Topologies
[training topology]
@beam/sweepHyperparameter sweep
A fan-out over sweep points from a common baseline, one merged model per point, then a final merge.
Static membershipBuiltinActivedistributed run
[topology identity]
Details
What version 1.0.0 declares.
- Package
@beam/sweep- Membership
- Static
- Engines
- distributed run
- Tags
- hyperparameter-sweep, rounds, synthetic
- Step kinds
- local, gather
- Latest tag
- latest -> 1.0.0
- Publisher
- registry
- Visibility
- Public
- Checksum
sha256:f3a55d0b7f9fb1ae07cd01a75a770d066a1f7fc11356adb5e34f134546a33ceb
[roles]
Roles
Who takes part, and how many members each role holds.
Trains the baseline, merges each point and the results.
Every eligible room member trains each point; a gather holds at most 15.
[slots]
Slots
The actions a run binds: what each slot consumes and produces.
model* -> model, report?
Merges models (one point's workers, or every point); its report holds the loss.
model?, neighbor? -> model, report
Trains with the point's hyperparameters (the baseline with the defaults).
[round]
Step kinds
The steps of one round, in order.
- localrole workers · slot train
- gatherfrom workers · to lead · object model · slot merge