FA-98900 / Gradient accumulation weighting / Member archive

A changing loss scale is applied after gradient summation · case 05

A changing loss scale is applied after gradient summation.

Member previewVariant 5 · 3 implementations · 7 checks per implementation

Case contract

Each microbatch sum uses its own positive loss scale. Unscale each sum before combining and divide by total examples; zero-scale contributions are omitted.

Why this case matters

A deterministic toy model isolates one local accumulation contract.

One recorded failure

Sample boundary fixture

This sample comes from the broken implementation of a controlled reproducer.

Boundary fixture	Actual	Expected	Outcome
different scales	`3.3333333333333335`	`5`	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.