your model.
apple silicon.
shipped.
You shouldn't spend a week fighting rope scaling, tokenizer templates, and quant recipes.
We port it, quantize it, benchmark it, hand you weights that actually work on the chip you care about.
custom-quoted, per model.
Every project is different — model size, quantization target, eval requirements, deadline, memory budget. Send the HF link and the constraints; I'll reply within 24 hours with a fixed price and a timeline.
> included in every engagement
- → Public or private delivery — your call.
- → Work done on-device; no cloud leakage of weights.
- → You own the outputs. Commercial-friendly.
- → Fixed price. No hourly surprise bills.
- → Convert fails on unsupported ops? You don't pay.
> FAQ
Will my model run on my specific Mac?
I bench on the exact chip before delivery. If it won't hit your memory budget, I'll tell you before you pay.
Weight confidentiality?
Signed NDA on request. All work happens locally on my hardware — no weights leave the building.
What if conversion hits unsupported ops?
You don't pay. I'll document what's blocking and suggest alternatives or upstream fixes.
Can I hire you ongoing?
Pro tier includes 2 weeks of support. For longer, see the Custom tier — retainers available.
How is pricing decided?
Per model. After you send the HF link and constraints, I'll quote a fixed price inside 24 hours. No hourly surprise bills.
Payment / invoicing?
50% upfront, 50% on delivery. Stripe, ACH, or wire. Company invoicing fine.
> request a quote
I'll reply within 24 hours. No newsletter, no funnel, just a human reading the form.