FA-98372 / Token sampling filters / Member archive
Token sampling: Repetition penalty pushes positive and negative scores alike · case 02
Repetition penalty pushes positive and negative scores alike.
Case contract
For seen tokens and positive penalty, divide nonnegative logits by penalty and multiply negative logits by penalty; unseen logits stay unchanged.
Why this case matters
A finite, deterministic token-selection contract isolates one decoding operation from model inference and runtime behavior.
One recorded failure
Sample boundary fixtureThis sample comes from the broken implementation of a controlled reproducer.
| Boundary fixture | Actual | Expected | Outcome |
|---|---|---|---|
| both signs | {"a": 2.0, "b": -1.0, "c": 3.0} | {"a": 2.0, "b": -4.0, "c": 3.0} | Failed |
MEMBER ARCHIVE
The complete case is available to members.
This record includes three runnable implementations, regression fixtures, execution results, and source hashes.
Member access is invitation-based. Sign in with your invited account to inspect the sources.
Sign in to the archive ↗