one-time per run).
- No hidden post-training fees.
FAQ
What dataset format do I upload?
Upload your dataset in the supported format listed in Dataset Requirements. We validate structure before payment to confirm it's ready.
What happens if validation fails?
You'll see the reason and can fix your dataset. If validation fails before training starts, you are not charged.
How long does a run take?
Run time depends on dataset size and queue conditions. You can track progress from your run dashboard.
What exactly do I download?
You'll download an adapter package, a run summary, and basic logs. Optional add-ons can include a detailed logs pack and two guidance PDFs.
How long do downloads stay available (Launch vs Orbit)?
Launch keeps downloads for 4 days and Orbit for 7 days. You can extend retention to 30 days with an add-on.
Can I re-download later?
Yes, anytime during your retention window. After it expires, files are automatically deleted.
Do you host my model?
No. BeaverYard delivers downloadable artifacts. Hosting and deployment are your responsibility.
Can I delete my data sooner?
Yes. Automatic deletion happens after the retention window, and you can request early deletion through support.