Sgdr Step 1
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
Sgdr Step 1 has 3 facts recorded in Dontopedia across 2 references.
3 facts·3 predicates·2 sources
Maturity scale
raw canonical shape-checked rule-derived certifiedOther facts (3)
The long tail: predicates that appear too rarely to warrant their own section. Filter or scroll to find a specific one. Each row links to its source.
3 facts
| Predicate | Value | Ref |
|---|---|---|
| Runs Full Epoch With Cosine Decay | LR: high → floor | [1] |
| Involves | Cosine Decay | [2] |
| Learning Rate Direction | High to Floor | [2] |
Timeline
Timeline axis is valid_time — when each source says the fact was true in the world, not when Dontopedia learned about it. Retracted rows are kept for provenance; coloured stripes indicate the context kind.
—
runsFullEpochWithCosineDecayblah/watt-activation/part-24
LR: high → floor
—
involvesblah/watt-activation/24
ex:cosine-decay
—
learningRateDirectionblah/watt-activation/24
ex:high-to-floor
References (2)
2 references
ctx:discord/blah/watt-activation/part-24ctx:discord/blah/watt-activation/24- full textwatt-activation-24text/plain3 KB
doc:agent/watt-activation-24/5da841db-60e0-40fc-b5cf-eafaf36ee8d7Show excerpt
[2026-03-06 15:32] xenonfun: should we be restarting from the last iter as it learned more, and just adjusting learning ranges a bit closer? ⏺ Yes, good instinct. The "best loss" checkpoint is often just a lucky easy batch — the model at t…
See also
Keep researching
Missing something or suspicious of what's here? Kick off a research session — a Claude agent will investigate, cite its sources, and file new facts into a dedicated context you can review before accepting into the shared view.