promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_

Lab — ship the adapter — step 3 of 3

ch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialistch 58 · fine-tune a small specialist3/3
Checkpoint

One last thing before we move on. pass this to mark the lesson done, or skip and keep moving. hop to the next when you're ready.

Checkpoint. Write emit_adapter_receipt(payload, train) that returns a dict with these keys, then print the four receipts as name kind verdict.

Receipt fields:

  • job, base_model, method copied from the payload
  • kind from train["kind"] — "fixture" or "live-t4"
  • verdict from audit_qlora_payload(payload) (same rules as the write step). A fixture is allowed to pass when the payload is honest. A live T4 with a leaky payload is still a hold.

Expected output:

fixture-good fixture pass
live-good live-t4 pass
fixture-leak fixture hold: train/eval ids overlap
live-70b live-t4 hold: base must be Qwen2.5 0.5-3B Instruct

full-screen editor opens — close anytime to keep reading.

open the editor to run it.