FA-99326 / Distributed training aggregation / Open access

Gradient bucket reconstruction discards global offsets · case 01

Gradient bucket reconstruction discards global offsets.

Verified by executionVariant 1 · 6 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

Different workers may contribute different bucket widths; bucket ownership is defined by global offsets.

THE FAILURE

Different workers may contribute different bucket widths; bucket ownership is defined by global offsets.

Unsuccessful approach: Appending payloads by width can put values into the wrong bucket.

Case contract

Given [offset,values] and total length, reconstruct exactly covered list or return None for gaps/overlaps.

Why this case matters

A toy distributed-training contract makes replica and shard contribution rules explicit without claiming framework compatibility or performance.

1 / The failure

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(chunks, total):
    return [v for s,vs in chunks for v in vs]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('chunks reversed',solve([[2,[2,3]],[0,[0,1]]],4),[0,1,2,3])
check('overlap rejected',solve([[0,[0,1,2]],[2,[2,3]]],4),None)
check('gap rejected',solve([[0,[0]],[2,[2,3]]],4),None)
check('empty tensor',solve([],0),[])
check('singleton chunk',solve([[0,[N]]],1),[N])
check('out of bounds',solve([[0,[0,1]]],1),None)
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
chunks reversed	`[2, 3, 0, 1]`	`[0, 1, 2, 3]`	Failed
overlap rejected	`[0, 1, 2, 2, 3]`	`None`	Failed
gap rejected	`[0, 2, 3]`	`None`	Failed
empty tensor	`[]`	`[]`	Passed
singleton chunk	`[1]`	`[1]`	Passed
out of bounds	`[0, 1]`	`None`	Failed

SHA-256 / c0403751c0b27211a4355a4c30e1be47cd9b435e385ee369b15bb4e7a28c5a78

2 / The unsuccessful fix

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(chunks, total):
    out=[None]*total
    for s,vs in chunks:
        for j,v in enumerate(vs):
            if 0<=s+j<total: out[s+j]=v
    return out
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('chunks reversed',solve([[2,[2,3]],[0,[0,1]]],4),[0,1,2,3])
check('overlap rejected',solve([[0,[0,1,2]],[2,[2,3]]],4),None)
check('gap rejected',solve([[0,[0]],[2,[2,3]]],4),None)
check('empty tensor',solve([],0),[])
check('singleton chunk',solve([[0,[N]]],1),[N])
check('out of bounds',solve([[0,[0,1]]],1),None)
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
chunks reversed	`[0, 1, 2, 3]`	`[0, 1, 2, 3]`	Passed
overlap rejected	`[0, 1, 2, 3]`	`None`	Failed
gap rejected	`[0, None, 2, 3]`	`None`	Failed
empty tensor	`[]`	`[]`	Passed
singleton chunk	`[1]`	`[1]`	Passed
out of bounds	`[0]`	`None`	Failed

SHA-256 / 70c688f564fa364063ee6aac95dad4e433548a25db92031822a25aab49d2bd8f

HELD IN THE MEMBER ARCHIVE

The verified repair and its recorded checks are member-only.

This mechanism has 6 recorded checks per implementation. The open-access tier publishes the failure and the unsuccessful fix; the repaired source that passes every check, and the observations that prove it, are available to members.

Every case sharing this mechanism uses the same contract and the same repair, so this one record is held back for all of them.

Member access is invitation-based. Sign in with your invited account to inspect the repair.

Verification & scope

Inputs are already materialized Python values from a fixed toy collective; no transport, failure recovery, optimizer state, or real model behavior is represented. This reproducer isolates one failure mechanism. Results cover the supplied fixtures. Variants within a family share a test contract and should remain grouped when constructing evaluation splits. Related mechanisms with a shared evaluation_group must also remain together; these controlled models are not independent production incidents.

Observations recorded using Python 3.12.14 at 2026-09-30T16:48:23.079335+00:00.

Case digest / 1a8f623a0869cdda428fd14b17c2f8b2f0c97489046d5da23a89631fc1418929