$
$
$
01 / Pricing
Flexible pricing
Pay per run, or save up to 30% on every run with BeaverYard Pro. Memberships also include Priority Processing and priority support.
Validate before you pay — failed runs cost nothingFine-tune any LLM.
Fine-tune any LLM.
Your dataset,
your adapter.
Upload a JSONL. We check it free. You see the price. We train. You download PEFT-compatible LoRA weights. No platform lock-in. No hidden tiers. Runs from $5.
Check first, pay afterRuns start at $5LLaMA, Mistral, GemmaAPI included
support replies dataset
USE CASE
customer_replies.jsonl
0 support examples
json
json
json
json
json
json
json
LORA CORE
ADAPTER
artifacts.tar.gz
download ready
Checks passed · price confirmedprice locked at upload
your assistant is ready
Base models
LLaMA · Mistral · Gemma
LLaMAMistralGemma
Format
PEFT-compatible LoRA
Runs from
$5
Validation
Always free
PASS
Delivery
Adapter weights
adapter_model.safetensors
From dataset to adapter
Five steps. No surprises.
Every checkpoint is validated before payment. The price you see is the price you pay.
01
Step 01 · Upload
Upload dataset
JSONL format, any size within published limits.
.jsonl
drop here
02
Step 02 · Validate
Free validation
Format, size, token count, and record limits — checked before payment.
format · jsonl
size · within cap
tokens · counted
records · valid
03
Step 03 · Review & pay
Review & pay
See your assigned tier and exact price. Pay only when eligible.
$5–9
Launch tier · per run
04
Step 04 · Train
Training runs
We fine-tune using a standardized recipe with the base model you chose.
05
Step 05 · Download
Download artifacts
Adapter weights and two customer guides. Yours within 7 days.
What makes us different
Built around your workflow.
$
$
$
{
}
What you get
Everything you need after training.
Three artifacts per completed run. Available to download for 7 days from completion.
adapter ready
PEFT compatibleplug & play7-day access
01
Adapter package
- adapter_model.safetensors
- PEFT-compatible format
- Drop into any inference stack
Training Summary
What your adapter learned
Behavioral changes
Recommended next steps
02
Training Focus Summary
- What your adapter learned
- Behavioral changes to expect
- Plain language, no jargon
Download artifacts
available for 7 days
Load & test your model
with your own data
Ship to your product
no platform lock-in
03
Usage Guidelines
- How to load your adapter
- Where it performs well
- Where it may struggle
How we operate
Designed for transparency, not complexity.
Validation always free
Format, size, token count, and record limits — all checked before you're asked for a card. If your dataset can't run, you don't pay.
llama
mistral
gemma
Standardized recipes
One LoRA recipe per base model family, tuned by us. No knobs to misconfigure, no parameter-tuning rabbit hole.
dataset
adapter
One path, on purpose
One LoRA recipe per base model. We're not pretending to be a hyperscaler — we're optimizing one path so you don't pay for choice you don't need.
7
DAYS
Your data, your weights
7-day retention on artifacts, then deleted. We don't resell datasets, we don't host inference, we don't keep your model.
Transparent pricing
Reference bands. No guessing.
BeaverYard auto-assigns the right band after upload and validation. The higher required tier wins when dataset size and tokens disagree.
★ Most common
Launch
Datasets up to 300 MB · Up to 9.5M tokens
$5–9 / run
- Adapter weights + config
- Training Focus Summary
- Usage Guidelines
- 7-day download window
- API access included
Orbit
Datasets 300–600 MB · Up to 12M tokens
$15–25 / run
- Adapter weights + config
- Training Focus Summary
- Usage Guidelines
- 7-day download window
- API access included
Save 20% with Plus ($3/mo) · 30% with Pro ($5/mo) · PAYG always available
Membership tiers
BeaverYard Plus and Pro.
Optional. Pay-as-you-go is always available, no membership required.
Pay-as-you-go
$0/ month
Default. No commitment.
- ✓Per-run pricing at standard tier rates
- ✓Free validation before payment
- ✓Full API access
- ✓Standard queue priority
Plus
★ Most popular$2.50/ month$3
Billed $30/yr — Save 17%
- ✓20% discount on all training runs
- ✓Priority queue placement
- ✓Free validation, full API access
- ✓Pays for itself after ~2 Launch runs
Pro
Max savings$4/ month$5
Billed $48/yr — Save 20%
- ✓30% discount on all training runs
- ✓Highest queue priority
- ✓Free validation, full API access
- ✓Best value above ~3 runs/month
Common questions
The honest answers.
Which base models can I fine-tune?+
LLaMA, Mistral, and Gemma — the three open-weight families with mature LoRA tooling. We pin specific revisions per family to keep training reproducible.
What format does my dataset need to be in?+
JSONL with one example per line. Our validator walks you through expected fields. Wrong shape? You'll see exactly what's off — for free, before payment.
What if my run fails?+
If validation fails, you don't pay. If a run fails after we've started training, we re-run it once at no charge or refund. We'd rather refund you than have you doubt the bill.
How long are my artifacts available?+
7 days from completion. Download the adapter, the Training Focus Summary, and the Usage Guidelines, then we delete our copy.
Do you host inference?+
No, on purpose. You take your weights to whatever inference stack you already use — vLLM, TGI, SGLang, transformers. No platform lock-in, no second bill.
Is there an API?+
Yes, on every plan including PAYG. Trigger runs, poll status, fetch artifacts. Wire it into your CI in an afternoon.
Ready to train your custom model?
Upload your dataset, get a free eligibility check, pay only when your run qualifies. Runs from $5.